Has a spark1047
2026-09
B iCode Positioned as an offline-first AI development platform plus agent/workflow toolbox, TUI-friendly with self-held data, hitting the real need for localized agent development. 238 stars in two days is a good start. Monetization can be open-source core plus enterprise private deployment, but the track is extremely crowded (Cursor, OpenClaw, various CLI agents), and insufficient differentiation is the biggest risk.
A lightweight, extensible, fully offline development platform and agent/workflow toolkit for the AI era, with an intuitive TUI, easy deployment, and complete control over your data, agents, and workfl…
B oxbox A tiny JS/TS/JSX parse-and-transform library from unjs built on oxc, filling a gap for lightweight AST tools in the JS ecosystem as a Babel alternative an order of magnitude faster. unjs has strong trust in the Node toolchain and is easily absorbed as a standard dependency; the risk is the fast-evolving oxc/sWC ecosystem requiring continuous maintenance.
🐂 Tiny, portable JS library for parsing and transforming JS/TS/JSX, powered by oxc
B Show HN: NSL – WSL for Linux Uses a single VM plus systemd-nspawn to replicate a WSL-style multi-distro development experience, solving the pain of atomic-distro users repeatedly reinstalling environments—a real developer workflow improvement. 74 HN points acknowledge the need. It could become a standard Linux dev environment tool, but differentiation and maintenance sustainability are risks against mature options like Toolbox and Distrobox.
One of the things that Windows really got right is WSL2. I drive an atomic Linux distro for daily use, but wanted a way to develop with multiple different distros with that same WSL UX. NSL is my answ…
B esp-sdr Turns a cheap chip into an SDR using ESP32's undocumented raw I/Q capture, a genuine hardware-level hack with high novelty. It could spawn low-cost radio monitoring and teaching networks, but the risk is undocumented reverse-engineered features being killed by firmware updates anytime, low duty cycle limiting practical use, and commercial use requiring dedicated SDR chips.
ESP-SDR uses the undocumented raw I/Q capture functionality of ESP32 family chips to turn them into (low-duty-cycle) Software Defined Radios
B heapspace Uses Three.js and ClrMD to turn .NET memory dumps into rotatable 3D visualizations, hitting the pain that heap structure is hard to grasp intuitively in .NET performance troubleshooting. The .NET ecosystem lacks such visualization tools, making this scarce supply, with a path to becoming a paid diagnostic tool like JetBrains or Razorpop. The risk is a narrow audience and a .NET community smaller than JVM's.
Heapscape: a local Three.js and ClrMD explorer for .NET memory dumps
B RuiC-motion-reel Generates 15-second animated films from code, with vector supersampling, point cloud and color-separation rendering, outputting 1920×1080/30fps plus a finished-film directory. Suited to people making data-visualization shorts without opening AE; render quality and the actual effect of the "15 styles" need the finished films to judge.
Generate 15-second motion graphics films with code · Agent Skill. Delivers 1920×1080 / 30fps. Self-built rendering engine: 2× supersampled vectors/text + true 3D point clouds + screen-print four-color…
B nuyoah-outfit-asset-generator Generates a full asset board with separated parts from clothing reference images, directly serving game and virtual fashion production pipelines, where part-splitting consistency is a real industry pain. As an MIT Skill it can embed in agent workflows and could become a standard skill for digital asset generation. The risk is unstable generation quality and part-splitting precision, with similar 3D clothing generation approaches evolving fast.
MIT Skill that generates a full outfit and a separated-parts asset board from a clothing reference image
B Show HN: Jevstiller – Distill Jev into a local model, with a disagreement bound Jevstiller proposes distilling Jev into local models with a 'divergence upper bound' guarantee, adding verifiability to lightweight decision models—a rare piece of rigor in the Jev ecosystem. 59 HN points show community interest. But the theoretical guarantee is far from production-ready, and the whole Jev paradigm is unvalidated by the market, with an unclear commercialization path.
B Show HN: Real-time Solar System with 526k asteroids and all tracked satellites Turns 526,000 asteroids plus all orbiting satellites into a real-time 3D visualization, with the data pipeline and rendering scale showing real engineering capability. 81 HN points prove its appeal for public and educational scenarios. Commercialization lies in space education and B2B satellite operations visualization, but such one-off demos rarely grow into standard dependencies; it depends on sustained maintenance and data updates.
B local-enough Estimates whether a background AI task can run on your own hardware: accuracy, p50/p95 latency, cost per thousand runs, and break-even volume, plus a router that picks the cheapest model meeting a quality bar. It's practical for small teams making localization decisions; the catch is needing labeled data first.
Measure whether a back-office AI task can run on your own hardware: accuracy vs gold, p50/p95 latency, $/1k tasks, amortised hardware cost and break-even volume, plus an OpenAI-compatible router that …
B jevwright End-to-end tests written as user operation steps, where AI only locates controls the first time and then runs via recorded playback, with assertions still decided by code. Attractive to QA teams tired of fragile selectors; but there's no data on locator accuracy or playback stability, and strong rivals like Playwright are ahead.
Browser tests for business flows, written as the steps a user takes. AI finds the controls once, recordings replay for free, and your code decides the verdict.
B Show HN: Pac-Bench – How well can models one-shot a Pac-Man game? Pac-Bench evaluates models head-to-head with a single prompt to generate a Pac-Man game, a lightweight benchmark of harness plus model capability. Its HN score of 78 shows interest in one-shot generation quality, and it could gain momentum as a cited community benchmark, but the task is too narrow and easily replaced by more comprehensive coding benchmarks, limiting commercial value.
Benchmarks how well Harness+models can create a Pac-Man game from a single prompt: “Create a Pac-Man game in a single HTML page” Each model gets one shot — no follow-up prompts or fixes.
B robolings A rustlings-style robotics exercise set using only NumPy, covering kinematics, dexterous hands, diffusion policies, and flow matching, suited to people who want hands-on practice rather than reading papers. Its restrained scope and lack of heavy dependencies are strengths, but the exercise set is just starting with no automated grading loop, so completeness remains to be seen.
Small exercises for robot learning, rustlings style: kinematics, dexterous hands, control, diffusion policy, flow matching. NumPy only.
B XelImageFormats---50-image-formats-for-TPicture-webp-jp2-heif-avif-heic-psd...- Adds decoding support for webp, heif, avif, psd and 50+ other formats to Delphi's TPicture, with a concrete need and clear boundaries. Such libraries are scarce in the Delphi ecosystem and it works once installed; the cost is that it only matters to Pascal desktop developers.
XelImageFormats - 50+ image formats for TPicture (webp, jp2, heif, avif, heic, psd...)
B pixeltop A real-time system monitor for the Omarchy desktop, emphasizing theme-matched colors and inspired by btop. A single Go binary, install-and-go, with clearly stated boundaries; the only selling point is theme adaptation, since such monitors are plentiful.
A live system monitor for Omarchy that puts your theme first. Inspired by btop.
B Nostalgex Wraps Plex, Jellyfin and Emby libraries into a retro cable TV schedule, with a native tvOS app plus a web tuner, sure to amuse self-hosting fans. The idea is likable and the boundaries are clear, but it is only an early version, and library compatibility and actual experience are unverified.
Retro cable TV guide for your Plex, Jellyfin or Emby library. Native Apple TV app + web tuner.
B Scrubbers A set of twelve feel sliders for SwiftUI, callable with one view plus one enum, with restrained naming and scope. It is not a new problem, but it breaks down drag feel cleanly, so iOS interaction people can copy it directly; the limitation is that it is more a visual experiment than a general control library.
Twelve sliders you'll want to touch, for SwiftUI. One view, one enum: Scrubber(.jelly, value: $value). Liquid Glass, jelly, elastic, fluid, squiggle, thermostat, swing, mood and more.
B chatgpt-receipts A Chrome extension that labels each ChatGPT reply with the model that actually answered, and warns when it's silently switched. A small, precise pain point for people who depend on a specific model; the limits are that it only covers the web UI and relies on page structure, so it may break at any time.
Chrome extension that shows which model actually answered each ChatGPT reply, and flags it when you get rerouted.
B OpenRayneo A Bluetooth bridge for RayNeo iO glasses on macOS, providing local HTTP notifications and teleprompter control. It targets a specific gap in specific hardware, with use and limits clearly stated; the audience is only Mac users of these glasses, but for them it's nearly the only choice.
Unofficial macOS Bluetooth bridge for RayNeo iO glasses: local HTTP notifications and teleprompter control.
B limen A local proxy in front of coding agents that tracks what they resend and compiles scaffolding out of prompts, non-intervening by default. The idea goes a step deeper than typical token panels; for now it's only measurable, so you'll have to test how much context it saves.
Local, opt-in proxy for coding agents: measures what your agent re-sends, compiles the scaffolding out of the prompt, and enforces nothing until you say so.
B china-market-data Wraps A-share quotes, candlesticks, fund flows and announcements into a CLI skill for agents, a pragmatic direction: letting Claude Code query data directly. Long-term usability depends on the data source, which is not explained; similar needs are currently mostly handled by libraries like akshare.
China market data / China stock market data Skill: A-share quotes, K-lines, tick data, research reports, capital flows, announcements, ETF/futures/macro. A Python CLI for Claude Code and AI Agents. Ch…
B xiaomi-LYWSD03MMC-to-solar-meter Turns an idle Xiaomi temperature and humidity sensor into a real-time photovoltaic power meter, showing generation power and self-use ratio in Home Assistant. Extremely niche but extremely specific, exactly the kind of hardware hack you can follow after ten minutes; the prerequisite is having the same sensor and solar setup.
Turn a Xiaomi LYWSD03MMC humidity monitor into a live solar meter for Home Assistant – PV output in kW and self-consumption %
B steno A local Whisper dictation tool for macOS that records while you hold a hotkey, also handling meeting notes, ready to use once installed. The need is real but the field is crowded, with VoiceInk, MacWhisper and Superwhisper all doing the same thing, and no differentiator is visible here.
Free, open-source dictation and meeting notes for macOS. Hold a key, speak, and your words appear at the cursor – transcribed locally with OpenAI Whisper.
B iigs-doom Running Doom on the Apple IIgs, requiring a ZipGS or TransWarp GS accelerator—a hardcore retro port. Worth a look for those into the 65816 platform and extreme ports; practicality is near zero and few can run it on real hardware.
Can it play Doom? Doom on the Apple IIgs with a ZipGS or TransWarp GS accelerator.
B pi_plays_pokemon Lets a coding agent play Pokemon from pixels and keypresses, essentially an end-to-end agent sandbox. The gameplay is not original, since Claude and Gemini have played it, but the code is readable and can be turned into your own agent benchmark, and with the author's reputation it is worth ten minutes.
pi (coding agent) plays Pokemon from buttons to pixels
B noterepo A local-first macOS journal app written by Flavio Copes, who has a long writing reputation, so the code style is probably fine. But the local notes space is already packed tight; unless it clearly differs in launch speed and pure keyboard flow, there is no reason to migrate.
A quiet, local-first daily notes app for macOS
B Sims3-ApexRadiance A lighting and performance mod for The Sims 3, with a description specific enough to mention nighttime light placement, water reflections and multi-core texture compression, showing real work on an old engine. For people still playing the third game it is a solid improvement, but it piles on many features, so stability must be tested yourself.
Lighting, visuals and performance mod for The Sims 3. At night, lamps really light the ground, lots, walls, roofs and snow; ponds glow and mirror their shore. Fewer stutters: faster file lookups, lot …
B vocab-flashcards A single-file HTML flashcard app: parses PDF/PPTX/DOCX locally to extract words, has a built-in offline dictionary, and looks up words via Baidu and Youdao. It's genuinely ready to use for vocabulary learners, but similar tools are plentiful; its only edge is direct access in China and single-file distribution.
Single-file English vocabulary flashcard tool: locally parses PDF/PPTX/DOCX to extract words, with a built-in offline vocabulary and online lookups via Baidu Translate and Youdao Dictionary through do…
B adu-motion-video An agent skill that assembles narration, subtitles and audio into editable HTML/JS scenes and exports local MP4, with export validation—more substantial than pure prompt collections. Still, it's in the crowded short-video auto-generation lane, and output quality is unclear.
AduNext motion-video skill: editable HTML/JS scenes, talking-head layouts, captions, audio and verified local MP4 export.
B steam-frame-nixos Gets a Valve Steam Frame VR session running on NixOS, an experiment explicitly labeled a spike, honest and technically substantive. It has reference value for the NixOS-and-VR intersection, but coverage is extremely narrow and upstream changes could topple it anytime, making it more notes than tool.
NixOS with Valve's VR session on the Steam Frame (spike)
B floorplan-3d From the name, a web tool that turns 2D floor plans into 3D views, a likable direction that hit hundreds of stars in a day, showing real demand. But the repo has no description at all: input format, export support and accuracy are unknown, and stars rising faster than docs is all the more reason to wait for it to explain itself.
B zhijian-receipt A pure front-end receipt generator that auto-calculates prices, exports images and high-res PNGs, handy for small shops or people making mockups. Single-purpose but honestly described; there are already many similar online receipt tools, and no differentiation is visible.
Paper Space · Receipt Generator | A simple receipt-making tool with automatic calculation, image import, local drafts, and high-definition PNG export.
B go-link Lets a computer run MAME and pushes video, audio and voice over WebRTC to friends in a browser, so online arcade play needs no client, and the Go plus Pion choice is fitting. But the claim of not needing ROMs needs explanation, and remote co-op already has mature alternatives like Parsec and Fightcade.
Play classic arcade games online with friends, straight from the browser. Your computer runs MAME (libretro) and streams video, sound and voice over WebRTC; friends join with a link and a PIN, no ROMs…
B milk A lightweight X11 desktop and window manager written in Odin, pitched as beginner-friendly. Choosing Odin for a WM is interesting, but the WM space is saturated, and "beautiful, beginner-friendly" are self-assessments with no config examples to back them up.
A lightweight, beautiful, and beginner-friendly X11 desktop and window manager.
B xss-specialist A continual-learning XSS detector that uses the KEV list as admission and verifies findings with browser execution, turning AI vulnerability claims from guesses into reproducible evidence for bounty and pentest automation budgets, with 116 stars in 1.2 days showing good momentum. It could sell an API or integrate into pentest platforms, but general agents plus browser automation already cover most cases, so it must stand on accuracy benchmarks.
KEV-gated continual-learning XSS specialist with execution-authoritative browser verification
B jeff A fine-tuned zero-shot text classification model meant to replace frequent classification API calls, drawing 219 stars in 0.3 days and showing strong developer demand for low-cost local classification driven by bills and data egress. It could become hosted inference or a classification SDK, but general embeddings plus few-shot prompting already cover most cases, and the casual naming and questionable reproducibility mean benchmark data is needed.
Fine-tunes of Qwen3.5 and Gemma 4 for zero-shot classification
B casita A content-addressed object store in Rust from the Cachix team that can serve as a build cache, artifact repository and deduplicating storage base. 60 stars in 1.9 days is modest, but the company behind it has a clear path: become a standard dependency in the CI/build ecosystem, then monetize a hosted service. The risk is mature alternatives like casync and cloud vendor artifact stores.
A content-addressed object store, written in Rust.
B iterleak It statically checks for goroutine leaks from uncalled iter.Pull stops, targeting a new pitfall introduced by Go 1.23 iterators—a small, precise cut. The audience is limited to iter package users, and it only pays off once wired into CI.
Go analyzer that reports iter.Pull and iter.Pull2 calls whose stop function can go uncalled and leak a goroutine.
B system-design-trainer A system-design interview whiteboard tool with a real-time capacity estimation model, closer to the real interview process than pure problem drilling, drawing 93 stars in 1 day during job-hunting season. Mock-interview subscriptions or training lead generation offer a clear monetization path, but educative, Exponent and interviewing.io already have mature offerings, so it must differentiate through the real-time model and interaction.
Whiteboard for practicing system design interviews with a live capacity model
B Show HN: Free alternative to graphics design giants A WebAssembly-based browser raster and vector design tool, free without registration, with 101 points driven by strong sentiment against Adobe subscriptions and high immediate value. It could pursue traffic-based ads or paid team collaboration, but Photopea, Figma and Canva AI already own mindshare, and the free no-account model itself lacks a clear paid conversion path.
A webassembly web based app that help editing raster and vector on browser without any subscription or signups
B Show HN: PaperMono, e-ink fridge magnet shopping list with mobile web page Turns an M5Stack e-paper display into a fridge shopping list synced to a phone web page, with 127 points coming from the story of real household daily use and fully vibe-coded development rather than the technology itself. It could extend to hardware kits or template sales, but it is essentially a personal hobby project and would need reminders, recipes and other scenarios to become a product; the M5Stack ecosystem is niche.
Fridge magnet shopping list on an M5Stack PaperMono (ESP32-S3, e-ink touchscreen), synced with a phone web app over Wi-Fi. Works offline, ~2,400 lines of C++. This is a great wee device! BLE, Wi-Fi an…
B mixtape An app store for OnionOS on the Miyoo Mini Plus, letting you browse, install, and update community ports over Wi-Fi, saving manual file copying. The audience is small but real.
App store for OnionOS on the Miyoo Mini Plus: browse, install and update community ports over Wi-Fi. Built by Claude.
B CnC-FPS-Unlocker Unlocks Red Alert 3, C&C3 and Kane's Wrath above 30 FPS while keeping normal game speed, a specific small fix rather than a cheat that old players benefit from immediately. Compatibility with the Tacitus mod platform is a plus; the limitation is that it only covers three old titles.
Run Red Alert 3, C&C3 Tiberium Wars and Kane's Wrath above 30 fps at normal game speed. Works with Tacitus.
B byterange It focuses solely on RFC 9110 Range parsing and limits range counts and multipart amplification attacks, a rarely handled detail in Go file serving; restrained in scope and accurate in description, but the standard library plus a few lines could do it too.
RFC 9110 HTTP Range parsing and serving for Go, with limits on range count and multipart amplification.
B seiso Defines a Markdown spec and linter for AI-generated project documentation, addressing the new problem that neither humans nor agents can read AI docs well, with 115 stars in 1.4 days reflecting a real pain point for AI-heavy teams. If it becomes a de facto CI standard it could support team subscriptions, but the need is still new and may be absorbed directly by documentation frameworks or native IDE rules.
A Markdown convention and linter for project docs written by AI and read by humans and agents
B go-acs A single-binary TR-069/CWMP ACS with embedded SQLite for managing ONTs and FTTR hosts: view status, change config, run diagnostics. Self-hosted ACS has long had only GenieACS as an option; this is precisely described and easy to deploy, suited to home-network tinkerers.
Single-binary, zero-external-middleware TR-069/CWMP ACS (Go + SQLite): manage ONTs and FTTR hosts — view info, change config, run diagnostics, manage sub-devices and terminals
B tekken3-expanded Puts all 18 Tekken Tag Tournament characters into Tekken 3, playable in every mode with matching moves, a substantial mod. The catch is you need the original disc and arcade ROM, so the install bar is high.
The 18 Tekken Tag Tournament characters in Tekken 3: playable in every mode and fought by the CPU. Tekken 3's cast can use their TTT movesets. Built from your own Tekken 3 disc and arcade ROMs.
B papermono-shopping-list An e-paper shopping list firmware with a FastAPI service and phone web sync, corroborated across GitHub's 140 stars and HN's 127 points, with the author saying the family uses it daily. It could become an M5Stack template or paid kit, but it remains a single-scenario personal project, and maintenance and category expansion will decide whether it moves from demo to reusable product.
E-paper shopping list for the M5Stack PaperMono, with a FastAPI server and phone web UI
B tiny-classifiers Distills a 17M text classifier from large-model labels, trains in a minute, and quantifies when it beats an API, targeting cost and data-privacy pain points with a tutorial and 10-task results. It could become classification-as-a-service or an embedded SDK, but general embeddings plus few-shot prompting already cover most cases, so credible benefit data is needed.
Train a 17M-parameter text classifier in a minute, with labels from a big AI model, and know when it beats an API. Recipe, tutorial, results on 10 tasks, phone trainer.
B st8ks A cross-platform Kubernetes desktop client whose selling point isn't another GUI panel but actively pointing out what's broken in a cluster and guiding fixes. The author has a reputation for terminal tools and the description is restrained; it faces direct competition from Lens, Headlamp and Aptakube.
A fast Kubernetes desktop client for Linux, macOS and Windows that finds what is broken and helps you fix it
B aztree Parses Azure bills to explain cost composition, targeting FinOps as a real enterprise paid pain point, with 56 stars in 0.8 days showing decent growth while Azure billing is complex and official attribution is weak. It could be SaaS with savings-based pricing or consulting lead generation, but existing tools like Azure Cost Management and Infracost already cover ground, so it needs differentiation in anomaly detection and attribution granularity.
Understand Azure billing
B ufo-core Positions itself as a business agent operating system aiming to unify enterprise process automation, a direction with clear budgets but an extremely crowded field. Momentum is modest at 81 stars in 1.7 days, with no unique data, channel or deployment cases; without top-customer endorsement it will struggle to become a standard dependency against orchestration frameworks and big-vendor platforms.
Business agent operating system
B nestjs-quota A NestJS library for API quotas and usage metering, clearly positioned to fill the gap where Throttler handles rate limiting but not metering. Small teams can pick it up directly, but with only 8 stars and no integration examples or load tests, the production-grade claim is unsupported.
Production-grade API quota enforcement and usage metering for NestJS
B arc-draw It narrows SAP architecture diagrams: describe in natural language, and an MCP service places BTP components using official icons and connection syntax. A must-have little tool for SAP solution architects, restrained in scope and honest in description; the limitation is that its audience is almost only that group.
ARC-DRAW is an SAP-aware Model Context Protocol (MCP) server for SAP Solution Architects. You describe an architecture in plain language; the AI agent places the correct SAP BTP icons and follows the …
B private-diary A fully offline calendar-style private diary as a PWA plus Android APK, with no accounts and no network, aimed at people who just want local storage; the description is clear and not oversold. The limit is that local diary apps are plentiful, and encryption and sync details are not covered, so there is no clear reason to pick it.
A fully offline, calendar-style private journal — PWA + Android APK. No account, no network, bilingual (中文 / English).
B Rapid-CRM An open-source billing and subscription analytics dashboard for Shopify app developers, covering plans, usage billing and MRR/churn with a restrained scope. It saves independent Shopify app builders from building their own billing backend; the main risk is that connecting to the Partner API is unverified.
Open-source billing and analytics platform for Shopify apps: plans, subscriptions, discounts, usage-based billing, and MRR/churn/LTV analytics from Shopify Partner data
B airsup-lab Disassemblable procedural 3D models of a fusion reactor, Raptor engine, and F1 car in the browser, aimed at science display and Three.js learners, selling the ability to take things apart rather than flashy rendering. The limitation is that detail and physical plausibility are unstated, making it more portfolio than reusable library.
Machines you can take apart in your browser: fusion reactor, Raptor 3, Cybertruck, humanoid, black hole, fighter jet engine, F1 car, drone. Three.js, procedural, real physics.
B testdrive3-enhanced It makes an SDL3 native port of 1990's Test Drive III, adding resolution and view distance—a sizable effort, honestly described; but you must supply the original assets and compile it yourself, and similar retro ports are already plentiful.
Enhanced native SDL3 port of Test Drive III: The Passion (1990) - smooth high-resolution 3D view, longer draw distance, smooth steering
B openOMSI Claims to rewrite OMSI 2 from scratch in Rust while staying compatible with all maps and mods, requiring only a legitimate copy of the base game. The direction is tempting, but claiming full compatibility a day after the repo was created is nearly impossible to deliver, more a vision post. Check whether a runnable build exists before spending time.
OMSI 2 recreated from scratch in Rust, compatible with every map and mod (requires the original OMSI 2)
B cf-mp3-player A small C embedded hardware player for MP3/WAV/FLAC using CF cards as storage, of reference value to people working with microcontrollers and audio decoding. The limit is heavy hardware dependence with no schematics or bill of materials, making it hard for others to reproduce.
A simple MP3/WAV/FLAC player using CF card
B minimax-h3-lora-colab-skill It packages MiniMax H3 LoRA training as an installable Codex skill running on Colab, with a clear goal and concrete steps; but the audience is very narrow, it depends on a specific model and compute, and it's just a thin layer of glue.
Installable Codex skill for training MiniMax H3 Ref2VA LoRAs on Google Colab
B snap A small tool that makes you vanish from Google Meet with a snap, triggered by microphone detection of a pop. The idea is fun and the implementation clean, but its usefulness is narrow—a few uses per meeting at most—making it a good-looking demo.
vanish from google meet with a finger snap mid meeting
B handdrawn-diagrams A Codex skill for hand-drawn-style technical diagrams with an Excalidraw look and local PNG rendering, with a specific direction. But similar drawing skills have clustered lately, results depend entirely on prompts, and the repo itself offers no evidence.
Chinese hand-drawn technical mechanism diagram Codex Skill: Excalidraw style, few words, clear hierarchy, with a local PNG renderer.
B smartatelier Local inventory for electronic components and 3D printing filament, with photos, videos, colors, and project links, plus a mention of AI assistance with human verification. The target is concrete (small maker studios), but a spreadsheet could do the same job, and the AI part only adds complexity.
Inventaire local de composants électroniques et filaments 3D : photos, vidéos, couleurs et projets avec assistants IA et validation humaine.
B OpenCrossing-Xbox A native Xbox port built on the Animal Crossing GameCube decompilation, with real engineering effort and technical difficulty. Worth a look for console reverse-engineering and porting fans, but it is currently WIP with no playable build or build instructions, so ordinary developers cannot use it.
Native original Xbox port of Animal Crossing (GameCube decomp)
B MoniVol A macOS menu bar utility that adjusts volume for external HDMI/DP monitors via DDC, with a clear scope and honest naming. The problem is real, but MonitorControl and BetterDisplay already cover the same need, and 8 stars hardly shows a difference.
Free, lightweight macOS menu bar app that enables system volume control for external HDMI and DisplayPort displays.
B genpark-speculative-decoding-verifier-skill A verifier for draft-target speculative decoding with rejection sampling, acceptance-rate telemetry, and speedup estimation, aimed at teams running their own inference. The technical points are concrete, but the audience is very narrow, it is packaged as an agent skill, and install steps and benchmark data are missing.
Draft-target model speculative decoding verification engine with rejection sampling, acceptance rate telemetry, and dynamic speedup ratio estimation.
B arena-skill Has Claude generate hundreds of strategies for the same task and pit them against each other to pick one, distributed as a Claude Code skill. The idea is entertaining and thought-provoking, but it is essentially prompt orchestration, and the fight-style marketing tone outweighs any explanation of results.
When Claude keeps giving you bad answers, make 100 versions of it fight to the death. Same task, 100 different strategies, a bracket, one answer left. Free Claude Code skill.
B craftsail-growth A self-hosted AI SEO dashboard that tracks how ChatGPT, Perplexity and others mention your brand. The problem is genuinely new and commercial equivalents are mostly paid, but the intro is a stacked feature list full of growth talk and hasn't proven it can actually capture answer citations.
Self-hosted GEO & AI SEO dashboard. Track how ChatGPT, Claude, Gemini, Perplexity, DeepSeek and more mention and cite your brand, compare share of voice with competitors, audit your site for AI crawle…
B Apple-Tiles It assembles Bad Apple animation from multiple synchronized Win32 windows, cleanly implemented and described in one line. "Bad Apple running on X" is already a fixed meme, so it's more spectacle than utility.
A recreation of the classic **Bad Apple!!** effect rendered as an animated mosaic of apples, using multiple synchronised Win32 windows.
B Polymarket-Perpetual-Bot A market-making, edge-detection and cross-market arbitrage toolkit around Polymarket's CLOB, with paper trading and risk controls, drawing 191 stars in 2 days on prediction-market hype. It could sell signal subscriptions or strategy SaaS, but regulatory positioning and Polymarket's official API policy are hard constraints, and an unofficial community project will struggle to charge compliantly.
Polymarket trading bot for prediction markets — browse CLOB markets, watch the order book in the terminal, run edge detection, liquidity provision and cross-market arbitrage strategies with paper trad…
B logo-design-skill Packages logo design methodology, SVG craft, testing tools and a 1,400+ case library into a cross-agent skill, with 181 stars in 1.7 days showing agent skills are a new distribution channel. It can be sold directly, serve as a design-tool acquisition funnel, or become an agent asset library; the risk is the case library being distilled and copied and general models already generating images, so the testing tools and structured process must carry the moat.
A comprehensive logo-design skill for Claude, Gemini CLI, Codex and other AI agents: principles, process, SVG craft, testing tools and a 1,400+ logo reference library.
B brewreel Writes a product brief, lets a cheap model like DeepSeek pick shots and write copy, and outputs a vertical promo video in one command, with 3 recipes, 6 industries and ad-law checks aimed at SMB marketing. 66 stars in 1.9 days makes it one of the most directly monetization-oriented projects here, chargeable per video or as SaaS; the risk is unstable output quality and margin pressure from big-vendor template tools and model upgrades.
Craft BrewReel: lets cheap models like DeepSeek make good-looking vertical promo videos. Write a product brief, AI picks shots and writes copy, one command to produce the video. 3 recipes, 6 industrie…
B rawmakase A free Lightroom alternative written in Rust covering Linux and macOS, with 96 stars in 1.4 days reflecting photographers' long-standing desire for a modern RAW editor. GPU-accelerated batch processing and non-destructive editing could support a buyout or subscription creator-tool business and a shot at becoming a standard photography workflow component, but darktable and RawTherapee are free incumbents, Lightroom's ecosystem is sticky, and cross-platform C++ engineering is huge.
Free Lightroom alternative for Linux and macOS (possibly Windows too - not tested)
B Show HN: TinyAIArena watch AI agents battle it out Four models battle in real time on an 8×8 grid with spectating and replays, turning AI benchmarks into entertainment, and 93 points show real demand for AI spectating. It could monetize as an engagement display layer, model-evaluation marketing or a white-label API, but entertainment is strong on novelty and weak on return visits, and leaderboards like LMArena could add battle modes and crush it.
Did you ever click on an “AI Arena” expecting glorious battle and instead get a boring benchmark? If so, this project is for you: proper life-or-death fights between four models on a picturesque 8×8 g…
B x-scraper-no-api Scrapes X data using your own browser session, exports LLM-ready JSON/CSV and ships an agent skill, hitting the free-API-quota pain point with 120 stars in 1.9 days. It could monetize as a data subscription or scraping API, but X's escalating anti-scraping, account bans and competition from existing scraping services mean compliance risk keeps it from becoming a standard dependency.
Free, self-hosted X (Twitter) scraper no API key, no credits. Uses your own browser session via Playwright. Export tweets, search results & timelines as LLM-ready JSON/CSV/Markdown. Ships an AI agent …
B awstree Visualizes AWS bills as an understandable tree, with 68 stars in 1.2 days. FinOps is real budget and enterprises pay to save money, so the monetization path (team SaaS, bill analysis API) is clear; the risk is that mature competitors like infracost and CloudHealth already occupy the space, so it needs more intuitive bill attribution and cross-cloud support to differentiate.
Understand AWS billing
B Show HN: Lofi Cities – Pixel-art city nights with browser-generated lofi A pixel city night scene with browser-generated real-time lofi music, where 142 points prove shareable ambient experiences spread well. Practical value is limited, but as a consumer product it has clear monetization options: scene-pack subscriptions, musician or brand collaborations, and API licensing; the risk is Lofi Girl and others already occupying the space, return visits depending on continuous content, and a low ceiling.
B VitaOS A modern homebrew launcher for the PS Vita, gathering games, video, music, apps, and a homebrew store into one menu. The handheld homebrew scene has long lacked a good-looking launcher, and the author keeps the scope tight; the limitation is serving only the small circle still tinkering with the Vita.
A modern home screen for the PS Vita: games, films, music, apps and a homebrew store behind one fast menu.
B 8cryptdo Reverse-engineering notes and companion tools for 8BitDo controller firmware encryption, written for people who want to modify firmware or study its update protocol. The topic is niche but concrete, with docs and scripts together; the limit is a very narrow audience, and it may break after firmware updates.
Reverse-engineered docs and tools for 8BitDo firmware encryption
B lemo-opuscar A collection of prompts and code samples for 39 film styles, essentially a prompt catalog letting users pick a style for an agent to direct. 348 stars in 1.6 days is decent momentum, but content repositories have no network effects and are easily copied, so monetization is limited to courses, templates or funneling to its own generation tool; style copyright disputes and upstream model updates quickly make prompts obsolete.
39 film styles, each a reusable style prompt plus a short film made entirely in code by Claude Opus 5.5. Pick a style, bring your own story, and let your agent direct. | Opus5.5 x 39 种影片风格：风格提示词 + 纯代码…
B Trade-Bot-Hyperliquid A read-only Hyperliquid account health checker: computes ROI, win rate, and drawdown, and identifies martingale/grid position-adding. Useful for copy-traders or self-risk checks, and it explicitly states it doesn't trade or need API keys; the limitation is it covers only one exchange, and you must verify the metric definitions yourself.
Independently analyze ROI, win rate, and PnL using the public Hyperliquid API, detect martingale/grid recovery strategies, and assess maximum drawdown and risk-adjusted returns. Read-only terminal che…
B paperlight Local full-text document search written in Tauri and Rust, finding files by name, directory, or body text inside PDFs and Office files, read-only and offline. The need is real and the description honest, but DocFetcher, Recoll and similar mature options already exist, so it needs stronger differentiation.
Every document on your PC, one search away. Find PDF, Word, Excel and PowerPoint files by name, folder or the text inside. Fast, private, read-only. Built with Tauri + Rust.
B memory-forest An Anki plugin that grows your daily review record into a forest, one tree per day. Small, restrained, and genuinely installable, it's a pleasant positive-feedback device for long-term card reviewers; the feature itself is light, a nice-to-have.
Memory Forest: an Anki add-on that grows a forest from your study history, one tree per day.
B skryba A macOS dictation tool where you hold Fn to speak and text lands in any input field, backed by Groq Whisper. Honest description and install-and-go, but Whisper dictation apps are already plentiful, and it doesn't explain why it's worth switching from something like Superwhisper.
Hold Fn, speak, and your words appear wherever you type. A small macOS dictation app using Groq's Whisper.
B Layerbase-Sequence-Player A lightweight image sequence player and converter supporting EXR, color space conversion, and cryptomatte. Compositors reviewing shots need this combination, and old tools like DJV and mrViewer haven't moved in years; the limitation is only C++ source, with no build or performance details.
A fast and light image sequence player and converter, supporting exr, color space transforms and cryptomatte
B comfyui_ensemble_upscale A ComfyUI node that upscales images with one to three ESRGAN models, avoiding the detail-smearing of simple averaging. The problem is defined specifically, showing the author understands existing flaws; the limitation is working only inside ComfyUI, with results still needing your own comparison.
ComfyUI node that upscales an image with one to three ESRGAN models without averaging the detail away.
B rain-go An MCP game platform where humans and AI play cards and board games at the same table: 10 games, seat-based multiplayer, server-side isolation of hidden cards. Using MCP for entertainment rather than tool calls is novel and it addresses cheating, but game depth and online deployment are unverified.
West Window · an MCP game platform where humans and AI play board and card games at the same table · 10 games, seat-based multiplayer, server-side hidden-hand isolation
B OmaPhoto Ports the macOS layered image editor Compositor to Linux/Omarchy, with 172 stars driven mainly by Omarchy community buzz, filling Linux's gap for a modern layer editor. It is practical but has a narrow audience and low willingness to pay, more like a personal project, and must compete with GIMP/Krita and other new editors for the same users.
Layered image editor for Linux and Omarchy, a port of Compositor
B TostAI-Sprite-Sheet-Studio A tool that converts raw video into game-ready sprite sheets for pixel or action game developers, saving frame-by-frame cutout and alignment. Clear positioning and low barrier; output quality depends heavily on input video, and the repo offers no comparison samples.
Turn a raw video into a game-ready sprite sheet
B sakura-print A small service that lets phones print and scan through a home Linux/CUPS printer, solving the common iPhone-can't-find-the-scanner headache. The idea is simple and the license clear, but it heavily overlaps with the existing CUPS+Avahi and scanservjs combination.
Print and scan from phones through one home printer (Linux, CUPS, AirPrint). GPL-3.0-or-later.
B hmg_nemesis Hands-on research on rooting modern Kia Gen5W head units with OTA, real material for automotive security researchers rather than a wrapper. But the audience is narrow, it needs the matching model and hardware, and risk and rollback notes are missing; it gathers scattered owner-forum posts into one repo.
Gain control of your Hyundai Motor Group vehicle, NEMESIS provides a way to execute software as root. Compatible with Gen5W AVNT systems with OTA update support.
B BlutterAndroid Brings Blutter into an Android app for on-device Flutter/Dart AOT reverse engineering, explicitly listing 70 Dart versions from 2.14 to 3.13, a restrained scope. It is a real addition for Flutter reverse engineers; the limit is a very narrow audience, and you need a target APK to verify it.
App-integrated Blutter: on-device Flutter/Dart AOT reverse engineering for Android (Dart 2.14~3.13, 70 versions)
B omarchy-sequencer A TR-808-style step sequencer for Omarchy, with built-in drum kits, sounds, per-track effects, and recording; the description is specific and not oversold. It's a fun little tool, but the audience is limited to users of this Arch derivative, and LMMS and Ardour already cover the functionality.
A step sequencer for Omarchy: TR-808 drums, riffs, 90s rave sounds, effects per track, free sound packs and recording
B ServerBox Turns an old Android phone into an always-on Linux server without root, with optional Alpine/Debian/Ubuntu and built-in SSH. The need is real, but the approach overlaps with proot-distro and Linux Deploy, and it collides with a well-known server monitoring project's name, causing confusion.
ServerBox: turn a used Android phone into a Linux server, no root required. Alpine, Arch, Debian, Kali and Ubuntu with an always-on SSH server.
B openhouse-3d It solves camera poses from listing photos, procedurally models in bpy, then renders a walkthrough video; the pipeline is concrete enough for real-estate agents or virtual-tour builders to follow directly. Compared with general 3D reconstruction tools it does only this narrow line, which is actually clear. Limits: it depends on Blender and a GPU, and shot quality depends on the input photos.
Turn real-estate listing photos into a 3D house model and a cinematic walkthrough video. Photo camera solving, procedural Blender (bpy) modeling, virtual staging, headless Cycles rendering and photo-v…
B ghostty-config A TUI for configuring the Ghostty terminal, turning options like themes and fonts into an interactive interface instead of hand-editing config files. The scope is small and clear and it works out of the box; the ceiling is low, and Ghostty veterans may not need it.
TUI for Ghostty terminal customization
B Multi-Agent-Game-Localizer A multi-agent game localization pipeline: Claude orchestrates, local small models translate, and patches are generated on-device. The approach is concrete and saves API costs, appealing for fan translation; the hard parts are terminology consistency and patch injection, and the repo hasn't provided test results.
Multi Agent game localizer: Claude as the Orchestrator, a local small model as the translation sub-agent, generating the localization patch on the local machine
B depcart Searches ten package sources including npm, PyPI, crates.io and Maven from a VS Code sidebar, letting you pick a version and get the correct install command, saving browser switching in multi-language projects. The feature is narrow but clearly explained; the problem is that plugins like Version Lens already cover most of it.
VS Code extension: search npm, PyPI, Go, crates.io, Maven, NuGet, RubyGems, Packagist, pub.dev and Hex from one sidebar, pick exact versions, and get the right install command for your project.
B DeadCells-PS5-Native Relinks the PS5 version of Dead Cells into a native Windows program rather than emulating it, an unusual and more aggressive approach than typical emulators. But it requires your own console game files, raising legal and availability problems; for now it's more a tech demo.
Dead Cells for PS5 as a fully native Windows program: relinked, not emulated. Built on AnyPS5.
B digislicer A C tool for slicing on the Elektron Digitakt Mk1, targeting the real pain of sample users chopping long loops into slices. But the README is one line, with no install method or supported formats, so it's useless outside the Digitakt community.
Slicer for the Digitakt Mk1
B shotcandy A pure frontend tool that turns screenshots into attractive shareable images, free, running in-browser, install-and-go. Cleanly done, but the space is crowded (Pika, Screely, Shots) and there's nothing notably new beyond self-hosting.
Turn any screenshot into a beautiful, share-ready image in seconds. Free, open source, runs entirely in your browser.
B Hasselblad-X-System-CIM-Firmware-Research-Feature-Extensions Independent research into focus, menus, and object recognition in Hasselblad X2D firmware, a rare look at high-end camera reverse engineering. The topic is fresh, but no conclusions or reusable tools are given, benefiting only a few Hasselblad users and firmware enthusiasts.
Independent research on Hasselblad X2D/X2D II firmware, autofocus, menus, and object recognition.
B unquarantine A menu bar utility that automatically clears the quarantine flag on downloaded files, saving you from typing xattr each time. The author writes it plainly, but it's essentially a wrapper around one command that also bypasses Gatekeeper's first-run check, a real security cost.
A menubar app that automatically removes quarantine flag from all your downloads
B lain-psx-decompiled A decompilation of the Serial Experiments Lain PS1 game plus a native cross-platform client, a niche reverse-engineering result useful to PS1 engine researchers or those wanting to run this obscure game on modern systems. The description is one sentence, so completeness and build steps need verifying yourself.
Decompilation of Serial Experiments Lain (PlayStation) and a native Windows, macOS and Linux client
B Red-Team-GOAD-Lab-Proxmox A setup guide for a GOAD red-team lab on Proxmox and pfSense, including C2 and redirectors. It saves effort for those practicing internal pentesting, but it's essentially documentation that can't avoid the official GOAD material and brings no new tools.
Build guide for Red Teaming home lab. GOAD lab setup in Proxmox and pfSense, Operator/C2 and Redirectors.
B MacStatusBarAndDock Adds a macOS-style menu bar, Dock and windows to jailbroken iPads, rootless and explicitly targeting iPadOS 15-16, with a cleanly stated scope. It is genuinely useful in a small circle, but the OS versions are old, the audience is only jailbroken users, and completeness rests on the description alone.
A macOS-style menu bar, Dock and windows for jailbroken iPads (rootless, iPadOS 15-16).
B hermes-virtual-office A 3D office for a fleet of AI agents: desks, meetings that auto-generate minutes, screen peeking and kanban sync, all stateless and easy to run. It's a strong demo but more toy than tool; real agent management still relies on logs and task queues.
A 3D virtual office for your AI agent fleet — desks, meetings with auto-generated minutes, live screen-peeking, and a Kanban board synced to Hermes. Next.js + Three.js, stateless, no DB.
B pokeemerald-3Ds-dualscreen Ports the Pokémon Emerald decompilation project to 3DS, adding dual screens, glasses-free 3D and a claimed fully 3D world. Decomp ports are already common, and a repo a day old claiming a full 3D remake deserves skepticism; wait for real hardware footage.
Native Pokémon Emerald port for Nintendo 3DS, featuring dual-screen gameplay, stereoscopic 3D, and a fully 3D world.
B okeylitctl A Linux controller only for verified HP OMEN four-zone RGB keyboards, with a modestly stated scope and no compatibility overclaiming. The problem is a tiny audience and an unexplained whitelist mechanism; openrgb covers similar needs more broadly.
Restricted Linux controller for validated HP OMEN four-zone RGB keyboards
B MagibuMizan Aims to read typed probabilities directly from open-source language models for decisions rather than generating text. The idea is unusual and appealing to those wanting classification or scoring alternatives; but it doesn't explain how to call it or what the output looks like, so it's closer to a manifesto.
Typed probabilistic decisions from open language models, without text generation
B Recat Collects security findings produced by agents like Hermes into a local workbench for review, evidence inspection and report export, a concrete scenario. But startup and asset models are unclear, and 8 stars in a day does not show real use; existing vulnerability management dashboards already cover most of this need.
A local workspace for security findings. Review findings from Hermes or other agents, inspect evidence, and export reports. Recat supports Web, Smart contract, and Other assets, including mobile, desk…
B slopbrake A proposal for putting brakes on agent-generated code: deterministic checks, re-review with fresh context plus direct fix commits, and one-way-door gating. The direction is worth watching, but it leans on "every rule has bitten someone" anecdotes with no install or integration path, so it reads as a manifesto.
Brakes for the software factory: deterministic checks, a fresh-context reviewer that commits its fixes, and one-way-door triage for coding agents. Built on Matt Pocock's skills. Every rule proven to b…
B Thinai Runs small models locally on Android and exposes a local inference API for other apps to call directly. The idea is not new—PocketPal and ChatterUI already occupy the space—and the local API is the only differentiator; with 9 stars in two days, the supported model list is still unseen.
Thinai is a privacy-focused Android app that runs small language models locally on your device, with a local API for integrating and accessing on-device AI inference.
B adaptix-kerbrute Puts Kerberos user enumeration and password spraying into in-process execution within AdaptixC2, avoiding dropped files and child processes. It's a practical stealth improvement for red teams, but essentially a kerbrute port; defenders will read it to learn evasion techniques.
In-process Kerberos user enumeration and password attacks for AdaptixC2 — no process spawn, no file drop.
B taiga-s1 Uses a 1.2M-parameter small model to generate CAD parts directly in FreeCAD, a specific idea with restrained ambition. But there are no samples of what such a small model can produce, so at this stage it's more an experiment log than something usable.
Taiga-S1: a 1.2M-parameter System-1 model that builds CAD parts in FreeCAD
B rovar A Figma-like desktop design editor written in Rust and GPUI. New stack and big ambition, but a repo just over a day old is mostly skeleton; the real difficulty in design tools is rendering detail and collaboration, and Penpot is a more realistic open-source alternative.
A Figma-like desktop design editor built with Rust and GPUI.
B OrbitalNote A cross-platform editor for Org-mode users, combining notes, calendar, files, and optional end-to-end encrypted sync. The direction is clear, but "Native preview available now" reads like pre-launch copy, maturity is unknown, and orgzly and Emacs already cover most needs.
Your notes. Your calendar. Your files. Free, open-source Org editor for desktop and mobile, with optional end-to-end encrypted Sync. Native preview available now.
B genspark-claw Provides 5 production-grade SKILL.md files plus a zero-dependency installer and MCP server for Genspark Claw, OpenClaw and Hermes, seizing the gap where the agent ecosystem lacks distribution channels; 69 stars in 0.9 days is strong momentum. The business path is clear: a skill marketplace or enterprise skill pack subscription. The risk is that cross-host protocols are not yet standardized and it could be absorbed by official skill libraries.
🦀⚡ Community skill pack & CLI for Genspark Claw, OpenClaw & Hermes agents — 5 production-ready SKILL.md skills (deep research, reports, slides, browser automation, AI call prep), zero-dependency Node.…
B muse2api Wraps a reverse-engineered Muse into an OpenAI-compatible interface covering chat, text-to-image and text-to-video with multi-account rotation and 48h auto-renewal; 51 stars in 1.4 days directly solves developers' integration cost for non-standard models. The business path is a model relay/aggregation service; the risk is that reverse engineering and ToS violations lead to upstream bans, and the compliance of the account pool model is the biggest variable.
Reverse-engineered Muse (muse.ai) wrapped as an OpenAI-compatible API, supporting chat, text-to-image, text-to-video/image-to-video, multi-account pool rotation and 48h auto-renewal. OpenAI-compatible…
B fastframe Provides a shared foundation layer (routing, state, window conventions) for egui apps, targeting the gap where Rust native apps lack framework consensus; 22 stars in 1.6 days is decent momentum. The business path is becoming the standard dependency for Rust desktop app scaffolding with hosting and enterprise support. The risk is that the egui ecosystem is still smaller than Tauri/Slint and the framework unification route is easily absorbed by official capabilities.
egui on rails: the shared foundation for native Rust apps built with egui
B agent-office Puts Claude Code workers in a 3D office with shared terminals, voice collaboration and synced GitHub issues/PRs, turning abstract multi-agent collaboration into a visual space, with 24 stars in one day showing decent growth. The business path is a team-level agent monitoring and collaboration interface via virtual-office subscription, but the playful shell struggles to replace Linear-style lists and depends heavily on a single agent runtime.
A cartoon 3D office where your team hires Claude Code workers at desks, shares live terminals, talks over voice, and tracks GitHub issues and PRs.
B deno-lockfile-diff It translates deno.lock changes into supply-chain signals—new raw URL imports, integrity changes, source migrations—exactly the most time-consuming and easily missed part of lock-diff review. 16 stars in two days is modest, but supply-chain security is something enterprises pay for, and it could grow into continuous CI auditing, competing with or being acquired by platforms like Snyk.
Turn a deno.lock diff into prioritised supply-chain signals — new raw-URL imports, integrity changes, registry moves.
B awesome-opus-5-5-videos With 150 stars in two days, it indexes 1000+ Claude Opus 5.5 generated videos by theme, visuals and production frame source, effectively a demand map and traffic entry for AI video creators. Momentum is among the highest in this batch. The business path is traffic referral, tutorials or production service matching rather than a dependency library; the risk is rapid obsolescence as models iterate and a pure list being hard to build on.
Source-linked guide to 1,000+ videos people made with Claude Opus 5.5: what they're about, how they look, how the frames were made.
B emu A proof of concept for running native macOS apps on iOS, which if feasible would open a huge iPad productivity gap; 22 stars in two days shows no small interest. But commercialization is stuck on Apple's sandbox and distribution policy, shipping as-is violates rules, leaving only jailbreak or enterprise signing channels; technically it relies on architecture translation and is exposed to Apple blocking long term.
A proof of concept for running native macOS apps on iOS
B evident-charts Makes coding agents follow honest charting standards when drawing charts and check each one before handing it over. The rules cover axes, scales and common misleading tricks, far more useful than draw it nicely, though it only supports a few hosts like Claude Code and Codex.
Teaches AI coding agents to make clear, honest charts, and checks each one before you see it. For Claude Code, Codex, Cursor, and more.
B Deobfuscator-Luraph-V15 Deobfuscates and devirtualizes Luraph v15-protected Roblox Luau scripts; 44 stars in 0.8 days shows script analysts need it, and the technical bar (IR recovery, constant decryption) forms a moat. Commercially it could be reverse engineering services or an anti-protection subscription, but it is in a constant arms race with new Luraph versions and the use is gray (cheats/plagiarism), with compliance risk limiting scale.
A standalone, high-performance deobfuscator and devirtualizer for Roblox Luau scripts protected by Luraph v15.
B claude-x-jev Uses TypeSafe's Jev classify/score/gate model to make high-frequency decisions for Claude, leaving the main model only read/write judgment, hitting the two pain points of agent pipeline latency and token cost; 16 stars in 1.4 days. The commercialization path is LLM routing and cost optimization middleware (LiteLLM-like). The risk is tight coupling to a single model provider and decision quality needing benchmark backing.
Fast, cheap, typed decisions for Claude Code. Jev (TypeSafe's decision model on OpenRouter) sorts, checks, scores, gates and verifies at 0.3s and a fraction of a cent per item. Claude keeps the readin…
B redis-guard Blocks dangerous commands like FLUSHALL and KEYS by policy before they reach production, a defense-line tool that saves you the one time it fires, with very low integration cost; two days and 20 stars matches a real pain point. The business path is clear: command auditing and policy control can be packaged as a DB security SaaS. The risk is that Redis official ACL already handles most blocking, so it must win on better default policies and audit experience.
Block FLUSHALL, KEYS and other dangerous Redis commands by policy, before they reach production.
B surface-field A themeable dot-and-line field component whose pointer-following lighting gives interfaces a lightweight animated background, ready to use with high visual recognition; 31 stars in 1.5 days is good momentum among this batch of frontend components. The business path is a component library or design system plugin; the risk is that decorative components are highly commoditized and easily copied by CSS animation tutorials, making a paid moat hard.
Themeable React field of dots and lines that makes room around surfaces, with light that follows the pointer
B chess-postmortem-skills A Claude Code skill combo for chess game review: voice notes, Stockfish annotations, an HTML viewer and commentary video in one pipeline, turning engine output into a review a human can retell; 51 stars plus 70 Show HN points is dual-source endorsement. Commercialization is constrained by roughly $15 in token cost per game and a niche audience, making it better as an agent skill combo pattern to reuse in teaching scenarios.
Claude Code skills for chess post-mortems: Stockfish-backed annotated PGN, HTML viewer, narrated video
B Show HN: A Claude Code skill to analyze your chess games Uses a Claude Code skill for chess review: voice notes in, Stockfish-annotated PGN and commentary video out, solving how tedious and hard to remember pure engine branch browsing is. 70 Show HN points plus 51 stars in two days is real usage feedback. The business path is narrow, roughly $15 token cost per game is hard to scale, and it is more a sample of the agent-skill pattern migrating to teaching and content generation; the risk is being niche and easily replaced by general AI analysis.
Hello HN, It started as an experiment: can Claude play chess properly if it uses vision instead of PGN notation? Somehow it can. The next experiment was to see whether Claude + Stockfish could explain…
B keephaven A NixOS image turning a Raspberry Pi 5 into a private home cloud with six self-hosted apps preinstalled, plus a recovery partition and signed updates, emphasizing reproducibility and rollback more than Umbrel. Suited to people who write Nix, with extensibility and hardware compatibility as the main limits.
Private home cloud OS for the Raspberry Pi 5: a NixOS image with six self-hosted apps, a recovery partition and signed updates
B finder-alias A JS library for parsing and creating macOS Finder aliases, filling a small long-standing gap for a clean API in the Node ecosystem. The author is a prolific macOS tool maintainer with sound boundaries and naming, though its use is narrow and non-macOS projects can't use it.
Resolve and create macOS Finder aliases
B company-brain supermemory open-sources its internal "company brain": a Slack teammate that remembers everything the team says and can pick up work, backed by real internal use and nearly 500 stars in a day. The limits are that self-hosting cost and permission design go unaddressed, and Slack AI assistants are already crowded.
Open-sourcing our company brain - A teammate in your Slack that remembers everything your team says, and can go do the work.
B Cohort A Material 3 panel for power, battery, fan, and keyboard settings on Lenovo Legion laptops under Linux, a real need since LenovoLegionLinux has long been command-line and kernel modules only. The scope is clear and the positioning sharp; the limits are 7 stars and under two days, with no word on how many models it covers.
Power, battery, fan and keyboard settings for Lenovo Legion laptops on Linux, in Material 3
B bundle-dupe-check Finds packages bundled multiple times at different versions in Vite/Rollup/esbuild and shows the import chain, targeting bundle bloat and phantom dependencies that are hard to trace by hand, and frontend teams can wire it into CI immediately. 16 stars in two days is usability evidence; the business path is merging into bundle-analysis or observability SaaS, but existing tools like webpack-bundle-analyzer and npq cover part of the ground, and a single-point lint is hard to charge for independently.
Find the same package bundled twice at different versions in Vite, Rollup or esbuild, with the chain that pulled in each copy.
B sqlite-wal-doctor Diagnoses WAL lock contention, unbounded WAL growth and checkpoint starvation, the most common and hardest-to-pin-down class of SQLite production failures; two days and 20 stars shows ops people feel it. The business path is an embedded database observability service (think pganalyze); the risk is that SQLite users are mostly lightweight scenarios with insufficient willingness to pay and diagnostic demand density.
Diagnose SQLite WAL misconfiguration — lock contention, unbounded WAL growth, checkpoint starvation.
B copycat-skill A Claude Code skill that captures multi-viewport screenshots of a target URL, extracts design tokens, then iterates pixel comparison to approach a clone. A runnable workflow for frontend replication, though copyright risk and heavy dependence on the target site limit it.
copycat - Claude Code skill: URL → pixel-perfect website clone. Playwright capture (3 viewports, folds, sections, hover), design-token extraction, pixel-diff verification loop. README EN/FR.
B YAUI A Unity UI approach implementing Flexbox layout on GameObject, from an author with a track record in Unity tooling and close in spirit to Web's Yoga. Worth a look for anyone suffering with UGUI layout, though at 22 stars the ecosystem is young and it faces direct competition from UI Toolkit.
Yet Another Unity UI: a fast, Flexbox-based UI system on GameObjects.
B gcdatlas An explorable map drawn with ASCII characters and WebGL, going from Earth all the way to the edge of the observable universe, with restrained positioning and no hype. It's not a production tool, but as a science showcase and geometry rendering exercise it's worth ten minutes; depth can't match Space Engine.
The real universe, drawn entirely in ASCII. An explorable WebGL atlas from Earth to the edge of the observable universe.
B recurrent-transformer The author's usual style: as soon as a paper appears, a readable minimal implementation, this time Harvard's recurrent Transformer paper. It saves reproduction time for those following architecture experiments; but it's exploratory code, with training scale and conclusions unverified, so don't treat it as a benchmark.
Explorations into the Recurrent Transformer proposed by Costin-Andrei Oncescu et al. of Harvard University
B typed-lm Turns Llama, Qwen and similar models into typed decision APIs with a single forward pass, implemented in Rust with Candle, plus LoRA/QLoRA training and quantization. Worth a look for structured output from small models, though at 9 stars, whether accuracy can replace constrained decoding is unverified.
Single-forward-pass semantic routing in Rust: turn dense LLMs (Llama, Qwen, Mistral, Gemma) into a typed decision API with LoRA/QLoRA training and FP8/FP4 quantization, built on Candle.
B bytex An event-driven trading engine for .NET, emphasizing fixed-point amounts and one strategy shared across backtest, paper and live, in the spirit of NautilusTrader. It claims eight exchange integrations on day one, which reads more like a vision; live trading depends on code maturity.
Event-driven algorithmic trading engine for .NET. Write a strategy once and run it unchanged in backtest, paper and live trading, deterministic, with fixed-point money and adapters for Binance, Bybit,…
B reelql A Claude skill plus API that turns video links into structured JSON for agents. Input and output are clearly stated, but underneath it's likely a wrapper around existing video models, and promises like "any video link" are overblown; video understanding APIs are plentiful, so first check how it handles long videos and failure cases.
Give your agent eyes: any video link in, one typed JSON out. A Claude skill + API.
B promo-video-skill A skill pack for coding agents that uses Remotion to generate vertical promo videos, claiming even cheap models can produce clips. The scenarios are concrete (catering, e-commerce, tourism), but results depend on template quality, and skill repos are highly homogeneous; watch a demo first.
Lets cheap models like DeepSeek make good-looking vertical promo videos (software/food service/e-commerce/education/beauty/tourism, Chinese-English bilingual | A skill that lets low-cost LLMs make pol…
B DORA A heterogeneous geospatial reasoning benchmark for disaster response, testing LLM agents on real emergency tasks, a more interesting question than yet another code leaderboard. But it accompanies a paper and targets researchers, and the "Oral" label is hard to verify.
[NeurIPS 2026 Oral] DORA: Can LLM Agents Respond to Disasters? Benchmarking Heterogeneous Geospatial Reasoning in Emergency Operations
B SeeCut Automatically rough-cuts talking-head or digital-human footage into short videos and exports layered CapCut drafts, a narrow niche with real appeal for talking-head creators. But similar auto-editing tools are plentiful, and with 13 stars in a day and a half the output quality is unproven; it currently looks more like an idea.
An AI editor that watches its own cut: talking-head / AI-avatar video → auto-edited short video. 网感口播精剪：数字人/真人口播 → AI 自动精剪成网感动效短视频 → 可选剪映分层草稿
B paint-mv-skills A skill pipeline that turns songs and lyrics into watercolor-style MVs, explicitly naming p5.brush, headless Chrome and ffmpeg, with honest steps and no hype. The problem is similar agent video skills are flooding in, and quality differences only show in the output.
Agent skills that turn a song and its lyrics into a hand-painted watercolour music video (p5.brush + headless Chrome + ffmpeg), following PDoomVideo's pipeline
B ballast A SwiftUI macOS tool for visualizing and reclaiming disk space, tackling the real problem of deleting without breaking the system. It faces DaisyDisk and OmniDiskSweeper, so it must prove its safety boundaries, and at 11 stars it's unclear how conservative its cleanup logic is.
See where your Mac's disk space went, and get it back without breaking anything. Native SwiftUI, macOS 26.
B pcybox-attackgraph Builds a simulated infrastructure in the browser to watch how attacks move laterally and verify whether your defenses hold, sitting between a range and a detection lab. The direction is interesting, but with 10 stars in a day and no docs, maturity is unknown.
Open source browser lab to build a simulated infrastructure, watch an attack move through it, and check whether your defense works.
B motion-video-skill Turns a brief or screenshot into a frame-by-frame rendered HTML/CSS animation MP4 with sound effects, and the Chromium plus ffmpeg pipeline is clearly explained. But agent animation skills like this are piling up these days, and without seeing output it's hard to say it beats hand-written templates.
Agent skill that turns a short brief, screenshots or a repo into a directed motion-graphics MP4: it directs a concept, builds HTML/CSS scenes, and renders them frame-exact with Chromium and ffmpeg, wi…
B clodfarm Aggregates multiple Claude Code accounts into a task farm, pacing sub-agents by each account's 5-hour and weekly quotas. The pain point is real, but it's essentially concurrency around subscription limits, with stability and compliance both questionable; similar agent orchestration projects are plentiful, and it offers no verifiable difference.
clodfarm (say it out loud): a farm of Claude Code agents. Plant a mission, they split it into sub-agents, open the work, and pace themselves on each account's real 5-hour and weekly usage. Steer it fr…
B ComfyUI-Fizgig-H3-Still Extracts true single-frame stills from MiniMax H3 in ComfyUI using one frame's latent and clean decoding, without extra model downloads. Very small in scope and honestly described, saving a step for H3 still-frame users, though it only serves this one workflow.
True single-frame stills from MiniMax H3 in ComfyUI: a one-frame latent and a clean decode, at any size. No extra models.
B ai-sni-proxy Uses an overseas host plus an SNI proxy to let restricted regions access Claude, ChatGPT, and JetBrains AI without a full VPN. The approach is honest and the steps concrete, but SNI proxying is old technology and circumventing regional restrictions carries stability risks.
Reach Claude, ChatGPT, Gemini and JetBrains AI from a country they don't serve — without a VPN: one hosts edit and an SNI proxy abroad
B aws-drawio-diagram-skill A skill that lets Claude Code draw AWS architecture diagrams in draw.io format, a real need with a small enough scope. But there are no sample files proving usable output, and similar diagram skills are multiplying fast, making its difference hard to see.
A Claude Code Skill that draws AWS architecture diagrams in draw.io format
B motion-video-skill An agent skill that one-click produces 1080p videos with AI voiceover, subtitles, and generated music, directly matching content creators' needs; 47 stars in 0.7 days and an MIT license aid diffusion. It could monetize through skill subscriptions or API billing, but the risk is replacement by one-click video products like Runway and Pika and many free similar skills.
Agent skill that produces beat-synced 1080p motion-graphic videos in HyperFrames (HTML + GSAP) with an AI voice-over, karaoke captions, SFX and generated music. MIT.
B json-guard Limits parse depth, key count and length to reject JSON bombs before allocation, directly closing the DoS attack surface; 51 stars in 1.6 days shows security and backend teams can use it immediately. It could become the default secure parsing library in the Node ecosystem and monetize via compliance audits, with the biggest risk being Bun/Node native protections or zod competitors absorbing the capability.
JSON.parse with hard limits on nesting depth, key count, string length and input size. Refuses parser bombs before they are allocated.
B omnibin Puts every binary in nixpkgs history directly on PATH, solving the core friction of wanting to use Nix immediately; 76 stars in 0.9 days is strong momentum. It has potential to become a standard tool for Nix users with monetization via hosting or support, but the risk is that official nix shell/flake experience improvements gradually erode its necessity.
Every binary nixpkgs ever shipped, on your PATH.
B unreal-home-wizard Uses Codex to turn room and home photos into a walkable, reviewable UE project, automating the path from photography to 3D scenes; 49 stars in 1.7 days is decent momentum. It could charge per use for real estate showcases and game prototyping studios, but the risk is quality depending on LLMs and UE version compatibility, with image-to-3D services like Meshy as strong substitutes.
Turn room and home photographs into reviewable, walkable Unreal Engine projects with Codex.
B pg-outbox A driver-free, dependency-free Postgres transactional outbox that plugs in with any query function, cutting distributed consistency boilerplate; 20 stars in 1.6 days is normal growth. There's a path to becoming a Node standard dependency with monetization via a hosted service, but the risk is that Debezium/CDC and logical replication replace the outbox pattern.
Transactional outbox for Postgres with no driver dependency — pass in any query function.
B agentic-cuda-optimizer A langgraph workflow drives an agent to run the CUDA harness, read benchmarks and nsight profiles, and automatically modify kernels, closing the performance tuning loop; 27 GitHub stars plus 36 HN points gives dual-source momentum. It could be a performance optimization SaaS or embedded in CI, but the risk is poor reproducibility of tuning results and competition from NVIDIA's auto-tuning tools.
A langgraph based workflow with a C++ CUDA harness to optimize CUDA kernels
B umbod Lets teams connect internal APIs and MCP on their own infrastructure and manage agent permissions by group, hitting enterprise agent governance and data egress pain points; 20 stars in 1.6 days but lacking customer evidence. The enterprise subscription path is clear with high deal size, and the risk is that model gateways and identity vendors are also extending into the permission layer.
Agent access on your own infrastructure. Connect internal APIs and MCP servers, manage permissions by group, and let teams choose their agents.
B Show HN: Agentic CUDA Kernel Optimizer Uses langgraph to orchestrate agents running C++ harnesses and reading nsight results to auto-edit CUDA kernels, automating the manual performance-tuning loop; HN 36 points and 27 GitHub stars corroborate. It could offer optimization services or join MLOps platforms, but results are hard to reproduce and NVIDIA's own toolchain covers this.
Hello; I was working on optimizing some CUDA kernels and I thought may be it is a good oppurtunity learn langgraph as well. I created a simple C++ CUDA Test Harness and handed that to AI agents. They …
B Show HN: Jev Plays Pokémon Red Lets low-latency model Jev play Pokémon Red and streams token costs, testing long-horizon agent planning; 259 points is the batch's strongest signal, with open source plus streaming driving recurring traffic. It could funnel to an agent runtime or model promotion, but entertainment projects rarely become products and it depends heavily on one model's capability.
Hey HN! Wanted to share a fun project I've been hacking on. Given Jev can make decisions really fast (but not fast enough to play Doom yet sadly), I wanted to try and push it to play a more complex ga…
B ai-system-design Step-by-step explanations of LLM, RAG, and Agent-based system design, filling a gap in interview and architecture learning material; 154 stars in 0.8 days is extremely strong momentum, proving real content demand. It can monetize through courses, consulting, and a newsletter, but the risk is that content repos have a low ceiling and are easily replaced by top education institutions' content.
AI System Design - Learn how to design AI systems built on LLMs, RAG, and AI Agents step by step.
B collabosm Gives a reproducible recipe and measured throughput for running a 125B MoE on a single Colab A100, directly solving the cost of personal large-model experiments; 70 stars in 1 day is strong momentum. It could become a reference implementation for inference deployment and funnel to hosted services, with the risk that weights and runtime versions expire fast and vLLM official docs easily supersede it.
Run Qwen3.8-Flash-Next (125B-A6B MoE) on ONE Colab A100-80GB High-RAM: pinned runtime, weights from HF, OpenAI-compatible endpoint, measured 2.8k-3.9k t/s prefill / 90-97 t/s decode, plus a verified p…
B motion-graphics Packages motion graphics as a skill Claude Code and Codex can call directly, riding the agent-skill ecosystem boom; 29 stars in 0.5 days is decent momentum. It can monetize through skill marketplaces or paid skills, but the risk is that the official skill ecosystem and many free similar lists quickly dilute its value, and it lacks exclusive capabilities.
Motion-graphics skills for Claude Code and Codex.
B flux Connects an Omarchy host's files, clipboard, notifications, media, and camera to the phone, solving the desktop-phone split; 46 stars in 0.2 days is a very good initial pace. It could monetize via open-source plus hosted dual tracks, but the risk is that KDE Connect, LocalSend, and Snapdrop already cover most functions, leaving differentiation and willingness to pay unproven.
Connect an Omarchy computer to your phone: files, clipboard, notifications, media, camera, and more
B ft-motion Treats motion graphics as a pure function of time, rendering headless Chrome and compositing with ffmpeg, plus a brief-to-video prompt for coding agents, filling a gap in agent-produced video; 20 stars in 0.4 days. It could charge via skills or API, but risks competition from After Effects scripts and similar agent video skills.
Motion graphics as a pure function of time — Canvas 2D scenes, headless-Chrome rendering with motion blur, ffmpeg, synthesized sound, and a brief-to-video prompt for coding agents.
B ComfyUI-Fisher-Pose A pose and camera editor in ComfyUI for Qwen Image portrait generation, with a built-in skeleton library and free multi-camera views, addressing exactly the pain of uncontrollable composition in AI images. Directly usable for illustration workflows, though tightly bound to a specific model and plugin ecosystem.
ComfyUI pose and camera editor for Qwen Image 2.1: VNCCS realistic human figures + one-click OpenPose posing (built-in 209 skeletons) + free-view multi-person camera | Pose & camera studio with VNCCS …
B Show HN: Paper-docx – agent-native Python-docx fork with 78% fewer DOCX failures Forks python-docx and cuts DOCX generation failures by 78%, hitting the frequent pain of LLM-written documents breaking formatting; 6 Show HN points is low but the repo has about 100 stars. It could become a standard dependency for agent document generation with a hosted service, but upstream may absorb the patch or other docx libraries replace it.
B Show HN: Ekselio – Loveable for finance workflows (local first) A local-first financial workflow generator: LLM orchestration, browser execution, exportable audit packages and QuickBooks integration, hitting the black-box problem in finance automation; 25 points is early validation. The subscription model is clear, but it is a solo developer competing with Power BI, Monarch and other mature finance tools for enterprise budgets.
Hi everyone, I am KD - Back in my college days, I dabbled with coding, learned the basics, HTML, CSS etc. but somehow I ended up in Finance which consumed the next 20 years. Then, during covid I picke…
B VeriTile A Lean 4 framework for formally reasoning about Triton-style kernels, targeting the nearly untouched corner of GPU operator correctness verification, meaningful to researchers in compilers or formal methods. The limits are a very high barrier and unclear use cases; at this stage it's more paper companion than usable tool.
Lean 4 framework for reasoning about Triton-style kernels.
B keelflow A community continuation of Flowise: it keeps only the Apache-2.0 code and focuses on security fixes, with existing flows and data migrating directly. It does save effort for teams with compliance needs who want visual agent orchestration, but it's still a low-code LLM platform with limited innovation, and maintenance effort needs checking.
Build AI agents and LLM workflows visually. Security-maintained continuation of Flowise: security fixes, Apache-2.0 only, drop-in for your existing flows and data.
B SFSymbolReplacer A small tool that globally replaces macOS system SF Symbols with custom PNGs, doing one thing with a direct name. Suited to theme customization and UI experiments; the limits are the risk of touching system resources and a very narrow audience.
Replace system SF Symbols with your own PNGs, system-wide.
B DLSSNR-AMD An attempt to run DLSS neural rendering models on AMD RDNA4 via Vulkan, a hardcore experiment moving Nvidia-exclusive tech to AMD cards, appealing to graphics reverse-engineering fans. The limits are unclear validation and being far from usable, with only a one-line description.
DLSS 5 neural rendering (NR) model on AMD RX 9000 (RDNA4) through Vulkan
B Seep-Reverse-Lab Turns multi-platform reverse engineering and CWE-602 client authorization bypass auditing into an agent-native workbench, letting models follow decompilation, protocol and auth logic, serving high-frequency mobile security and pentest grunt work; 88 stars shows security researchers willing to try. But IDA, Ghidra and Burp dominate, and big-vendor toolchains may absorb it.
Agent-Native multi-platform reverse engineering and CWE-602 client-side authorization audit workbench.
B rooms Makes each project a room, switching the whole window layout for that project in one click, hitting the pain of repeatedly rearranging windows when running multiple macOS projects; 273 stars in 1 day shows real demand. But window management is a red ocean: Moom, Rectangle, free Raycast plugins and built-in Stage Manager, so independent buyout is hard to scale.
Switch between projects on your Mac. Every project is a room: its windows, laid out, one shortcut away.
B Show HN: Radix – Visual UI for agentic programming It turns one-off agent-generated artifacts into persistent, commentable, editable interactive widgets on local disk, trying to make the chat window no longer the only interface to an agent. The direction is right, but Claude artifacts, Canvas and notebooks already cover most scenarios, 15 points is low, and Pro revenue depends on replacing in-IDE experimentation.
Hey HN, I'm Jordan from Radix. Radix is a UI tool for programming agents. You prompt your agent to generate a workspace for a task you're working on and get an interactive widget that persists locally…
B SuperMario3DWorld-VR A native OpenXR VR mod for Super Mario 3D World on Cemu, supporting stereo rendering, 6DOF, and first person. It's very specific for Wii U or VR players and technically substantial; the limitation is dependence on Cemu and the game itself, narrowing the audience.
Native OpenXR VR mod for Super Mario 3D World (Wii U) on Cemu, with stereoscopic rendering, 6DOF head tracking, diorama and first-person modes.
B agents-dont-sleep A tray utility that prevents lid-close sleep only while agents like Claude Code or Codex are working, restoring afterward; the niche is tiny and the need real, spanning macOS/Windows/Linux, but it depends on recognizing each agent's process.
Close the lid, let your agents cook. A tiny tray app that keeps your laptop awake (lid shut, in your bag) only while Claude Code, Codex & friends are working, then lets it sleep. macOS · Windows · Lin…
B voxel-musou A build-free browser voxel Musou game in Three.js, with Zhao Yun versus 300 soldiers. Both completeness and shareability are good, worth seeing for game devs and graphics fans, but it's a demo with limited gameplay depth.
Dynasty Warriors-style voxel action game in the browser: Zhao Yun vs 300 soldiers. Three.js, no build step.
B quiver-ae A billion-scale vector retrieval system for GPU and SSD, aimed at ANN infrastructure builders: spilling indexes that don't fit in VRAM to disk while keeping latency down, described without hype. The limitation is it's a pure research prototype with no install instructions or reproduction data, similar to FusionANNS and PipeANN.
A low-latency, high-throughput, and billion-scale GPU-SSD vector search system.
B amfi-allow Lets ad-hoc signed binaries on macOS carry Apple private entitlements by cdhash without patching amfid, aimed at macOS developers doing signing or security research. The problem angle is very narrow and accurately described; the limitation is the high platform and permission barrier.
Let an ad-hoc signed binary carry Apple-private entitlements on macOS, by cdhash, without patching amfid
B pixel-office A Windows desktop app that puts multiple agents in a pixel office to collaborate, directly producing pptx/docx/xlsx/pdf, using your own key with approval before execution; the feature list is concrete, but the office-visualization wrapper has appeared many times this year, and reliability depends on how it handles failures and concurrency.
A Windows desktop app for visual AI agent teamwork in a pixel-art office. Create PowerPoint, Word, Excel and PDF files, switch AI models with your own API keys, and approve plans before agents act.
B ipc-edge A PLC data-collection and edge gateway for industrial sites, unifying multi-brand protocols with MQTT offline buffering and reporting, described specifically enough to judge who it suits; the limit is that such tools depend heavily on field protocol coverage, and the README lacks a supported PLC model list and deployment package.
Cross-platform PLC communication and edge gateway tool — unified collection across multi-brand protocols, MQTT offline caching and reporting, collection and cloud upload in one program.
B cookiebannerbench It quantifies the cost of consent banners across the same set of pages: appearance timing, size, viewport occlusion ratio, and layout shift, with metric definitions more concrete than saying experience is bad. The limitation is that the publisher is itself a CMP vendor, so sample selection and data need independent verification.
An open benchmark for what consent banners cost the pages they sit on — time to banner, payload, viewport blocked and page stability, measured on identical pages.
B digiemu An emulator running Elektron Digitakt mk1 firmware behind a clickable panel, with real-time audio and bundled .syx files; very appealing to niche hardware users and reverse-engineering/embedded enthusiasts, but limited to one model and unofficial.
Elektron Digitakt (mk1) emulator: runs the Digitakt's own firmware behind a clickable front panel, with live audio. Portable Windows app, bring your own .syx. Unofficial.
B ghost-editor An AI editing agent skill for talking-head short videos, featuring 7 styles, safe subtitles, and reverse-engineering from reference clips. The angle is specific and shareable, but it depends on HyperFrames and Claude Code, and actual output quality remains unverified.
AI video editor for talking-head reels: 7 styles, face-safe captions, motion scenes, reverse-engineer any reference edit. A Claude Code / agent skill on HyperFrames.
B transgran-monstr Hashes phone numbers from CRM exports before sending them to large models while preserving relationships. Useful for teams needing AI customer-data analysis within compliance bounds, but it only covers Russian CRM scenarios and has few samples.
Anonymizes CRM exports before sending them to neural networks. Phone numbers become hashes, and joins survive.
B tev1 A decision model Together fine-tuned on Qwen3.5 4B with open weights. It's a small-model reference for people doing agent orchestration, and the institutional backing lends the stars some credibility, but there are no evals or usage instructions, so the payoff is unclear.
Open-weight, Jev-inspired decision model finetuned on top of Qwen3.5 4B
B jevlint Writes semantic lint rules for code review issues that deterministic linters can't express, aimed at teams wanting to automate review. The angle is clear and it's not another formatter; the limitation is reliance on model judgment, with false positives and configuration cost still unseen.
Semantic lint rules for the code-review questions a deterministic linter can't express
B roblox-game-boilerplate An agent-friendly scaffold for Roblox projects: Rojo, Wally, typed networking packages, data stores, and AGENTS.md, more thought-out than most templates; but scaffolds themselves offer limited novelty, and value depends on whether these conventions keep up with Roblox and Rojo version changes.
Roblox game boilerplate for AI agents: Rojo, Wally, typed Packet networking, ProfileStore data, strict Luau and an AGENTS.md
B easyplot An agent skill for research data analysis that produces publication-grade charts in R or Python; the pain point is clear and it's more focused than generic plotting prompts, but it depends on an agent environment and an R/Python stack, and chart reproducibility isn't shown.
AI Agent Skill for scientific data analysis and publication-ready figures in R and Python.
B glance-speedlab A local-camera inference latency test bench backed by measurement data, paired with the author's Glance. The methodology is honest and useful for on-device vision practitioners to compare against; but it's tightly bound to one product, so its value is limited without Glance.
A measurement-first lab for low-latency local camera inference with Glance.
B wow-ai It connects Claude, Codex, and Grok inside a World of Warcraft addon; since the addon sandbox cannot reach the network directly, there must be a bridge layer, which is worth a look. But the README is one sentence, with no communication method or ban risk, and 80+ stars look more like player curiosity than utility.
Addon to use AI (Claude, Codex and Grok) inside World of Warcraft (WoW)
B opencode-gpt-live Adds real-time voice calls to OpenCode, running GPT-Live via a ChatGPT subscription. Attractive for those wanting to dictate to a coding agent, but it relies on a subscription voice channel, so compliance and stability risks need checking first.
Talk to OpenCode out loud: real-time GPT-Live voice calls on your ChatGPT subscription
B ollaya Locally pulls open decision models like Laya, decider, NLI, and GLiClass and serves them via a TypeSafe API, positioned as 'Ollama for decision models.' Promising for local classification/NLI serving, but the ecosystem and model quality remain unverified.
Run open decision models locally: pull and serve Laya, decider, NLI and GLiClass behind a TypeSafe-compatible API. Ollama for decision models.
B muster-ww2 Procedurally generates WWII soldiers, tanks, artillery, and aircraft with three.js into a browsable 3D archive or sandbox, with a clear scenario for historical display and teaching and a restrained scope. The limitation is that it's a content project, with model accuracy and licensing unstated.
Muster: a hands-on 3D archive of WWII military models. Browse, configure and arrange soldiers, tanks, artillery and aircraft on a diorama display. All models procedurally built in three.js + TypeScrip…
B Gaussian-Render-Capture A Blender add-on that renders models into COLMAP captures ready for training, feeding Gaussian Splatting pipelines. Useful for people making synthetic data, with a small clear scope; but it depends on the whole 3DGS toolchain, so the audience is narrow.
Blender add-on: renders a model into a ready-to-train COLMAP capture for Gaussian Splatting (Postshot, LichtFeld Studio)
B code-video An agent skill that researches a topic then generates a hand-drawn plus 8-bit style promo MP4, aimed at content or marketing people needing quick short videos. The workflow is concrete, but output quality and the required agent environment are unstated, so it reads like a demo.
Agent skill: research a topic and render a hand-drawn + 8-bit promo video (MP4)
B JEV-Paper-Radar Every morning a model reads arXiv and picks papers by interest, aimed at researchers tracking the frontier; cost estimates and fork-and-use are described concretely. The limitation is reliance on the Jev service, unverified recommendation quality, and plenty of similar tools already.
Let Jev read every new arXiv paper each morning and surface the few you should read. Plain-English interests, calibrated probabilities, ~$0.06/day, fork and go.
B claude-animation-skill Breaks hand-drawn 2D animation into detail settings, ant skeletons, watercolor brushes, and frame-by-frame reproduction for an agent to generate with canvas and ffmpeg; it's more specific than most skill packs, but output depends heavily on the model, and the repo lacks a complete film and failure samples.
Hand-drawn 2D animation written as code: detail bible, a detailed ant rig, brush/watercolour pen, frame-matched recreation loop, synthesised sound. Node canvas + ffmpeg.
B MemBodied The official implementation of recurrent associative memory for vision-language-action models points in a direction useful for embodied AI; however, it is paper companion code without weights or training details, making external reproduction hard, so it suits researchers following along rather than direct use.
Recurrent associative memory for vision-language-action models. Official implementation of MemBodied.
B win-plus-leave Locks the screen when a USB drive or YubiKey is pulled, closing the gap where Win+L gets forgotten, and can be deployed via Intune. The scenario is concrete and the PowerShell script easy to audit, but similar USB/Bluetooth lock tools already exist and the star count is very low.
🪟 + Leave: the Win+L you never forget. Locks Windows when your YubiKey or USB key leaves the laptop. Intune-ready.
B sonic_for_a3 Agibot's SONIC adaptation for the full-size humanoid A3 targets researchers who own the robot or work on whole-body control. The official backing suggests the code may actually work, but a one-sentence description and missing startup instructions make it look like an internal engineering release.
SONIC adpation for new full size humanoid agibot A3！
B MKVDCU-Recomp An unofficial project recompiling Mortal Kombat vs. DC Universe to PC, aimed at game preservation or recompilation enthusiasts. The technical direction is specific, but with only one sentence it doesn't say compatibility or whether the original files are needed.
An unofficial port of Mortal Kombat vs. DC Universe to PC (More platforms coming)
B SnafflePy A Python rewrite of the Windows credential-hunting tool Snaffler, filling a real gap for Linux-side internal pentesters; but the README is just "Snaffler in Python," with no rule coverage, run instructions, or differences from the original, and it can't even be installed today.
Snaffler in Python
B blender-claymation-skill A Claude skill that generates Blender stop-motion clay animation from one prompt, chaining storyboard, materials, and rendering; the direction is concrete, but it only gives prompts and a demo, with no render times, failure rates, or Blender version requirements, so usability must be tested yourself.
Claude skill: stop-motion claymation films in Blender from a single prompt
B incremental It expresses object construction as a typed fold, where each step knows the exact shape and the final type proves completion, a rare convergent idea in type gymnastics; the audience is only heavy TS type users, similar small libraries abound, and the practical gain over a handwritten builder is unclear.
A typed fold over object contributions: build an object piece by piece, know its exact shape at every step, and prove when it is complete.
B blender-game-skills Claude Code agent skills for Blender game artists, turning concept art into riggable game assets and validating against reference images. The direction is specific and suits indie game teams, but it depends on the agent and a Blender environment.
Claude Code agent skills for game dev in Blender. Turn concept art into rigged, game-ready 3D assets, measured against your references.
B CVE-2026-84543 A technical disclosure and PoC for a macOS SMB kernel vulnerability, with a CVE number. Directly valuable to security researchers, but the number and impact need verification; ordinary developers should only read the disclosure, not run the PoC.
Technical disclosure and PoC for CVE-2026-84543, a macOS SMB kernel vulnerability
B goutoujunshi-jev-chat A floating-window Mac tool that reads WeChat screens, analyzes conversation relationships, and drafts replies. The need is clear, but it screenshots private chats at no small privacy cost, doesn't explain model or data handling, and similar 'AI advisors' are already plentiful.
Doghead Strategist Chat: Mac WeChat screen reading, relationship analysis and reply-draft floating window.
B lifeline Remotely controls your local Cursor/CodeBuddy from a phone: view sessions, approve actions, dispatch tasks. The scenario is real, but solutions like Happy Coder already cover it, and the repo doesn't explain how the communication channel is secured.
Remote-control the Cursor / CodeBuddy IDE on your machine — watch sessions, approve actions, and send tasks from your phone.
B container A TanStack project running a JS/TS sandbox in the browser, heading straight into StackBlitz and Sandpack territory; with one sentence and single-digit stars, it's unclear whether it targets doc embeds or IDEs, and no differences from WebContainers are shown, so just watch it for now.
A browser-powered sandbox for JavaScript and TypeScript development.
B openchatx-mcp Turns ChatGPT into a runtime that can control the local machine, call local tools, and forward to your own model via one MCP connection; the idea is on track, but MCP bridges are already plentiful, and it needs to show permission boundaries and tool registration, offering positioning only.
Turn ChatGPT into a local agent runtime. Operate your computer, use local tools, discover MCP servers, and delegate work to your own models through one MCP connection.
B shipvideo Enter a URL or a one-line request and it produces a product launch video, using an agent to write HTML and server-side rendering; this is a crowded AI-generated-materials space, and the README reads more like a landing page, with no rendering pipeline or controllability details, so output quality cannot be judged.
Paste a URL or a prompt, get a launch video. Opus 5.5 writes HTML, an OpenComputer serverless agent renders it.
B cve-2026-87902-wordpress-lfi-lab A reproduction environment for an unauthenticated LFI in WordPress, plus a scanner and PoC, with affected versions noted and an authorized-testing-only caveat, written properly. But such repos have very short lifespans, and value depends on whether the vulnerability is real and the PoC runs, so verify it yourself.
Reproduction lab + URL-list scanner + PoC for CVE-2026-87902 / GHSA-7hp8-65ch-5whp — WordPress get_page_template() unauthenticated LFI to conditional RCE (WP 4.7.0-7.1.1, fixed 7.1.2). Authorized/defe…
B zh-tech-writing An Agent Skill based on Ruan Yifeng's Chinese technical writing guidelines, emphasizing short sentences, plain language, and removing AI tone. Directly useful for Chinese technical writers, but it's essentially a prompt/spec wrapper, hardly a standalone tool.
An Agent Skill for writing Chinese technical documentation, based on Ruan Yifeng's "Writing Style Guide for Chinese Technical Documentation": short sentences, plain language, no AI tone.
B fathom A depth-style Alt+Tab for Omarchy that layers windows by recency and overviews workspaces. The interaction idea is fresh and scope restrained, but it only serves Omarchy users, limiting its generality.
A depth-based Alt+Tab for Omarchy: every window as deep as it is old, every workspace at a glance, any theme.
B NotProton It brings a Steam Play/Proton-style compatibility layer to macOS so Mac users can run their Windows game libraries, a long-unsolved and highly shareable pain point, starting fast with 52 stars in 0.9 days. It could go subscription or partner with publishers, but the biggest uncertainty is facing Apple GPTK, CrossOver, Whisky and Valve's stance, with high compatibility and legal risk.
Steam Play for macOS
B 3dicon Generates looping 3D icons with real alpha transparency from a single prompt, directly cutting the asset cost of product landing pages and app icons—transparency is exactly where most AI image tools fall short. 134 stars in 0.2 days shows developers readily accept "skill as tool." Icon-pack subscriptions or a design plugin could monetize; the risk is official AI features from Figma, Spline and others swallowing it.
One prompt in, a looping animated 3D icon out — with real transparency. A Claude Code skill.
B cogsend A self-hosted social media scheduling tool that runs on the user's own Cloudflare account, solving expensive subscriptions, account bans and data leakage—attractive to indie developers and agencies, with 112 stars in 1.9 days from self-hosted community spread. A hosted version and team collaboration could be paid; the risk is that open-source peers like Postiz and Mixpost are ahead, and platform API policies are tightening.
Self-hosted social media scheduler. Write once, publish to any connected platform. Runs on your own Cloudflare account.
B FoxDevStudio Provides a modern IDE and 64-bit runtime for Visual FoxPro 9, directly solving the problem that many legacy enterprise systems cannot move to 64-bit and lack modern tooling; such users have strong willingness to pay and few alternatives, and 74 stars in 1.2 days shows the old ecosystem still has heat. The risk is the group naturally shrinking, Microsoft's migration narrative, and the huge compatibility workload for language and runtime.
A modern IDE and 64-bit runtime for Visual FoxPro 9 applications.
B Show HN: Ive Sent It – online courier for files, with signed proof of delivery Turns large-file delivery into a "courier" with signed delivery receipts, solving the proof problem in legal and finance scenarios where the recipient denies receiving it—more compliance-oriented than WeTransfer. 18 points on HN shows the pain point is understood, and SaaS subscription plus enterprise compliance is a clear monetization path. Risks: email and cloud storage vendors adding native sign-off, and compliance qualification barriers.
B paragravity Provides non-invasive multi-account isolation and sandboxing for Google Antigravity, solving the problem of developers running multiple agents in parallel with state bleeding between them; it rises naturally with Antigravity's popularity and could extend into a general IDE/agent multi-instance manager charged per seat. Risks: tight binding to a single platform, so an official multi-workspace feature would kill it, and multi-account strategies may violate ToS.
Native, non-invasive parallel multi-account & sandbox manager for Google Antigravity
B ai-engineering-course A free structured curriculum spanning machine learning, Transformers, LLM fine-tuning, RAG, agents, evaluation and safety, filling the gap in systematic AI engineering teaching material; the author has reach in developer communities, and 75 stars in 0.7 days shows steady growth. Monetization comes from course and corporate training funnels; the risk is content decay requiring constant updates, with DeepLearning.AI and others already occupying the space.
AI Engineering Course - A free and complete AI Engineering Course to learn AI Engineering step by step - from Machine Learning, Neural Networks, and Transformers to LLMs, Fine-Tuning, RAG, AI Agents, …
B threejs-architecture-effects Packages the know-how for generating interactive self-assembling Three.js buildings into an agent skill, letting coding agents directly produce usable 3D scene code—hitting the gap where agents lack domain skill packs, and 175 stars in 1.9 days shows skill distribution has momentum. Monetization could be a skill marketplace or subscription; the risk is skills being easily copied or absorbed into model vendors' official skill libraries.
An Agent Skill for building interactive, self-assembling 3D architecture with Three.js. 一个用 Three.js 生成可交互、可动态建造 3D 建筑的 Agent Skill。
B dpapi-toolkit Drop in any Windows DPAPI artifact and it identifies the format and points out the required master key; give it the key and it decrypts offline, with a CLI and web interface. It saves work for forensics and incident response; compared with dpapick it is more oriented to single-file investigation, and its coverage still needs testing.
Drop any Windows DPAPI artifact and it identifies the format and the exact master key it needs, then decrypts once you supply the key. Offline, CLI + web UI.
B skipper A git MCP service for agents: local repos are read in-process via gitoxide, while remotes cover GitHub, GitLab, Gitea, and Forgejo by reusing each platform's CLI. It saves you from piling up gh calls yourself; the limitation is that it depends on each platform's CLI already being logged in.
MCP server for git repositories: local git in-process via gitoxide, plus GitHub, GitLab, Gitea and Forgejo through their CLIs
B YuE2-Studio Packages YuE2 as a native Windows app that generates full songs with vocals locally, and can also edit scores and replay precisely. It removes the Python environment hurdle, greatly lowering the barrier for people who want to play with music generation on their own machine; the limitation is GPU demand, and output quality is up to the model.
Local AI song generator with an editable score — YuE2 on your GPU: full songs with vocals, sheet music, covers, exact replay. Native Windows app, no Python, installer with auto-update.
B gut Adds an LLM judgment branch to ordinary Elixir control flow, treating model output as one condition in an if; the scope is restrained and the claims are not exaggerated. Useful for people writing BEAM services who want to embed a single classification in business code; how retries and cost are handled is not yet explained, and both scale and reliability remain to be verified.
Use LLM judgment in regular Elixir control flow.
B AstrLink A local AI gateway that unifies routing across users' subscriptions and APIs with an on-device privacy emphasis, solving agents filling in keys everywhere and picking models per task. The author maintains New API, bringing ready distribution and engineering credibility, and 54 stars in 1.8 days is steady growth. Monetization is enterprise gateway and quota management, with the risk of direct overlap and severe homogenization against LiteLLM and one-api family products.
A local AI gateway for AI agents, unifying your subscriptions and API providers with smart routing and on-device privacy protection.
B llm-multi-agent-game-evaluation It tests LLM strategic reasoning, deception and social norms with reproducible game environments such as RPS, Hawk-Dove and Liar's Bar. But such game benchmarks are already common, and no actual results are given.
A dynamic multi-agent game framework for evaluating LLM strategic reasoning, communication-driven decision-making, deception, and social-norm behavior across RPS, Hawk-Dove, and Liars Bar environments…
B bendc A self-hosting compiler for Bend 2 written in Bend 2, a rare and hardcore goal. It is a useful reference for language-implementation enthusiasts, but neither completeness nor practicality is there yet.
A self-hosting compiler for Bend 2, written in Bend 2.
B qu-ai-wei Merges three sets of Chinese writing rules for "removing AI flavor" and resolves seven places where they contradict each other, packaged as a Claude skill. People who need to polish Chinese drafts can use it directly; the limitation is that the rules come from empirical corpora, so results vary by genre.
A Claude skill for stripping AI tells from Chinese drafts.
B codex-routing-detector It probes which model actually answers behind a Codex request, to catch silent swaps. That's practical for users who suspect downgrades and hits an industry black box, but the decision basis and false-positive rate still need checking in the code.
Find out which model actually answers your Codex requests (detects silent model substitution, e.g. gpt-6-astra served by gpt-5.6-luna)
B hetzner-cloud-audit-skills It packages Hetzner Cloud attack paths, config drift and cost auditing into an agent skill, requiring evidence-backed conclusions, with a narrow and clearly defined scope. It suits small teams on Hetzner for periodic checkups; it depends on an agent host and covers a limited set of resource types.
Evidence-backed attack paths, drift, and cost auditing for Hetzner Cloud.
B windows-disk-cleaner-skill Brings WizTree-like fast NTFS/ReFS scanning to Windows disk cleanup and packages it as an agent-callable skill. People who want to script cleanup on Windows benefit directly, and 22 stars in a day shows demand; the limitation is that it only covers those two filesystems.
A Windows disk cleanup Skill based on Wiztree-like fast NTFS/ReFS scanning
B deepreseach-learing A LangGraph deep-research agent emphasizing resumable interruptions, evidence management, and cited reports, with engineering more complete than a demo; but such projects are everywhere, the author positions it as a learning reference, and the incremental value is limited.
Recoverable and auditable deep-research agent built with LangGraph. It plans research, searches the web, manages evidence and context, resumes interrupted runs, and generates cited reports—also design…
B jianying-automation Turns AI-generated storyboard assets into directly editable JianYing drafts, hitting a real short-video pain point: easy to generate, hard to get into editing software. The toolchain is concrete and the license is non-commercial; pyJianYingDraft already exists, so its edge depends on how long draft-format compatibility holds.
Jianying (CapCut) automated editing pipeline: AI storyboard assets -> editable Jianying drafts. cutcli+capcut-cli, assets keep their original names and full original audio, non-commercial attribution …
B qwen-image-2.1-studio A local Qwen-Image studio for Apple Silicon with a CLI and web UI, addressing the lack of CUDA on Macs. But Mac local image generation already has mature options like mflux and Draw Things, and a one-day-old repo hasn't proven it runs reliably.
Local Qwen-Image-2.1 studio for Apple Silicon: text-to-image, multi-image editing and prompt enhancing, with a CLI and a helmstudio web UI.
B drizzle A Drizzle integration module from the official NestJS org, giving teams already on Nest a maintained option for replacing TypeORM/Prisma. Little novelty; its value is entirely in the words 'officially maintained,' and the API design is worth a quick look.
Drizzle module for Nest framework (node.js) 🗄️
B NinjaMesher A generator that directly writes OpenFOAM polyMesh hexahedral grids via right-angle cuts, replacing the most tedious step in CFD preprocessing. The description is precise and the scope restrained, so OpenFOAM users can judge fit in ten minutes; the limitation is that it only covers this one mesh type.
Hex-dominant Cartesian cut-cell mesher for CFD, writing OpenFOAM polyMesh directly
B jev-browser Lets Jev decide control actions while Playwright executes, doing indexed browser automation to reduce the instability of agents operating web pages. The direction is right but the space is crowded (Playwright MCP, browser-use, Stagehand), 77 stars in 0.7 days is modest momentum, and it's tightly bound to the single CodexQA/Jev stack. Monetization is limited to enterprise automation services, and Playwright already holds the standard slot.
Jev Browser — indexed browser automation. Jev chooses the control, Playwright acts. A CodexQA skill.
B PrettyPitch Estimates F0 directly from singing voice without relying on scores, note onsets, or pitch annotations, a rare angle for people working on singing synthesis, transcription, and MIR research. The repo covers only this one thing and is quite restrained; but accuracy, datasets, and reproducibility are unverified, so for now it is more like a code appendix to a paper.
A singing F0 model trained without scores — no note onsets, no note pitches, no reference score
B termirai A remote management and AI ops platform for multi-device terminals, solving unified management of scattered server sessions and letting agents execute commands, but the track is crowded (Warp, Termius, tmux setups, various SSH MCPs). 52 stars in 0.5 days is middling momentum; commercialization can be team seats, with the risk of heavy security responsibility and high trust barriers, making it hard to become a standard in the gaps between giants.
Multi-device terminal remote management and AI operation platform.
B quiron A 'de-AI-ify' skill for Claude Code and Cursor that sells itself on quantifying rewrite results against a human baseline, somewhat more honest than slogan-style peers. But such prompt-layer tools depend heavily on context and many similar ones exist, so long-term differentiation is hard.
AI humanizer skill for Claude Code, Codex and Cursor. Removes AI slop and the signs of AI writing, then measures the rewrite against human baselines.
B DataFlow-AgentMM A framework from the DataFlow lineage for collecting and evaluating multimodal agent trajectories, a real direction. But the description stops at "framework," without saying how to install it or which visual environments it supports, and agent evaluation frameworks are already numerous, so the difference is unclear.
A framework for collecting and evaluating multimodal agent trajectories in visual environments.
B omacontra A five-level arcade boss rush on Linux and Hyprland, with dependencies and gameplay clearly written; it's a small but complete indie game. Whether it's worth it depends only on feel, and being tied to one window manager limits its audience, with many similar free action games around.
Omacontra: standalone five-stage arcade boss rush for Linux and Hyprland
B mu A coding agent that uses a small model as a judge to route work and a large model to do it, a sensible approach, and 49 stars in a day and a half shows people care about cost. But it is assembled on pi and AionUi, the coding agent space is already crowded, and there is no real comparison of results yet.
mu (μ): a coding agent that thinks before it acts. A small, fast judge makes the routine calls, the big model does the work. Built on pi and AionUi.
B roboto-physical-ai-actions Reviews robot-collected data and converts it into LeRobot v2.1/v3.0 datasets, with merging and version conversion. People building embodied-AI data pipelines can skip writing their own conversion scripts, and the scope is clear; the limitation is that it only covers the LeRobot ecosystem.
Roboto actions for physical AI: curate and audit robotics data, convert it to LeRobot (v2.1, v3.0) datasets, merge, up- and down-convert them
B llm-typesafe Simon Willison's Jev/TypeSafe integration plugin for the llm command line; the code quality is probably reliable and the style as restrained as ever. But it is only a thin adapter, and whether it is worth a look depends entirely on whether you have decided to use that model.
LLM plugin for accessing Jev and other TypeSafe AI models
B typist Converts photos into text art in the browser that can be pasted into Instagram or Telegram, with no install and a clear purpose, a clean little tool. But character-art generators already have many implementations, and it offers no better algorithm or new twist, so it is useful but replaceable.
Turn any photo into text art you can paste into Instagram, X and Telegram. Runs in your browser.
B yanwai An LSPosed module that shows emotion probabilities, subtext and reply suggestions in WeChat chats. The angle is specific and rare, but it relies on root/Xposed and a specific WeChat version, and the privacy cost is significant, so it's only for trying out.
A subtext assistant for WeChat chats: an Android LSPosed / Xposed module that shows emotion probabilities, implied meanings and communication suggestions based on context.
B dfirswarm The goal is multi-agent collaboration for digital forensics and incident response, and recording each step in a hash chain is a good idea. But talk of 'letting agents build their own tools' drowns out deployment and supported data sources, and a one-day-old project can hardly be called auditable.
A swarm of forensic agents, on your machine, on the record. Open-source agentic digital forensics and incident response under the AGPL: agents divide the case on a shared board, forge their own tools,…
B pasame A near-field file transfer tool with no install needed on the receiving end. The scenario is not new, since LocalSend and Snapdrop already cover it, so it needs clearly better discovery or transfer experience to be worth readers' time.
Pasá archivos a cualquier celular o computadora que esté cerca. Sin cable, sin cuentas, sin instalar nada del lado de quien recibe.
B JevBystander An Android accessibility screen reader that interprets WeChat messages and only pops up three hints: intent, emotion and urgency, without generating or sending anything, showing good restraint. But the scenario is very narrow, the basis and accuracy of the interpretation are unstated, and privacy is sensitive.
Android accessibility-based WeChat interpretation: screen reading only, only pops 3 Toasts (intent / emotion / urgency / suggestion), generates no reply text and sends nothing · zero third-party depen…
B obsidian-taskflow It adds an action workbench layer to Obsidian Tasks to 'reduce decision cost', which genuinely appeals to people with piled-up tasks. But the Tasks ecosystem already has many plugins and templates, it looks more like an interaction redesign than a new solution, and a day-old project has no usage feedback yet.
An action workbench for Obsidian Tasks that reduces decision cost and helps you start faster. 一个为 Obsidian Tasks 打造的行动工作台，减少决策成本，让你更快开始行动。
B droidmem An Android process memory read/write library using a kernel module with a /proc/pid/mem fallback, with a concrete technical path. But such tools are almost only used to cheat in games or scrape data, and the docs don't cover compatible kernel versions or the anti-cheat cost, so think carefully before installing.
Android memory reader/writer — kernel module + /proc/pid/mem fallback
B PickleChart An open-source order-flow and liquidity terminal, a rarity since such tools are usually paid software. But the Windows XP skin, 3D trade bubbles and '60FPS' promotion overshadow any explanation of data definitions and latency, making it look more like a demo.
🥒 High-performance, open-source Order Flow & Liquidity Trading Terminal styled in nostalgic Windows XP Luna. Features 60 FPS canvas charts, 3D trade spheres, resting limit walls, drawing tools, and lo…
B yard-sale A Wes Bos demo project: photograph secondhand goods to identify and price them, running on React plus Cloudflare Workers, good for learning how to write full-stack AI apps. But it is a teaching sample, not a product, the pricing itself is crude, and wrapper apps like this are a dime a dozen.
Yard Sale Gold — AI-powered yard sale scanner built with React and Cloudflare Workers
B rewindDV-LAB A MiniDV capture and repair tool for Apple Silicon, releasing only alpha binaries and test reports, with no source. Digitizing old DV tapes is indeed a gap, but a closed-source alpha isn't suitable for entrusting precious footage.
rewindDV LAB — native MiniDV preservation for Apple Silicon. Engineering alpha downloads, documentation and tester reports. No application source.
B go-jev The author is a trustworthy Go developer, but the content is just an SDK and CLI wrapped around TypeSafe Jev's typed decisions. Its value depends entirely on whether you've already chosen this model; the code itself offers no judgment worth reading, and 11 stars in half a day is just release noise.
Go SDK and CLI for TypeSafe Jev: typed decisions (yes/no, choice, score) from a model
B Kaghez A comic reader for GNOME written in Python, with an honest, understated description. But Komikku and similar readers already cover the need on Linux, so unless you are on GNOME and dislike the existing options, there is no reason to seek it out.
A manga reader for GNOME
B gpu-fan-control A script that controls fan speed by GPU temperature for a Corsair 1000D case with dual BOSS Storm3 controllers, written honestly and extremely narrow in scope. Only people with this exact hardware can use it, but the script is short enough to copy and tweak the curve directly.
GPU temperature-based fan control for Corsair 1000D + BOSS Storm3 (per-GPU curves, dual-controller)
B qwen-image-studio A local interface for running Qwen-Image on Apple Silicon, built on mflux plus ComfyUI's GGUF pipeline. It saves Mac users configuration work, but it is essentially a wrapper, and the description is a single Korean sentence.
An image generation studio for running Qwen-Image-2.1 locally on Apple Silicon (Next.js + shadcn/ui, ComfyUI GGUF, mflux)
B GitComment It adds a message board and visitor messages to a personal homepage using GitHub Discussions, with nine themes. It is a decorative gimmick that is easy to build but of limited value.
A comment section and Visitor Messages for your GitHub profile, powered by GitHub Discussions. Nine themes.
B llama-chat A native Android client for llama.cpp, with local inference that doesn't go through the cloud; the direction is right. But several mature clients already exist, the repo doesn't explain differentiating features like model management or background inference, and the star count doesn't show real usage.
Native Android client for llama.cpp
B jev-guard Adds a scoring guardrail layer to Claude Code to block unauthorized changes and unrequested deployments — a direction aimed at a real pain point. But the description piles on terms like 'probabilistic scoring' and 'evidence checking' without saying how to install it or what it blocks, and similar hook approaches are already plentiful.
Probability-scored guardrails for Claude Code: deny rule-breaking edits and unasked-for deploys, route your docs into each prompt, and check the final answer against the turn's own evidence.
B ad-radar A browser extension that collapses Xiaohongshu, Weibo, and Zhihu feeds by keyword and blogger, then uses your own model key to judge ads and topics. The idea is practical, but results depend on prompts and your own wallet, and similar blocking extensions are already plentiful.
Open-source browser extension: collapse content by keyword and blogger on Xiaohongshu, Weibo, X, and Zhihu; use your own Jev API key to identify topics such as ads, AI, military, and politics.
B guoapp A short-drama client built with Flutter and Go, using Actions to automatically build Android, Windows and iOS versions. The multi-platform packaging is tidy, but the content source depends on third-party platforms that may fail at any time, and it treads in copyright gray areas.
Hongguo Jian / Zhenguo Jian: a Flutter + Go multi-platform short-drama client, with GitHub Actions auto-building Android / Windows / iOS
B functional-emotions-video A hand-drawn-style music video made with multi-agent division of labor, including a self-written GPU brush renderer and storyboards — interesting as a workflow showcase. But it's an artwork, not a reusable tool; the scripts and render pipeline have no standalone docs, so others can't take it away.
A painted music video for "Functional Emotions", made by Claude Opus 5.5: custom GPU brushstroke renderer, storyboard, and 7 parallel chapter agents.
B MacOBlox Runs macOS Roblox on Linux via Darling — a rare approach, honestly stated in one sentence. But Roblox's client anti-cheat will almost certainly block it and Darling itself is incomplete, so it likely stops at the experimental stage: fine to watch, not to use as a tool.
Run the macOS Roblox client on Linux through Darling
B jev-chat-windows A Windows sidecar reply assistant for WeChat that screenshots and reads messages with local offline OCR before suggesting replies, addressing users' concern that chat content shouldn't leave the device; 252 stars on day one for the family shows the product form is validated. Monetization is likewise a subscription efficiency entry. Risks: overlap with the mobile version, WeChat version changes and compliance bans, and architecture dependent on OCR accuracy.
A reply assistant attached alongside WeChat (Windows 4.x): window screenshot + local offline OCR to read the other person's messages → Jev judges intent → 3 candidates filled in with one tap, sending …
B deepopen A non-autoregressive System 1 engine for structured typed decisions, replacing token-by-token decoding with one forward pass to cut the cost of high-frequency, low-value agent judgments; two-day 206 stars come from Jev话题 spillover. It could become a cheap decision dependency inside agents, but many similar djev/SemIf/laya projects and opaque evaluations make it hard to tell real from fake.
A non-autoregressive System 1 decision engine designed for structured type decision scenarios. DeepOpen Multilingual, non-autoregressive System 1 decision engine.
B ReCam An Android virtual camera that injects frames directly into cameraserver so any app sees a video file or desktop stream. Mobile security testing and camera development have real demand, and similar tools are mostly LSPosed modules; it needs system-level permissions, so ordinary users cannot use it.
Android virtual camera for security testing. Injects frames into cameraserver so any app sees a video file, desktop stream, or webcam instead of the real camera.
B WeChatBridge Forwards WeChat chat history to AI agents and Obsidian in one click, solving the collection problem of WeChat—the biggest data silo—for personal knowledge bases; 250 stars in two days shows real demand. Monetization is limited to one-time purchase or subscription, with a low ceiling. The biggest risks are WeChat ToS and privacy compliance, plus many similar export tools that easily break with client updates.
Native macOS tool for one-click forwarding of WeChat chat records to AI Agents and Obsidian
B imac-second-display Pushes low-latency Apple Silicon video to a 2010-era iMac as a second screen, with hardcoded H.264 over RTP, a virtual display, and zero-copy OpenGL, described down to the implementation level. The audience is narrow, but this is the kind of project that takes one odd idea all the way.
Low-latency screen streaming from Apple Silicon to a 2010 iMac on OS X 10.9. Hardware H.264 over RTP/UDP, virtual display, zero-copy OpenGL rendering.
B jev-test-filter Scores each test based on git diff, then outputs recognized filter arguments for vitest, Playwright, cargo test, and others, a plain idea that directly solves slow full test runs in CI. The author has a good reputation; the risk is reliance on the young Jev, with little data on effectiveness.
Score every test against a git diff with Jev, and emit the filter arguments vitest, node:test, Playwright, cargo test and go test already understand
B expo-adaptive-glass Provides an adaptive glassmorphism layer for Expo/React Native, adjusting blur, refraction, and highlights based on device capability and real-time frame rate. Matching the new iOS glass style is a real current need, and the restraint is right; the limits are covering only the RN ecosystem and lacking comparison with existing blur libraries.
Adaptive native glass surfaces for Expo and React Native. Scales blur, refraction and highlights to device capability and live frame performance.
B Show HN: JevBench, a reproducible benchmark for typed decision models A reproducible benchmark for decision models like Jev that return constrained options and probabilities, solving the lack of unified evaluation for non-autoregressive models; 53 points on HN shows topic heat, and public tasks plus scoring code boost credibility. If adopted it could become standard like lm-eval. Risks: vendor-led with the top of the leaderboard being its own model—a conflict of interest that's hard to sustain credibility against.
Hi HN! I built JevBench because Jev kicks ass, and the world deserves to know how the serious open source and fake lookalike projects really perform in comparison. Jev-class models return bounded choi…
B AltKeeper Moves AltServer into Docker on a home server to re-sign sideloaded iPhone apps over Wi-Fi, so you no longer need an always-on computer. The pain point is concrete and the approach clear, but it relies on Apple's signing mechanism and can break with any upstream change.
AltServer for iOS 17+ that runs on your home server (Docker): renews and installs sideloaded iPhone apps over Wi-Fi, no computer or VPN
B dolphinbench A benchmark plotting the Pareto frontier of agent memory capability, from the mem0 team, filling the gap where "memory" lacks a unified metric. Worth reading the eval design for agent infrastructure people; the limitation is that their own product is part of the evaluation, so watch for neutrality.
DolphinBench - Mapping the Pareto frontier of Agent Memory
B unreal-agent An async-first agent harness aiming to solve the engineering problems of concurrent orchestration for long tasks, but the harness space is already red ocean and the README's one-slogan description offers no differentiation; 895 stars in one day looks topic-driven rather than real adoption. Commercialization requires enterprise private deployment, but Claude Code and OpenHands already occupy the space. Risks: no moat, easily absorbed by frameworks.
Async-first agent harness
B nix-compact Compresses Nix build output to one screen while keeping full local logs and bounded diagnostics, and offers a flake interface, so anyone flooded by Nix logs will immediately get its use. The author has credibility in the Nix community and the scope is restrained; the audience is narrow, hence naturally few stars.
Native compact Nix build output with full local logs, bounded diagnostics, and a flake interface
B app-store-screenshot-skill An Agent Skill that turns raw app screenshots into App Store and Google Play promotional images. A handy little tool for indie developers with a concrete idea, but the 'pixel-perfect' claim lacks samples, and similar screenshot-template tools are plentiful.
Agent Skill: turn raw app screenshots into App Store and Google Play marketing images, UI pixel-exact
B tisco Ask questions about video transcripts in the terminal, verify evidence and cut clips, stringing transcript search and editing into one command-line pipeline for podcast and interview editors. The positioning is clear, but it lacks a demo and accuracy notes; the closest equivalents are GUI tools like Descript, and command-line versions are rare.
Question video transcripts, inspect the evidence, and organize clips from the terminal.
B ComfyUI-QwenImage21-LatentUpscale A latent-space 2x upscale node for Qwen Image 2.1 in ComfyUI, explained in one sentence with a very narrow scope, so image makers can install and try it. Similar upscale nodes are plentiful; whether it's worth it depends on releasing training details and comparison samples.
Learned 2x latent upscaling for Qwen Image 2.1 in ComfyUI.
B wet-recomp ReXGlue statically recompiles the Xbox 360 game WET into a PC version, explicitly without game files. Static 360 recompilation is rare and technically interesting, but it covers one title, needs your own game, and is narrow in practice.
Fan PC port of WET (Xbox 360) using ReXGlue. Does not include game files.
B Bluethread A dual-mode Bluetooth protocol stack written from scratch in standard C++, covering HCI/L2CAP/QoS with trimming support across Windows/Android/Linux/RTOS. Embedded Bluetooth developers will care, but BTstack and NimBLE already exist and this code's maturity is unverified.
Self-developed dual-mode Bluetooth stack in standard C++. Cross-platform for Windows, Android, Linux and RTOS. Modular design with thread-pool scheduler. Inter-module message delivery, serial executio…
B CVE-2026-43786 A PoC for a macOS CoreServices local privilege escalation that lets a normal app gain root. Useful for macOS security research and patch verification, but the repo has only the PoC and no analysis, so reference value depends on the details.
Proof of concept for CVE-2026-43786, a local privilege escalation vulnerability in macOS CoreServices that allows an application to gain root privileges.
B jeview A local visualization tool for Jev (TypeSafe) that shows every call the code makes in real time, explicitly declared unofficial. People already using Jev can skip digging through logs, but it depends entirely on upstream interfaces, has a narrow audience, and breaks when upstream changes.
An unofficial local visualizer for Jev (TypeSafe): a live view of every call your code makes. Not affiliated with TypeSafe AI.
B polytech-tree Turns the history of human technology into a three.js 3D tech tree, with data in plain JSON for easy PRs, which knowledge-visualization people will like. The highlight is the clean separation of interaction and data; the limit is that data accuracy can't be verified, making it more for viewing than research.
Interactive 3D map of the human tech tree, from the first stone tools to 2024: technologies stacked as a tower of eras in three.js, with prerequisite and relation edges, hover cards and a guided camer…
B FactorForge LLM-assisted quantitative factor research with DSL safety checks and cost-aware Top-K backtesting, clearly scoped and a ready scaffold for factor miners. But LLM-plus-quant projects are plentiful, so value is limited until validated.
LLM-assisted quantitative factor research with safe DSL validation, benchmark comparisons, and cost-aware Top-K backtesting
B CategoryTheoryStudio Draws commutative diagrams, 2-cells, adjunctions and 3D cubes in the browser, exporting SVG/PNG/tikz-cd. A handy specialized tool for category theory teaching, but online diagram editors already exist and it's a single researcher's project with unknown maturity.
Visual Category Theory Studio — draw commutative diagrams, 2-cells, adjunctions and 3D cubes in the browser; export SVG, PNG and tikz-cd. Concept by Dmitry Vostokov, DumpAnalysis.org
B jive A terminal coding agent that plans tasks as executable graphs, a clearer approach than linear chat, but this space is already crowded and its differentiation isn't enough to hold attention. Star growth is flat, with no demo or comparison, so not recommended for now.
A terminal coding agent that plans work as executable graphs
B laya-drift A small plugin for opencode users to observe agent behavior drift after long runs. Monitoring agent stability is a real problem, but the description gives no metric definitions or alerting method, and the audience is limited to opencode users, so watch for now.
opencode plugin to monitor agent drift over time
B MiMo-V2.6-Flash-DGX-Spark-Recipe A configuration record for running MiMo-V2.6-Flash with vLLM on dual DGX Sparks, covering tensor parallelism, KV cache and four image fixes, with measured baselines. It can save a day for the few who own this hardware, but it is dense and extremely narrow.
MiMo-V2.6-Flash-RL on 2x NVIDIA DGX Spark (GB10): vLLM TP2 + DFlash + vision/audio, fp8 KV 300K, four fixes for the stock image, measured baselines
B html-explainer Writes visuals in HTML and renders locally into explainer videos with voiceover, subtitles and covers, emphasizing audio-visual sync and being 'for agents.' The angle is concrete, but Remotion-style solutions are mature and the project has only a dozen stars with unproven stability.
Turn a topic into an explainer/analysis video with voiceover, subtitles, and cover — fully local, visuals written in HTML, stable audio-video sync. A deterministic HTML → MP4 rendering pipeline for AI…
B aayans-yc-indexor Describe a YC company in one natural-language sentence and logos float out of a physical pile ranked by match, covering 6,241 companies. Fun as a retrieval demo with a concrete description, but more toy than tool—you're better off browsing the YC directory.
Describe a YC startup in any words and the matching logos float up out of a physics pile, each with how likely it is. 6,241 companies, searchable by what they do, what their logo looks like, and what …
B 332_lab-jev-chat A Windows WeChat assistant that judges message intent and suggests DeepSeek replies, written in Kotlin. The need is real and growth is decent, but WeChat helper tools are heavily commoditized and rely on unofficial APIs that easily break.
Jev Chat Assistant for Windows - PC WeChat intent judgment and DeepSeek suggested replies
B abts A third-party Bilibili audiobook player touting no login and no ads, clearly appealing to Chinese audiobook listeners. But such third-party clients are already numerous, features and compliance boundaries are unexplained, and code quality cannot be judged, so it is not yet recommendable.
A third-party Bilibili audiobook player with no login, no ads, and a focus on listening
B jev-foundation-models A Swift 6 bridge layer connecting TypeSafe AI's "Jev" decision model to Apple Foundation Models. The engineering effort is small, and its value depends entirely on whether that external model actually works well; with no model source or evidence of results, it looks more like an ecosystem placeholder.
A lightweight, native Swift 6 bridge integrating TypeSafe AI's Jev System One decision model into Apple's Foundation Models framework.
B investigation-video-skill Packages the full pipeline of faceless long-form business investigation videos—word-level timestamped TTS, footage coverage, Remotion compositing, mixing, thumbnails and copy for eight platforms—into a Claude Code skill that produces a video from one sentence. It hits the capacity pain of batch account teams and could sell template subscriptions or SaaS; risks are platforms flagging low-quality content and footage copyright.
A complete production pipeline for faceless business/consumer research long-form videos: one sentence to a 10-minute finished video, open-sourced as a Claude Code skill (TTS word-level timestamps, foo…
B Show HN: Lossless-memory – a personal AI memory that never summarizes Full, timestamped long-term memory for personal AI without summarization solves the detail loss of summary-based memory. Its 55 points show real demand, but mem0, Zep and built-in IDE memories are all competing; technical moat is limited, so the path is either platform absorption or a privacy-focused local-storage subscription.
B dasheng Uses streaming ASR plus word-by-word judgment to give real-time pronunciation scoring for English reading aloud, solving the lack of instant feedback in speaking practice. 111 stars rides cheaper ASR and strong English-learning demand, and it could become an education SaaS or a scoring engine for exam prep; the biggest risk is established paid solutions like iFlytek already occupying the space, plus heavy reliance on third-party models.
Read Aloud — R2T2 streaming ASR listening, Jev word-by-word judging, English reading scoring
B editable-ppt-rebuilder Reconstructs images or scanned slides into editable PPT via OCR and image generation, emphasizing an auditable pipeline and strict quality checks. Useful for migrating old courseware, though the risk is that auto-generated layouts still need manual review.
Auditable workflow for rebuilding raster slides into editable PowerPoint decks with OCR, semantic assets, ImageGen, and strict QA
B crush-monitor Uses LLMs to analyze WeChat chat logs for sentiment, intent, and reply performance, helping people gauge relationship progress, a dating/social assistant tool. 114 stars show real consumer demand and easy spread, but it relies on WeChat data scraping with high compliance and account-ban risk, and similar products are numerous and hard to charge for beyond subscriptions; watch for now.
Crush affection monitor: uses Jev to analyze the emotion, intent, and reply performance of WeChat chats. Locally deployed, using your own API Key.
B self-compact-pi-agent Lets long-running coding agents monitor their own context usage and compress proactively, a real pain point in agent engineering. As a Pi extension it's very small in scope, and readers should note it's tied to a specific harness.
A Pi coding-agent extension that lets a long-running agent watch its own context and compact itself
B rizzo-flow Converges LLM output into typed decisions rather than free text, saving tokens and easing programmatic consumption, in the structured-output space. 223 stars in one day is a steep rise, but it is a local clone of the closed-source Jev, and multiple same-theme repos appeared around the same time, so the heat is questionable; no differentiation moat and limited commercial value.
The open, local take on Jev: typed decisions from an LLM, without generating a single token
B 2Wild-Image-Video-Gen-Studio A single-file, zero-dependency Python web panel for remotely driving a set of ComfyUI instances to generate images and video, suited to people with multiple render machines. The scope is restrained and the approach plain, but the models it names look unreleased, so real usability needs checking.
Plain-language web remote control for a fleet of ComfyUI instances (Qwen-Image-2.1 stills, MiniMax-H3 video). Stdlib Python, no dependencies.
B quicktunnel Hides Xray behind Cloudflare Tunnel with one command, requiring no public IP or open ports, with an install wizard and QR code. The problem is that many similar one-click scripts exist, and this one shows no visible improvement.
Xray (VLESS/WebSocket) behind a Cloudflare Tunnel (trycloudflare.com). One-command installer with a setup wizard, service, management CLI, and a terminal QR code. No public IP, no open inbound ports.
B CnaEmulator A standalone simulation and testing environment for Cobalt Strike .cna scripts, a genuinely rare niche. But it is very narrow, has only 12 stars, and shows no verifiable runtime evidence, so it looks like a personal experiment.
CnaEmulator is a standalone, general-purpose development, emulation, and testing harness for Cobalt Strike Aggressor Scripts (.cna)
B gem-x.cpp Converts NVIDIA's human pose tracking model to GGML for local skeleton extraction, a direction and taste that fit. But it takes half a day to build, with no examples or weights, so it likely will not run if installed now.
Track live or prerecorded humanoid movements to produce a 3D skeleton. C++/GGML conversion of NVIDIA's GEM-X
B pi-session-hub Aggregates and searches past sessions from different coding agents like Claude Code, Codex and OpenCode, and lets you resume them, hitting the real pain of losing context when switching tools. But it only works within the Pi harness, so the audience is very narrow.
Cross-harness session hub for Pi: browse, search and continue sessions from Claude Code, Codex, OpenCode, Crush and JCode without leaving your chat.
B minesweeper Turns Minesweeper into a benchmark for LLM agents: same board, up to nine models in parallel, unified timing and tool layer. It reveals planning ability better than a chat arena, though the task is single and stars are still few.
A Minesweeper benchmark for LLM agents. One identical board, up to nine models in parallel, one clock, one tool layer.
B mkwii-ps5 Uses WiiCompiled to statically recompile the Wii version of Mario Kart for PS5, with AGC rendering and DualSense support. The technical approach is solid and useful to console recompilation hobbyists, provided you have a homebrew-capable machine and the original disc.
Mario Kart Wii PAL on PS5 ; native static recompilation with WiiCompiled, AGC rendering and DualSense controls.
B Color-Pallet A pure-browser palette, gradient and contrast tool, cleanly made but not beyond existing tools like Coolors. Color tools are saturated, so only worth it if you need self-hosting or want to read the source.
A fast, open-source color toolkit for developers and designers. Explore colors, build palettes and gradients, convert formats and check contrast. Runs entirely in your browser.
B Light-sprite-renderer A small procedural sprite generation and rendering tool for Godot that works with light nodes. The idea is concrete and the scope restrained, but it serves only a niche art workflow, docs are thin, and stability is unclear.
Simple Godot-based tool for procedurally creating and rendering sprites to be used by Godot's light-nodes
B claude-vs-codex A 93-second pixel animation drawn in code, which the author admits was a single generation; it is fun enough. But it is an artwork, not a tool, with nothing reusable beyond amusement.
A 93-second pixel-art cartoon about Clawd and Codex, written entirely in code by Claude — one-shot, 1 user turn.
B ah-douyin-clean-downloader Feeds a Douyin link to Codex to download the original video without a watermark and archive it by creator, explained in one sentence. But watermark-free downloaders are everywhere; the only novelty is using an agent as command-line glue, and platform-rule risks go unmentioned.
Hand a Douyin link or share code directly to Codex to download the original video without platform watermarks and auto-archive it by creator
B sirca-shell The author's personal KDE Plasma 6 Wayland top bar and dock, complete with blur and popup animations. It is a personal desktop config upgrade with almost no reuse value for others, since Plasma's built-in panel suffices.
My own top bar and dock for KDE Plasma 6 on Wayland. Popups grow out of the bar, everything is blurred and see-through, light/dark and 7 colours.
B ninfer-ternary-bonsai-ada Runs a 27B ternary model at 2.125 bits per weight on NINFER, targeting Ada sm_89 and Linux. Ternary inference is interesting, but this is a derived config for someone else's engine, with narrow hardware and near-zero docs; treat it as an experiment log.
Ternary (2.125 bit/weight) Bonsai 2 27B on the NINFER engine — Ada sm_89, Linux. Derivative work of Neroued/ninfer; see README.
B Parasite-Eve-HD-Remaster Uses static recompilation to bring Parasite Eve to PC at 4K/60fps; the approach is credible and the wording admits it's in progress. But the repo is less than a day old with no downloadable build, so for now just watch.
Native PC port of PARASITE EVE via static recompilation — true 4K rendering, 60 fps, HD Remaster Not an emulator. Work in progress.
B laya-mlx Brings typed decision models from cloud LLMs to local Apple MLX inference, completing classification and short decisions in 7–14ms without generating text. 1,763 stars in 1.5 days mainly come from mutual promotion within the Jev/Laya ecosystem. If it truly becomes an on-device decision layer, private inference licensing is possible; the biggest risk is that the whole Jev ecosystem looks like mass-manufactured pseudo-hype lacking third-party reproduction.
Native MLX runtime for Laya typed decision models — 7–14 ms short decisions on M3 Max. No text generation, PyTorch, or cloud API.
B underclass An OpenAI-compatible account-pool proxy pins sessions to one account to preserve prompt cache, cools down until quota reset, and fails fast by earliest Retry-After when the pool is dry, targeting unused subscription quota. The author's traffic drives stars; it could be a team LLM gateway, but it violates subscription terms with high ban risk.
underclass: an OpenAI-compatible pooling proxy that pins sessions to one account (prompt cache stays warm), cools quota-exhausted subscriptions until their window resets, and fails fast with the earli…
B open-mcp-gateway Turns MCP servers into an OpenAPI 3.1 gateway usable directly as tools in Open WebUI, a positioning stated in one sentence. Open WebUI users can replace mcpo today, with performance as the pitch; limits are serving only this ecosystem and unverified maturity.
High-performance MCP-to-OpenAPI 3.1.0 gateway, built for the Open WebUI ecosystem.
B BootLife Runs Conway's Game of Life in a 512-byte x86 boot sector, using VGA memory directly as the grid, viewable in QEMU with one command. Anyone into assembly and low-level demos can read it in ten minutes; the scope is clean and the description honest. The limit is that it is a pure toy with 512-byte precedents, so do not expect extensibility.
Conway's Game of Life running directly from a 512-byte x86 boot sector, using VGA memory as the simulation grid.
B claudecut Strips Claude Code down to a single shell tool, a 36-token system prompt and limited context, to observe whether an agent can still work under a minimal setup. A good controlled experiment for anyone studying or trimming coding agents; the limit is that it is a test rig, not a daily productivity tool.
Run Claude Code with almost everything cut away: one shell tool, a 36-token system prompt, and a capped context window.
B crypto-tax-calculator A crypto tax calculator covering major wallets, exchanges and block explorers, solving multi-account data reconciliation at tax season where users naturally pay. Its 84 stars come from seasonal crypto tax demand, but Koinly and CoinTracker have years of funding, and data sources plus compliance trust are hard barriers—without differentiation it's hard to break out.
A comprehensive crypto tax solution supporting leading wallets, exchanges, and blockchain explorers.
B ritridata A Rust forensics-oriented toolkit that read-only parses MBR/GPT, publishes file signature definitions, and generates synthetic disk images. People building disk and recovery tools can skip writing the parser again; but three features are crammed into one repo, and long-term signature-library maintenance is the real cost.
Read-only MBR/GPT inspection, public file signature definitions, and synthetic disk image tools. AGPL-3.0-only.
B GazeVeil Uses AirPods connection status to tell whether you are at the computer, auto-hiding the screen when you leave, as a macOS privacy tool. The idea is concrete and narrowly scoped, smarter than generic privacy screens; limits are its AirPods and system-permission dependence, small codebase and thin docs.
AirPods-aware screen privacy for macOS
B li-ion-soc-soh-toolkit A zero-dependency C99 library for lithium battery SOC/SOH estimation, with Ah integration, OCV correction, an EKF calibration pipeline, and HPPC parameter identification scripts; the scope is restrained and the description doesn't oversell. Useful directly for BMS or battery experiments, but too new, lacking measured comparisons and tests.
Li-ion SOC/SOH estimation library (C99, zero-dependency) + PC calibration toolchain (Ah integration + OCV resync + EKF), with HPPC parameter identification scripts
B nice-mermaid.nvim Renders Mermaid diagrams directly as Unicode character art in Neovim, so you need not switch to a browser while writing docs. Clear scope, does one thing, works once installed; limits are that complex diagrams and narrow terminals blur, and heavy needs still require mermaid-cli.
Render Mermaid diagrams as Unicode terminal art in Neovim
B TinyGPU A 64-bit vector accelerator for the CVA6 RISC-V CPU, implemented in Verilog with a three-level memory hierarchy and unusually concrete technical detail. Valuable reference for architecture or RISC-V people, but useless to ordinary developers and lacking validation data.
Custom RVV 64 bit vector accelerator for the CVA6 RISC-V CPU. SIMD style layout with 3 memory hierarchies inside the accelerator.
B Glide LAN keyboard and mouse sharing for Omarchy, built on Lan Mouse, letting one set of peripherals control multiple machines. The description is honest and not oversold, and Hyprland users can just install it; but it's essentially a wrapper over an existing tool, and Deskflow and Synergy already cover the same need.
One desk. Every machine. LAN mouse and keyboard sharing for Omarchy, powered by Lan Mouse.
B footprint-osint A defensive tool that checks whether emails, passwords, domains, and usernames appear in public breach data, with a CLI and local GUI and no API key needed, practical for self-checks. But tools like holehe and h8mail are already mature, and it does not cover data sources or false-positive rates, making reliability hard to judge.
Defensive OSINT exposure checker: see what an email, password, domain or username leaks in public breach data. No API keys. CLI + local web GUI.
B jevmail A read-only local tool that sorts Gmail inbox into reply-needed, notifications, promotions, and spam, with honest cost and speed figures. But AI email triage is already crowded, and it is tied to an obscure Jev model, losing its selling point if the model changes.
Open-source AI email triage for Gmail. Sorts your inbox into Needs reply, Updates, Promos, Sales and Spam with Jev, TypeSafe AI's decision model, via Vercel AI Gateway. Read-only, runs locally, 1,000 …
B cida A Chinese-language Skill for setting rules for AI, requiring writing driven by task goals, textual evidence, and discourse mechanisms. It's genuinely somewhat useful for those needing Chinese copy quality; but it's still prompt engineering, results vary with the base model, and there's no comparative evaluation.
Cídá · A modern Chinese high-quality expression Skill. Lets AI follow task goals, textual evidence, and discourse mechanisms to say worthwhile ideas clearly, naturally, and with appropriate restraint.
B vksr The official implementation of an ECCV paper using Vecchia approximations for large-scale kernel surface reconstruction, with a clear problem scope and no exaggerated claims. Researchers in point-cloud reconstruction will care; but it serves only peers, with no easy interface or visualization, far from everyday development.
Official implementation of "VKSR: Scalable Kernel Surface Reconstruction Using Vecchia’s Approximation" (ECCV'26)
B steam-controller-calibration A small Python tool for calibrating Steam Controller stick drift, with a single purpose and an honest name, a real need for anyone still using this discontinued controller. But it targets one model only and doesn't explain the basis for its calibration method, reading more like a personal script.
Steam Controller Sticks Calibration Utility
B ThinkingOrbs A set of dot-matrix 3D loading animations for SwiftUI AI/agent interfaces, nine designs in two sizes, with restrained documentation admitting they are just animations. Saves some design effort for iOS AI interface builders, but the ceiling is a few loading components.
Dotted, honestly-3D loading indicators for AI and agent interfaces in SwiftUI. Nine hand-tuned designs, two tuned sizes.
B Multi-Agent A multi-agent orchestration engine featuring persistent sessions, durable mailboxes, no manager LLM, and per-task token accounting. Speaking in measurable cost and speed is the right direction; but the space is crowded, and it must beat LangGraph and CrewAI with real data.
One orchestration engine for multi-agent work: warm persistent sessions, a durable mailbox, no manager LLM in the loop — with per-task token accounting so speed and cost are measured, not assumed.
B bonsai2-small-gpu It targets inference of the ternary 27B model on consumer GPUs with tiered deployment and a 1.5x decode kernel, and the audience is clear. But it's full of jargon like prismml and qwen 3.8 mtp head, with no install or verification steps, and 36 stars in 1.4 days is hard to verify.
run ternary bonsai 2 27b well on the gpus people own: serve lines per vram tier, a 1.5x decode kernel for the prismml fork, the qwen 3.8 mtp head grafted back, sweeps by pr
B comfyui-studio-remote A zero-dependency stdlib Python web remote for multiple ComfyUI instances in plain language, which is a practical angle. But the description contains seemingly fabricated model names, so whether the README gives real integration steps is unclear; check the code first.
Plain-language web remote control for a fleet of ComfyUI instances (Qwen-Image-2.1 stills, MiniMax-H3 video). Stdlib Python, no dependencies.
B reader-source An AI reading workspace on macOS supporting PDF/EPUB/TXT, annotations and notes, and chatting with books, with local storage. Features are clear and there is a download link, but this is one of the most crowded tracks and it offers no new reading or retrieval method, so it is yet another AI reader.
Open-source native macOS AI reading workspace for PDF, EPUB, and TXT with highlights, notes, book chat, and local-first storage. Download: reader.marvy101.com/download
B Xover-windows A Windows client for P2P synchronized music listening, with a clear idea: several people hearing the same song at the same progress. Useful for small remote listening circles, but Spotify Jam and similar products already exist, and a Kotlin Windows client raises maturity doubts.
P2P listen to music together
B open-jev-typed-decision-engine Reproduces a 150M typed decision model, giving checkable numbers like 0.697 versus 0.727, better calibration, and half an hour of Colab training, more honest than its cluster peers. But the reproduced subject lacks a public source, so value depends on whether you trust the setup.
Open reproduction of TypeSafe Jev: a 150M typed decision engine (noul/choice/score in one non-autoregressive pass, calibrated confidence). 0.697 vs Jev's 0.727, 2.5x better calibrated, 4x faster, free…
B gps A gamepad button triggers speech-to-text into WoW chat, via a plugin plus a cross-platform helper. A thoughtful little tool for gamepad WoW players, and the description is honest; but the audience is tiny, injecting input into the game risks bans and compatibility issues, and voice input methods are the alternative.
GamepadSpeak: controller-triggered voice-to-chat for WoW Forever (addon + cross-platform helper)
B edge-scanner A real-time intraday US stock scanner with a self-arranging browser panel, saving short-term traders from writing their own screeners. But the data source, latency, and signal definitions are all unstated, so it can't be assessed before live trading; established screeners like Finviz and TradingView are more mature.
Real-time intraday stock scanner for US equities with a browser dashboard you lay out yourself
B otelyssey It aims to give various coding agents unified OpenTelemetry instrumentation plus an issue-driven auto-marketplace distribution plugin, hitting the real gap of unobservable agent behavior. But it's more vision than substance, with no clarity on which agent works today or what the data looks like.
OpenTelemetry plugins for every coding agent, in the Agent Plugins format — a marketplace that runs itself: open an issue, the repository validates, reviews, lists and follows your plugin.
B abide Makes coding agents follow project rules (naming, directories, commit conventions), addressing the common friction of agents that write code but violate team conventions, with 156 stars showing real demand from engineering teams; it could charge per seat as a CI or editor plugin, but Cursor and Claude Code are natively absorbing rule files, so the plugin slot is easily swallowed.
Make your coding agent abide by all your project rules
B jeff A self-hostable decision service positioned as a TypeSafe jev alternative, initiated by LangChain's author, showing typed decision has been noticed by the mainstream ecosystem; it is clearly valuable to teams unwilling to hand decision data to third parties, and 108 stars is a decent start, but it must match official accuracy while facing pressure from openjev and simple-jev.
A self-hosted drop-in replacement for TypeSafe's jev, powered by GliFormer.
B guizang-product-video-skill Reuses real product components and design language to generate software update promo videos from code, chaining storyboard, music, motion sound effects and rendering into a Claude Code/Codex skill, solving the cost of product videos for small teams; the author brings traffic, so a template marketplace or done-for-you service is viable, with the risk that end-to-end video generation models produce the video directly.
Guizang product video skill: reuse real product components and design language to make software update promo videos with code. Includes storyboard copy, original music, motion sound effects, and video…
B splash A local inference engine for Apple Silicon that schedules around the model itself, targeting the memory and throughput pain of running local LLMs on Macs; 293 stars in a day show people still seek easier options beyond MLX and llama.cpp, but those are already de facto standards, so a self-built engine can hardly charge license fees and must rely on hosting or enterprise support.
A local inference engine for Apple silicon, built around the model.
B jev-align Uses Jev plus GEPA to train calibrated classifiers from human feedback, turning 'label a few examples and ship a small model' into a reusable process, practically valuable for enterprise moderation and ticket classification; it could sell MLOps services, but the risk is that platforms like OpenAI absorb fine-tuning and calibration as hosted capabilities.
Build calibrated AI Functions from human feedback using Jev and GEPA.
B traceclause A local-first document requirements and evidence-checking workbench that outputs traceable citations and keeps human review, hitting the hard need for verifiable RAG in compliance, bidding and audit scenarios. Its 60 stars are small but the direction is right; it could be enterprise compliance software, with the risk of legal AI and general RAG platforms pressing down simultaneously.
Local-first document requirements and evidence review workbench with traceable citations and human review.
B geo-sleuth It infers a photo's shooting location from OSM geometry, skyline elevation, satellite imagery and street view and shows the derivation, turning the gray skill of geolocation into a compliant agent skill. Its 108 stars show real OSINT and news-verification demand; risks are privacy controversy and map API licensing costs, with monetization via investigation, insurance and other verticals.
An agent skill that finds where a photo was taken — OpenStreetMap geometry, elevation skylines, satellite imagery and street view — and shows its work. Works with Claude Code, Codex, Cursor, Gemini CL…
B openjev Implements a Jev-compatible open-source System One decision service on DiffusionGemma, replacing the commercial API with self-hostable inference, directly valuable to teams wanting typed decision in private environments; 123 stars show the ecosystem seeking open alternatives, but it must match official model quality and faces competition from jeff and simple-jev.
Open, Jev-compatible System One decision server on DiffusionGemma
B jev-pruner Uses Jev to decide which lines to keep before long Bash output enters the model context, saving tokens and improving agent judgment, a pain-point plugin Claude Code users can install immediately; 114 stars come from plugin marketplace spread, monetization can only be team editions or subscriptions, and the risk is agent platforms building output trimming in natively.
Claude Code plugin: trim long Bash output with TypeSafe Jev before the model sees it
B omarchy-iphone-mirror Projects an iPhone onto the desktop on Omarchy (Arch + Hyprland) and takes over input separately over USB or Wi-Fi, filling a gap in Linux phone integration. Omarchy users can try it directly, but it's tied to one distro config and the underlying implementation is unclear, so it's not portable.
On-demand iPhone mirroring and focused input for Omarchy over USB or Wi-Fi.
B mac-shortcuts-mcp Uses Apple Shortcuts as a bridge to give any agent read-only access to Mac calendars, reminders, notes, and clipboard, plus a catalog of 539 Shortcuts action parameters. The boundaries are honestly stated (read-only, local, no account), a rare restraint among MCP projects; the limit is macOS-only and requires building the shortcuts yourself.
Let any AI agent read your Mac — Calendar, Reminders, Notes, Contacts, Clipboard, Location, Weather — through Apple Shortcuts. Read-only, local, no account. Includes a harvested catalog of 539 Shortcu…
B Tcalm A zero-dependency CLI that tracks Islamic prayer times in the background on Linux/macOS, auto-pausing playback and popping a notification at each time. The scope is tight and usage clear, ready to use for developers who pray; it's single-purpose and doesn't aim to be more.
A lightweight, zero-dependency command-line utility for Linux and macOS that monitors Islamic prayer times (Adhan) in the background, automatically pausing media playback during prayer times and deliv…
B EdgeSSH A lightweight WebSSH workbench on Cloudflare Workers with host management, encrypted storage, SFTP, and process monitoring. Low deployment cost is the main selling point, suited to managing a few personal machines; Workers' limits on long connections and SFTP need to be spelled out.
A lightweight WebSSH workspace on Cloudflare Workers with host management, encrypted storage, SFTP, process monitoring and globe visualization.
B iphone-tailnet-bridge It proxies a paired iPhone's CoreDevice connection over Tailscale using Bonjour and socat, letting remote machines debug real devices. The scenario is specific and the approach clean, immediately useful to remote iOS developers; the limits are that it covers only one link and the docs are thin.
Proxy a paired iPhone CoreDevice connection over Tailscale using Bonjour and socat.
B cli-faq-shortcuts Mines your own Claude Code and Codex history for repeated prompts and turns them into short commands installed as Agent skills across three major CLIs. The angle is concrete and targets real repetitive work, but the implementation is likely just prompts plus scripts, so you'll have to test the speedup yourself.
An Agent Skill that turns the asks you keep typing into short commands, mined from your own Claude Code and Codex history. Works in Claude Code, Codex and Cursor.
B apk-reverse A one-stop script collection for Android APK reverse engineering that chains decompilation, static scanning and other common steps, plug-and-play for security practitioners and risk-control research, with 197 stars in 0.4 days from the Chinese reverse-engineering community; but jadx, apktool and MobSF already cover most needs, making commercialization hard and legal compliance the ceiling.
Suitable for Android APK reverse engineering analysis
B skyline-speeder It rewrites congestion control with eBPF struct_ops and a Rust control plane, speeding up satellite and cross-border links with non-congestion random loss—a real, under-served networking pain point, with 74 stars showing research interest. But the kernel 6.12+ requirement limits deployment, so commercially it's only dedicated lines or cloud vendor customization.
Sender-side TCP acceleration: eBPF struct_ops congestion control with a Rust control plane. For links with non-congestive random loss. Kernel 6.12 LTS+.
B Awwards-mcp Gives coding agents retrieval, screenshots, and design-element extraction from award-winning sites, pitched as a free Mobbin alternative, usable by frontend and UI people once installed. Concerns are its reliance on scraping, unclear long-term availability, and the recent cluster of "design inspiration MCP" projects.
Free, open-source MCP server that gives AI coding agents (Claude Code, Codex, Cursor, OpenCode, ZCode…) design inspiration from award-winning websites — search with real screenshots, pull design DNA, …
B coastal-simulation A coastal simulation solving the shallow water equations in real numbers with Three.js, featuring breaking waves, foam, wet sand and a shoreline walking view. Rare among WebGL water demos for explaining its solution clearly, it's good code to read for graphics people; the limitation is that it's a demo, not a reusable library, with no docs yet.
A real-time coastal simulation in Three.js, pairing a shallow-water solver with breaking swells, whitewater, wet sand, spray and shoreline exploration
B react-native-nitro-rolling-number A React Native rolling-number (odometer) component with a native implementation via Nitro Modules, by the author of react-native-zoom-toolkit, so quality has a baseline. Scope is restrained and it's installable today; the downside is many similar components, so only swap when you need native performance.
Native rolling number (odometer / ticker) view for React Native, built with Nitro Modules
B jev-semgrep Uses Jev to score each line of code semantically, supporting AND/OR/NOT combinations and cross-language, Japanese-English mutual search, making 'grep by meaning' a usable CLI; 74 stars show real demand for semantic search, but it overlaps with vector search and IDE semantic search, making it hard to charge independently or become a standard tool.
grep by meaning, across languages. TypeSafe Jev scores every line against a meaning; combine meanings with AND/OR/NOT. A meaning-based grep. Search English in Japanese and Japanese in English
B Pixel-Camera It backports the Pixel 11 camera Look, 5x portrait, and 10x zoom toggle to Pixel 6–10, with clear scope and device list, so older Pixel users can decide in ten minutes. The catch is it requires root/Magisk, and such camera feature ports have precedent, so innovation is limited.
Unlock newer Google Pixel camera features on older Pixel devices. This project brings Pixel 11's next-generation Camera Looks to Pixel 6–10, enables 5x Portrait Mode on Pixel 6 Pro to 10 Pro, and adds…
B 2ship2harkinian A native PC port of Majora's Mask via recompiling the N64 binary into executable code, a real contribution to the porting scene that's playable once built. The limit is you must supply your own ROM, the audience is narrow, and the repo itself has little info beyond a website link.
https://www.harbourmasters.org/game/2ship2harkinian
B ComfyUI-Hyperflow It wires Video Rebirth's 8-step LoRA into ComfyUI's native MiniMax-H3 without touching core files or needing a custom sampler, and the wording stays restrained. Useful for ComfyUI users chasing new video models, but it depends on a not-yet-common weight, so the ceiling is clear.
ComfyUI node pack for HyperFlow [Video Rebirth's] 8-step LoRA for MiniMax-H3 (video + audio), ported onto ComfyUI's native MiniMax-H3 model. No ComfyUI core files are modified; no custom sampler is ne…
B ddc A Rust rewrite of DEX-to-Java decompilation featuring progressive analysis and a bilingual CLI, where faster startup and lower memory matter for batch sample analysis; 115 stars show Android reverse engineering still lacks modern tools, but jadx has a deep ecosystem and accuracy is the hard metric, making direct charging difficult and leaving only enterprise security suite licensing.
DEX → Java decompiler in Rust — fast, progressive analysis, bilingual CLI
B hype A local Markdown presentation tool in C++ with a visual slide editor, taking a native route opposite Marp/Slidev. The one-liner is clear and suits frequent tech sharers tired of heavy web tools—check the demo; limits are just 11 stars in a day and no platform or setup details.
Simple Markdown presentations with a visual slide editor
B Audio_Bridge Turns Windows into a Bluetooth A2DP/LE Audio receiver so phone audio plays through the PC speakers, with a tray app and auto-reconnect; the need is real, the scope restrained, and the bilingual description not exaggerated. Novelty is limited, as store apps like Bluetooth Audio Receiver and AudioRelay already cover most of this.
Lightweight Bluetooth audio receiver (A2DP Sink / LE Audio) for Windows. Connect your phone or tablet to your PC and play its audio through Windows speakers. Tray app with auto-reconnect, performance …
B mpc2000-emulator An experimental emulator running the Akai MPC2000 sampler on the MAME engine, aimed at old-school beatmakers who can't afford the hardware. The direction is specific and the description restrained, but it calls itself experimental and has only 8 stars after a day, meaning nobody has really gotten it working; similar needs are currently met with existing MAME drivers and sampler software.
Experimental MPC2000 Classic emulator with a MAME-based engine.
B zia A Sine plugin that makes Zen Browser look and behave like Dia. Its 28 stars show fellow fans, but this is pure appearance mimicry while functionality stays with Zen, a typical 'yet another' that needs ten minutes and no bookmark.
A Sine mod that makes Zen Browser look and work like Dia.
B ghost A full backfill and read-only query API for ERC-20 fund flows on Robinhood Chain, clearly stating it pulls 1.4 billion transfers via HyperSync and stores them in ClickHouse; the scale and boundaries are honestly described. On-chain data analysts can pick it up directly, but it serves only one new chain, so its value rises and falls with that chain.
Fund-flow pipeline for Robinhood Chain (chain 4663): full ERC-20 Transfer backfill via HyperSync, 1.4B transfers in ClickHouse, read-only JSON API.
B vnpresence Syncs the cover, title, and reading time of the visual novel you're reading to Discord status; a small thing done cleanly, useful to VN players once installed, and the description isn't exaggerated. The limit is that many similar Rich Presence scripts exist and the technical depth is low.
Show the visual novel you're reading on Discord - cover art, title, and elapsed time, automatically.
B should-i-submit A pre-submission local self-check for bounty reports that helps authors avoid triage rejections. The audience is clear and the pain real, but it doesn't say where checks come from or whether there's a rule set, so results depend on author experience and it isn't worth a dedicated look yet.
A local, safety-first pre-submission reviewer for bug bounty reports.
B everything-search-skill Wraps Voidtools Everything's local index search as an agent skill, a piece Windows users genuinely lack, and 42 stars in 0.7 days shows the demand. But the implementation is a thin wrapper, and the description is stuffed with ⚡ and tool-name stacking, showing little craft.
⚡ Ultra-fast, index-powered local file search skill for AI agents (Cursor, Codex, PI-Desktop) using Voidtools Everything
B bend-sha256 Implements SHA-256 in Bend with a machine-checked correctness proof against the independent FIPS 180-4 spec, the most novel concept in this batch. Unfortunately its audience is only formal-methods and parallel-language enthusiasts, its practical usability is near zero, and it's better as reading than as a tool.
SHA-256 in Bend with a machine-checked functional correctness proof against a separate FIPS 180-4 specification.
B NobodyTdxTQ Generates A-share quant strategies from natural language and backtests them on backtrader, using the Tongdaxin TQ interface for data, with a web frontend that produces reports in one click, lowering the bar for beginners. The limits are the recent flood of similar natural-language stock-picking plus backtesting tools and compliance and stability questions around the TQ data source.
Generate quantitative strategies from natural language and backtest them, an A-share quantitative backtesting system based on [backtrader], with data from the Tongdaxin TQ API. Built-in web frontend: …
B live-jev Controls Ableton Live with one sentence in Japanese or English, summoned with ⌘⇧Space and executed on speech. Among the jev-prefix batch this has the most real need, but it depends on TypeSafe interfaces and doesn't say how broad the command set is or what happens on failure, so wait for it to grow.
Control Ableton Live with one short sentence (Japanese / English). Summon with ⌘⇧Space, type or dictate, done.
B Glyd A Rust compression library claiming smaller output than zstd -3 and decoding twice as fast as LZ4. The specific numbers are a plus, but 10 stars in a day and a 'world's fastest decoder' claim need your own benchmark; with so many compression options, unreproducible data means nothing.
The world's fastest-decoding open-source compression: fewer bytes than zstd -3, 1.3x its decode speed, 2x LZ4. Rust + C ABI + CLI.
B iPhone-Duo-App-Patcher Patches Mach-O version numbers, UIDeviceFamily, and plist so the iOS simulator renders a foldable dual-screen layout. It saves work for iOS developers targeting foldables and states its scope clearly, but it is a binary-patching hack with a tiny audience; the alternative is configuring the simulator yourself.
Patches Mach-O build versions, UIDeviceFamily, and plists to unlock dual-screen foldable layouts in iOS Simulator.
B Frida0xKit It bundles common Frida scripts into a modular toolbox for Android runtime instrumentation and reverse engineering. It saves digging through snippets, but overlaps heavily with Frida's official examples and objection, and no killer feature is apparent yet.
A modular toolkit for Android runtime instrumentation, reverse engineering, and Frida-based analysis.
B jev-reviewer A browser tool for systematic review data extraction with verbatim citations, pointing each item in RoB 2, QUADAS-2 and similar templates to the source line; this is the most solid angle in the Jev series. Unfortunately the batch of repos appearing together drags down credibility.
Data extraction for systematic reviews, quoted from the papers. Ask a trial report and its supplements your extraction form or a RoB 2, ROBINS-I, QUADAS-2 or TIDieR template; Jev points at the lines, …
B pi-prefix-cache-compaction Context compression for Pi that reuses prefix caches from vLLM/SGLang/llama.cpp to avoid cold-start recomputation. The direction targets agents' real costs, but it's tightly bound to one specific client with no docs or examples, making it hard for outsiders to adopt directly.
Pi compaction that reuses the server's prefix cache (vLLM / SGLang / llama.cpp Anthropic endpoints): no cold re-prefill, plus a warm-up so the next turn starts from cache
B openjev-sglang Reproduces Jev's API (prefill-only) with open models so callers aren't locked into a single closed model. The author has good credibility, and 154 stars in 1.2 days shows real de-lock-in demand. It could charge for hosted inference or enterprise private deployment. The biggest risk is upstream price cuts or open weights swallowing the demand, and interface changes breaking the compatibility layer.
Jev-compatible API endpoint based on open models (prefill-only)
B repopilot Uses the Codex SDK to turn issue and PR feedback into tested, reviewable changes, filling AI coding agents' missing verification loop. 113 stars in 0.8 days rides a self-hosted-plus-human-review narrative. It could sell an enterprise self-hosted agent. Risk: Devin and OpenHands already occupy the space densely, and it depends heavily on OpenAI SDK pricing and interfaces.
Verification-driven AI software iteration, powered by the OpenAI Codex SDK. Self-hosted agents turn goals, GitHub Issues and PR feedback into tested, reviewable changes, with bounded execution and hum…
B tax-doc-classifier Achieves strictly 100% accuracy on page-splitting for 261 IRS tax forms at about $0.001 per page, solving manual tax-form splitting. 112 stars ride the Jev decision paradigm plus verifiable low unit price. It could sell to tax SaaS or charge per page. Risk: the niche is extremely narrow, customers may build in-house, and cheaper general multimodal models could replace it.
Tax document page classifier built on Jev decisions. 100% strict accuracy across 261 IRS forms, ~$0.001 per page.
B remiqora Combines the open-source ACE-Step and YuE2 music models with stem separation, MIDI, LoRA fine-tuning and a multitrack DAW in one local interface, fixing a fragmented music-generation toolchain. 90 stars ride the window opened by maturing open music models. It could become a local creation suite subscription. Risk: Suno and Udio lead in cloud quality, and upstream model interface changes must be tracked.
Local AI music studio unifying ACE-Step 1.5 and YuE2-3B in one Vue interface — text-to-music generation, stem separation, MIDI transcription, and LoRA fine-tuning, with a built-in multitrack DAW.
B NanoJev A Python reproduction of Jev's parallel decision-making, dynamic candidates and end-to-end training pipeline, valuable as a readable minimal implementation for researchers. 370 stars in 1.3 days shows a hot reproduction wave around Jev. Monetization is thin—at best a reference implementation or tutorial funnel. The biggest risk is that an upstream closed-model interface change breaks all reproductions at once, and several already exist.
A nano replica of Jev: parallel decisions, dynamic candidates, and an end-to-end training pipeline.
B zv-e10-creators-app-patcher A Smali patch that makes Sony's Creators' App recognize the original ZV-E10, which Sony never officially supported, restoring USB/Wi-Fi live view, phone transfer and card import. It solves a concrete problem for owners of old cameras with features the vendor stripped out, and is clearly described; the limit is it works for one model only and may break on an app update.
Patch Sony's Creators' App for Android so it works with the original ZV-E10, which Sony never added: remote shooting with live view over USB and Wi-Fi, Send to Smartphone, and Import from the card.
B omarchy-android-mirror This mirrors and controls a phone from the Omarchy status bar via adb and scrcpy, over USB or wireless, with keyboard and mouse passthrough. It saves manual setup for Hyprland desktop users and stays narrowly scoped, but is meaningless outside that environment and is still essentially an scrcpy wrapper.
Mirror and control your Android phone from the Omarchy bar — USB or Wi-Fi, keyboard and mouse, via adb + scrcpy
B human-atlas A Vietnamese-language 3D human anatomy atlas built with WebGPU/Three.js, with male and female models and interactive lessons that open directly in the browser. It targets medical students and local-language teaching, fitting the need better than generic anatomy apps. The limitation is that nothing says who reviewed the content for correctness, which is critical for learning material.
Cơ thể người — Atlas giải phẫu 3D tiếng Việt, WebGPU/Three.js, mô hình nam và nữ, bài học tương tác.
B pg-jev Queries tables in natural language inside Postgres, skipping hand-written SQL, with the value of working right on the database. 164 stars in 1.2 days rides Jev ecosystem hype rather than its own reputation. It could charge for a hosted version or enterprise plugin, but text-to-SQL is a red ocean with Supabase and built-in database AI competing, and it depends heavily on a third-party closed model—almost no moat.
Ask your Postgres tables questions in plain language. A PostgreSQL extension powered by TypeSafe's Jev.
B wp2shell-PoC A PoC exploit chain for the CVE-2026-63030 and 60137 RCEs lets defenders immediately reproduce and verify patches. 163 stars in 0.3 days shows the vulnerability is fresh and many sites are self-checking. There is no business model; value lies in security vendor citations and author reputation. Risk: the script gets weaponized by criminals, and heat vanishes once patched—it's a response-speed race.
CVE-2026-63030 & CVE-2026-60137 RCE chain proof-of-concept
B drainflo A local device that drips vinegar into an air conditioner condensate drain on a schedule, with no network or cloud, explained clearly in one sentence and quite specific. It is a real fix for homeowners with clogged drains, but the audience is tiny and it requires soldering hardware, more like reproducible build notes.
Local, cloud free automatic vinegar doser for an HVAC condensate drain line
B cache-tax Keeps Claude Code prompt caches alive during idle periods and estimates the extra cost before sending on cold start. The angle is narrow but hits a real cost pain point, with restrained, non-hyped description. Limitation: it only holds under a specific CLI's cost model and breaks if caching strategy changes—a category users could script themselves.
Keep Claude Code’s prompt cache warm during breaks and show the estimated cost before a cold send.
B feelings Probabilistic conditionals in Ruby: Feelings(message).like?("genuinely urgent"). One sentence explains one thing, and both the naming and description are restrained and fun. The author is a veteran of the Ruby community, so the code style is trustworthy. The limitation is that it ultimately calls a model, with no word on controlling latency or cost, and the use case is experimental.
Probabilistic conditionals for Ruby: Feelings(message).like?("genuinely urgent")
B TapWall It turns macOS desktop icons into 3D objects that fall and collide into piles, built in Swift with SceneKit and no third-party dependencies, with scope clear in one sentence. Desktop tinkerers can get it running in ten minutes; beyond fun it has no other use, so don't expect a productivity tool.
Native macOS 3D icon interactive desktop: real icons drop and collide-stack, with full speed-up, slow motion and reset by video interval. Swift + SceneKit, no third-party dependencies.
B stack-attention Following a lead from an ETH paper, these are implementation notes attempting to merge differentiable stacks into attention mechanisms. The author has long-standing credibility in ML reproduction circles and cites sources without exaggeration. Readers want to see whether the code keeps pace; stars are few and content is early, so it's worth following rather than using as a tool.
For following a line of research that augments attention with a differentiable stack, beginning with DuSell et al. at ETH Zurich
B adnullenum This walks through anonymous AD enumeration via SAMR and LSARPC in one pass, needing no credentials and producing structured, reusable output. It saves work for internal pentesters and asset mapping, though enum4linux and rpcclient already cover most of it; the value is in output format and automation.
One-pass anonymous Active Directory enumeration over SAMR and LSARPC — null session, no credentials, with structured reusable output.
B code-map It renders a codebase as a zoomable GPU treemap with Makepad, plus a switchable 3D city view, visually novel and suited to quickly seeing structure. But it's more of a tech demo, lacking evidence of performance and usability on large repos.
Fly over a codebase as a zoomable GPU treemap (2D and 3D city), built with Makepad
B idempotency-shield An Express middleware for Idempotency-Key that replays cached responses for duplicate requests, with memory or Redis storage. It is a drop-in small piece, but several idempotency middlewares already exist, and coming from an organization imitating dashscope lowers trust.
Express middleware that makes handlers safe to retry — replays the cached response for a duplicate Idempotency-Key instead of re-running it, with pluggable in-memory or Redis storage.
B FaceMac This uses a local SFace model for macOS face unlock, then types your saved password after recognition, all offline. The approach is concrete and the description honest about filling in passwords, but handing passwords to a script is risky; Howdy on Linux is the closest comparison.
Open-source face unlock for macOS — look at your Mac and it types your saved password at the lock screen. On-device SFace, Face ID style notch animation. Free, private, GPL-3.0.
B LiteMD A lightweight macOS Markdown editor with live preview, twenty color themes, and format conversion, while files stay your own .md. The positioning is clear and modest, but this is the most crowded category, with Typora and MacDown already there, so it needs a stronger reason.
A lightweight Markdown editor built for macOS: live preview, 20 color themes, built-in format conversion, and files always stay as your own .md
B ProcessingField A SwiftUI progress component where a shape made of many small squares drifts, folds, and breathes with the task. The scope is tiny and the description restrained, suited to Apple platform developers wanting a different loading animation, but nothing more.
A SwiftUI progress field: a grid of marks whose sizes describe one soft shape that drifts, folds and breathes while your app is busy.
B teeho A diagnostic tool for Xiaohongshu notes, launched with one sentence and shipping an Agent Skill interface. The angle is specific and aimed at a clear group of content creators, more concrete than generic writing assistants. But it depends on platform data access (prone to breaking) and shows neither diagnostic criteria nor sample output, so it's unclear yet how it beats manual review.
题火 Teeho — a Xiaohongshu note diagnosis tool, start analysis in one sentence | Xiaohongshu / RedNote content audit, note analysis and insights | supports Agent Skill
B sahne-plus A desktop app for Kick streamers that alerts on donations, subscriptions, and gifts, with a concrete feature description and a license. But mature alternatives exist, and it is tied to the Persian-language KickBot ecosystem, so the audience is narrow.
Sahne+ (صحنه پلاس) — open-source Windows app for Kick / KickBot donation, sub and gift-sub alerts with local transparent media. Apache-2.0.
B pstack-codex It ports someone else's pstack skill pack to Codex, with a marketplace and installer. The port honestly credits its source, but 47 skills and 23 playbooks amount to a content repo whose value depends on the original author.
Codex-native port of Lauren Tan's pstack: 47 skills, 23 playbooks, agent personas, multi-agent workflows, and Benny automation adapters. Includes a Codex marketplace and installer.
B webuntis-family-cards Renders WebUntis timetables, homework, and exams as a family Lovelace card in Home Assistant. The positioning is small and precise, and HA users with school-age children can benefit directly. But several similar integrations exist, this project is only a view layer, and whether it adapts to different schools' calendar entities is the key question the description doesn't address.
Family-focused Home Assistant Lovelace cards for timetables, homework, exams and school-day overviews from calendar entities.
B kalshi-trading-bot A market-making and momentum strategy bot for the Kalshi prediction market, with paper trading, risk controls and position management, and an honest disclaimer. But prediction-market arbitrage bots are a category with public tutorials everywhere, the strategy has no backtest evidence, and similar projects are so numerous readers can't tell which won't be abandoned after two days.
Kalshi trading bot for prediction markets — browse event contracts and watch the order book in real time, run market making and momentum/reversion strategies with paper trading, risk limits and positi…
B doc-router A small tool that routes document pages to different OCR engines based on page content—a sound idea, since different layouts do need different recognizers. But the description is one line, with no supported engines, routing criteria or accuracy comparisons; as a freshly built experiment it's worth watching, but not yet a usable component.
A Document OCR Router to help route pages based on content.
B sfmc A static analyzer for JAR files that flags suspicious behavior in code, aimed at developers who need to check third-party Java dependencies. The pain point is real and comparable tools (decompiling plus rule scanning) have to be assembled by hand, so it's worth a look; the limits are that neither the rule set size nor the false-positive rate is stated, and the intro is only in Russian.
A static analyzer for JAR files. Analyzes whether the code contains anything that could cause harm
B Web-Pilot A local-first browser agent that turns natural-language instructions into Playwright actions and scrapes or downloads resources. The description is honest and the boundaries clear, but browser agents are the most crowded track right now, with OpenAI Operator, browser-use, and others already established; this project doesn't say what sets it apart on safety constraints or local models.
WebPilot is a local-first AI browser agent that converts natural-language instructions into safe, structured browser actions, navigates websites through Playwright, extracts useful information, semant…
B gliner2-ultrafast Runs GLiNER2 locally in the browser for entity recognition to drive automation, with open weights. Putting NER in the browser is a sensible direction and the author honestly says it's work in progress, but the repo is too early, with no speed comparisons or usage examples, so for now it's just one to watch.
Real browser automation with local GLiNER2 inference and open weights. Work in progress. Contributions welcome.
B amb3r-slam A companion repo for a SLAM paper claiming kilometer-scale operation with a layered backend, potentially valuable to people working on robot localization. But only the paper title is present, with no code structure, data or reproduction notes, so quality can't be judged.
[arXiv'26] AMB3R-SLAM: Kilometer-scale SLAM with Hierarchical Backend
B design-systems Ten single-file design systems where you can open a browser and copy tokens, useful for frontend developers who want a decent-looking interface fast. But it's essentially a sample collection with no versioning or maintenance commitment, and full open-source design systems offer far more.
Ten design systems. Ten files. No build step, no framework, nothing to install. Open one in your browser, copy the tokens, ship something that looks like yours.
B yote A zero-dependency React form component with a unified prop vocabulary and built-in styles and animations. The positioning is clear and the description restrained, but forms are a red ocean, libraries like react-hook-form already cover most needs, and no unique capability is visible.
Form inputs for React. Styled and animated out of the box, with one prop vocabulary shared by every field. Zero dependencies.
B Show HN: Aclif – Agent CLI framework: one grammar, canonical names across SaaS Unifies various SaaS CLIs under one syntax and naming convention so agents can call different services with the same commands, addressing fragmented tool invocation; its 29 points were relatively high that day, and default schemas can be discovered without credentials. It could become a standard interface for the agent tool layer, with the risk that vendors may not maintain the adapters.
Simple discovery of the provider's default schemas is possible without any credentials (since they are built into the binary). If you want it to fetch customizations of your specific instance, give it…
B Show HN: OJ – A drop-in replacement for Vite in Rust Rewrites Vite natively in Rust to fix slow React builds and cold starts; 399 stars in one day shows real developer frustration with Vite's performance limits. Commercialization means acquisition by a big company or becoming React's default build chain, but the biggest risk is that VoidZero's Rolldown/Rspack already hold the ecosystem slot and plugin compatibility is very hard to catch up on.
B skillbox This is a self-hosted, versioned agent skill library supporting MCP and client scopes, solving the lack of versioning, permissions, and auditing when teams share skills; the author is a well-known developer with built-in traffic. It could become the standard layer for enterprise skill distribution, but the risk is that the official skill repository is free and more deeply integrated.
Self-hosted, versioned skills library for AI agents. MCP, scoped clients, and optional Jev recommendations.
B lead This is a Postgres full-text search extension for non-production environments, letting development, testing, and CI run TIN-compatible search SQL and solving the old mismatch between local and online behavior. Backed by PlanetScale, it could easily become a companion dependency alongside TIN; the risk is that it only works within their cloud product, locking in the audience.
A non-production Postgres text-search extension for exercising TIN-compatible application SQL in development, test, CI, and staging environments
B Show HN: Craigslist for agent skills, curated by a human Turns agent skills into a marketplace where ordinary people can buy ready-made capabilities instead of writing them; its 20 points show HN curiosity about skill trading. Commercialization means a commission marketplace, but cold-starting both supply and demand is extremely hard, and the strongest competitors are Anthropic's official skill catalog and free GitHub collections.
hey, i'm nick (@skeptrune on X). i have been using ai for a lot of things i'm not an expert in and thought it would be nice to try and buy paid skills for helping it accomplish those tasks more effici…
B tradingview-optimizer A Chrome extension from Roboquant that grid-searches TradingView strategies with walk-forward tests and heatmaps, sparing you custom backtest scripts, so retail traders can start directly. The feature scope is modestly stated, but watch extension permissions and the usual overfitting pitfalls.
Free, open-source TradingView strategy optimizer for Chrome. Grid search, walk-forward, heatmaps. By Roboquant.
B commit-miner From a familiar name in security tooling, it uses Jev to classify commit diffs and info, identifying fixes, CWEs and change types and generating risk summaries. Scope is clear with a CLI to try, but results depend on Jev and are hard to reproduce independently.
Classify Git commit diffs and messages with Jev. Bug fixes, security fixes/CWEs, and change types.
B mu300-linux Runs Linux (Ubuntu 26.04 with a self-compiled 5.4 kernel) on a ZTE F50/MU300 5G hotspot, with boot tools, kernel config and modem watchdog debugging notes. Hardware hobbyists will enjoy these first-hand notes, though the audience is tiny and the work is clearly ongoing.
Linux (Ubuntu 26.04 + systemd, custom 5.4 kernel) on the ZTE F50 / MU300 5G hotspot (Unisoc T760 / UMS9620): boot tooling, kernel config, modem/PM watchdog bring-up, Wi-Fi, findings
B HFTENGINE A replay console for hftbacktest that visualizes the order book, queue position and matching latency, addressing unexplainable market-making backtests and hard-to-trace slippage. Quant teams pay well, so a commercial version is possible. Risk: it's essentially a wrapper around an open-source crate, and competitors are firms' in-house backtest visualizations.
Replay console for hftbacktest market-making backtests: order book, queue position of every resting order, feed and order latency, fills and the raw feed from the collector, on real recorded Binance F…
B github-ranking-audit It audits GitHub repository search ranking signals, helping maintainers diagnose why they cannot be found and which metadata to add; its 192 stars show exposure anxiety is widespread. It could extend into an open-source growth service or SaaS, but ranking rules are decided by GitHub's black box, the capability ceiling is low, and one redesign could render it obsolete.
Audit your GitHub repository search ranking signals. Checks name, description, topics, README, stars, forks, and activity.
B ManiLoop A closed-loop evaluation framework for VLA/VLM robot policies, already integrated with MuJoCo, LIBERO and LeRobot, directly usable by researchers comparing policies. The value is a unified evaluation standard; the limit is that it's new, covers few benchmarks, and tasks must be added yourself.
An open framework for closed-loop evaluation of VLA, VLM, and other robot policies across benchmarks, designed for fair and reproducible comparisons. Current integrations: MuJoCo tasks, LIBERO, GPT, a…
B surfacemap A reconnaissance tool for authorized testing that first organizes a target site's hosts, routes, parameters and trust boundaries into an evidence-backed handoff document before discussing exploitation, built on client-side JS analysis. Its positioning is restrained and audience clear; similar recon tools abound, and 7 stars means it's unproven.
Maps one authorized web origin's attack surface: hosts, technologies, routes, parameters, identities and trust boundaries, delivered as an evidenced, ranked hand-over document before a single payload …
B utuvo-orbit Small power, network, and audio toggles in the macOS menu bar, native SwiftUI, MIT licensed, bilingual Chinese-English, with a restrained scope and no hype in the description. Alternatives include Stats and OnlySwitch, so retaining users will depend on polish.
Your Mac. In one small orbit. A native macOS menu-bar app for power, network and audio. SwiftUI + AppKit · English / 繁體中文 · MIT.
B jev-on-a-laptop Reproduces Jev-style parallel type decisions on Apple Silicon, with benchmarks and an HF demo, rare hands-on material for people wanting constrained decoding on local 1.5B-8B models. Its value depends on whether the reproduction is credible; currently there are only self-reported numbers and no comparison with other baselines.
Unofficial study: Jev-style parallel typed decisions on stock 1.5B-8B models on an Apple Silicon laptop. Benchmarks, research notes, and a Hugging Face Space demo.
B bidcraft A conversational bid-document system using LangGraph to chain tender parsing, knowledge-base distillation and revision — a very concrete scenario that genuinely pains Chinese engineering bidders. Currently it's more an open-source shell of an internal product, lacking deployment docs and sample data; watch for now.
BidCraft · conversational bid document authoring system: AI intelligently drafts technical bids, supports tender document parsing, lot selection, bid writing, knowledge base distillation, and conversa…
B harness-bridge A bridge layer that points coding harnesses like Claude Code and Codex at any OpenAI/Anthropic-compatible endpoint, a real need. But it shipped a library, CLI, tray app and web UI in half a day, the scope is scattered, and it collides head-on with claude-code-router and LiteLLM.
Point any coding harness (Claude Code, Codex, OpenCode, Pi, OMP, Crush, Copilot, Grok…) at any OpenAI/Anthropic/Responses-compatible inference endpoint. Core library + CLI + macOS tray + web UI.
B gksdud A small macOS tool for quickly switching between Korean and English input methods, with a one-line purpose that Korean users would want. But Karabiner, Gureum and other alternatives are mature, and it doesn't say how it's faster or how to install.
A snappy macOS Korean-English input switcher
B cdn-tunnel Forwards TCP/UDP traffic through a CDN using SOCKS5 + HTTP/2, with password authentication and an Android VPN wrapper, described concretely without hype. Suitable for self-built tunnels on restricted networks, but wstunnel and cloudflared are already mature, and nine days of stars do not show stability.
TCP/UDP tunnel over a CDN (SOCKS5 + HTTP/2) with password auth, GET/POST/stream transports and an Android VPN wrapper
B fanduck Complete replication materials for the Microduck robot: live demo, bill of materials, 3D-printed parts and boards, so makers can build their own desktop robot pet. The value is in verified parts details; the limit is that it covers only this one piece of hardware and depends entirely on the author going forward.
Fanduck · Microduck has been successfully replicated! Sharing live demos, bill of materials, 3D printing, boards, build tools, and a 3D mirror workbench. Build resources from our completed Microduck r…
B fruitflysynth Turns a recently published neural dataset of Drosophila courtship song into an interactive synthesizer, combining science and playfulness, interesting to audio-synthesis or neural-encoding people. More fun than practical and small in code, it's an oddity worth a look.
An interactive resynthesis of a male fruit fly's mating sounds utilizing the recently mapped neural dataset.
B oneplus-negroni-arch-linux A firmware repo porting Arch Linux to the OnePlus 10 Pro — mobile porting that few tackle head-on, of scarce value to those following phone mainlining. The limits are obvious: only one line of description, so flashing steps and usability are all on you to figure out.
OnePlus 10 Pro arch linux firmware
B UCAS-Desktop A UCAS campus toolkit: tray resident, lecture booking, course planning, online courses and PDFs, gathering scattered scripts into one Windows desktop app — highly practical for students. But it serves only one school, and automated course selection carries compliance risk.
UCAS Windows desktop assistant: runs in background in the system tray, Qingxin Classroom, lecture booking, course selection planning and automation, UCAS online video/PDF; modular integration with det…
B phicomm-s7-ha Replaces the Phicomm S7 body-fat scale's firmware with ESP8266, auto-connecting to Home Assistant via MQTT, so owners of this old scale can revive it. Scope is tiny and the writing solid, but it serves one discontinued device, so coverage is inevitably narrow.
Phicomm S7 body fat scale ESP8266 replacement firmware — Home Assistant MQTT Discovery
B higgsfield-api-skill Wraps Higgsfield's image and video generation into a skill coding agents can call, with model docs and resumable tasks; convenient for existing users, but it's a third-party API wrapper replaceable by switching platforms.
On-demand Higgsfield API image and video generation for coding agents, with model-specific docs and resumable jobs.
B ReactantNitro.jl A Lux training wrapper from EnzymeAD that compiles and runs a whole experiment in one pass with Reactant, cutting boilerplate. It comes from a trusted team with a restrained scope, but it has only 8 stars, thin docs, and is basically useless outside Lux.
Reactant-first training for Lux models: declare the experiment, compile once, train without boilerplate.
B EchoMind A local voice-memory assistant: real-time transcription plus RAG Q&A based on what you have said, with a clear privacy story. But transcription plus RAG is already a common combination, and beyond Rewind and Limitless it will have to win on engineering details.
Local-first voice memory assistant — capture speech, transcribe live, and ask a private RAG chatbot grounded in what you've actually said. No cloud required.
B seijaku MengTo's Three.js Kyoto courtyard walkthrough offers procedural buildings, gardens and seasonal changes you can stroll through in the browser, well judged. But it's a piece rather than a tool; only the scene code is reusable, and ten minutes gets you the effect.
A continuous interactive walk through a Kyoto residence, with procedural Three.js architecture, gardens, seasons and responsive photography.
B llmwiki-operational A persistent knowledge system using LLMs to continuously maintain technical manuals and SOPs, targeting the problem of stale ops documentation. But it lacks getting-started steps and effectiveness evidence, and doesn't explain how it differs from wiki-plus-retrieval.
A persistent LLM-maintained knowledge system for technical manuals, SOPs, and operational documentation.
B jev-ultrafast Nearly two thousand stars in a day, but the repo has no description, leaving name and buzzwords to guess its purpose; the stars come from the Jev topic effect rather than verifiable output, so wait for clear capabilities and usage.
B typesafe-computer-use Replaces multimodal large models with OCR plus action classification for computer use, at about $0.0002 per step, directly targeting the pain of agents operating computers being too expensive. Its 62 stars in 0.3 days hit a cost-sensitive nerve, and a hosted API is possible; the risk is that native computer use from OpenAI/Anthropic keeps getting cheaper and will crush it head-on.
Computer use for about $0.0002 a step: OCR the screen, classify the next action with TypeSafe, click. macOS.
B bashka Statically verifies script safety before curl | bash and manages installed software in one place, addressing trust in blindly running install scripts. 52 stars in 1.1 days reflect a real pain point hitting supply-chain security. It has potential to become a standard install-step dependency; risk is that static analysis can be bypassed and Homebrew and nix already partly cover the scenario.
Stop running unverified bash scripts to install software! curl <url> | bashKA for static verification of bash script safty and managing installed software
B Show HN: Fail2zig. A single-binary fail2ban replacement written in Zig A Zig rewrite of fail2ban replaces the regex engine with a bounded parser, persists state in SQLite, and writes nftables directly via netlink, addressing the risk of attacker-controlled logs triggering regex backtracking blowups. 0.4.0 just shipped with a website; HN discussion is cool but the direction is solid. Monetization is enterprise support or hosted protection; the risk is fail2ban's entrenched ecosystem and mature competitor CrowdSec.
Built fail2zig because I like fail2ban's log-to-ban model but wanted a smaller runtime with explicit limits on the work attacker controlled input can cause. fail2zig is a single static binary with bou…
B agent-monitor Visualizes the execution traces of Codex and Claude Code subagents, breaking down tokens and cost. Heavy coding-agent users can use it right away, and it targets a real pain point. The limit is it only covers these two CLIs and must chase them whenever their formats change.
Visualize your Codex and Claude Code subagents. Advanced traces view, tokens breakdown, and cost usage is also supported.
B VFD-GLB-Importer Adds the missing one-click glTF/GLB import to 3ds Max 2024–2026, including PBR materials, hierarchy, skinning, and animation. A clear pain point for those using Max in real-time pipelines, with a specific feature list; limits are covering only three versions and single-person maintenance.
One-click glTF 2.0 / GLB importer for Autodesk 3ds Max 2024-2026. PBR materials, textures, hierarchy, skin and animation. Created by Fady Derias.
B r-mdir It brings DOS-era Mdir III's two-pane file management and function-key operation back to the terminal in Rust, with a restrained scope and honest description; worth ten minutes for anyone nostalgic for old-style operation or wanting an extremely light TUI file manager, at the cost of a Korean interface and an early ecosystem.
A two-pane file manager that brings the screen, per-extension colors, and function-key-centric operation of the classic Mdir III to the terminal
B AE_StaffGenerator An After Effects ScriptUI panel that generates staff notation and notes inside a composition, saving manual note placement for score-animation work. Scope is tiny and the description honest; the downside is it only serves a narrow need inside AE.
A ScriptUI panel that generates staff notation and notes in After Effects
B Show HN: How Stale Is Your AI? Release age and training cutoff for 20 models Compiles release dates and training cutoffs for 20 models to answer how stale a model's knowledge is, aiding model selection; 69 points in 0.5 days, dense and shareable. Could become a data API or leaderboard ads, but the data is manually maintained and easily copied by aggregators, so barriers are low and independent charging is hard.
B QuickPPPoE A portable Windows PPPoE dialer that adds the auto-reconnect the built-in dialer lacks, ready to use for campus and broadband users. Scope is restrained and the description honest; the downside is many similar small tools and limited functionality.
A lightweight, installation-free Windows PPPoE broadband dialing and auto-reconnect tool
B influxdb2tsfile Migrates InfluxDB 1.x/2.x and line protocol to Apache TsFile with schema mapping, parallel export and resumable transfers. Aimed at teams storing long-term time-series data, the problem is concrete and engineering details are solid; the limit is serving only the TsFile ecosystem.
Command-line tool to migrate InfluxDB 1.x/2.x and line protocol files to Apache TsFile — auto schema mapping, parallel time-sliced export, resumable and verified.
B avdslim Cuts Android emulator AVD memory from about 8GB to 1.5GB, addressing the pain of memory-hungry emulators on Apple Silicon and Linux, inspired by simslim. Its 49 stars in 0.7 days show real demand, and it could be absorbed into the Android toolchain; the risk is being squeezed out by official Google optimizations.
⚡ Drop Android Virtual Device (AVD) emulator RAM from ~8GB to ~1.5GB on Apple Silicon & Linux. Inspired by simslim.
B Show HN: Friday – Self-hosted persistent memory for AI coding agents (MCP) An MCP layer giving coding agents persistent memory, solving Cursor/Claude/Copilot amnesia across sessions; it rides both MCP and coding-agent hype onto HN, but only 12 stars shows it is very early. Monetization is team-shared-memory SaaS; the risk is dense memory layers like mem0 and Zep, and model vendors' built-in memory directly squeezing it.
B Switch-PC-Stream Uses an unmodded Switch Lite as a 720p60 streaming secondary screen; the idea is novel and the scope small. Worth ten minutes for anyone with a Switch who wants a second screen. The limit is reliance on a specific streaming chain, with no measured latency or image-quality data yet.
Low-Latency 720p 60 FPS PC Game Streaming on a Stock, Unmodded Nintendo Switch Lite
B vscode-mcp-api Lets web-based AI call local MCP tools through a VS Code extension to read and write files, targeting the real gap that ChatGPT's web version cannot reach the local environment. The idea is direct and small in scope, but it hands local file permissions to web AI, so you must weigh the security boundary yourself.
Let web AI easily use MCP tools to manage local files
B CustomDance A SIGGRAPH Asia 2026 paper repo on 3D dance generation using coarse-to-fine human interaction control, a novel direction suited to animation and embodied AI researchers. The limit is the paper code isn't yet proven reproducible, so ordinary developers can't use it.
[SIGGRAPH Asia 2026] CustomDance: Customized 3D Dance Generation with Coarse-to-Fine Human-Centered Interactive Control
B easy-unlocker A tool for managing AI agent secrets: biometric approval on your phone, end-to-end encryption straight into the target process, with plaintext never touching chat or logs. Teams deploying agents will care. The catch is it needs a companion phone app and the ecosystem is still tiny.
Secrets management for AI agents: approve API keys and credentials on your phone with biometrics, delivered end-to-end encrypted into the target process. Plaintext never enters the chat, logs, or rela…
B awesome-cloudflare-selfhosted Curates open-source apps that can replace SaaS inside your own Cloudflare account, solving product discovery; 265 stars in 1.6 days shows the Cloudflare self-hosting topic is hot. The path is becoming the standard index for this niche, sponsorship, or cloud-vendor deals, but awesome lists are heavily homogenized and easily forked, capping commercial upside.
🍊☁️ Open-source* apps that replace a SaaS product, running in your own Cloudflare account
B privacy-router Privacy routing for local AI requests: detect sensitive content, confirm once manually, then automatically block or allow based on the registered decision. The problem is well chosen, since local models also need data boundaries, but currently there is only a concept-level description and 8 stars, so whether it lands remains to be seen.
Local AI privacy routing: discover sensitive content, review once, enforce registered decisions on future requests.
B qr A pure-browser QR code generator that exports SVG/PNG, with minimal code and no extra features, showing good restraint from the author; worth a look if you want an offline local QR tool, though similar tools are everywhere.
A simple, browser-only QR code generator with SVG and PNG export
B AHCINT A universal SATA AHCI driver for Windows NT3.5 through XP, letting old systems recognize new drives. Retro PC enthusiasts can use it immediately and the work is substantial; the limit is a very niche platform and non-trivial installation.
Standard SATA AHCI Controller Driver for Windows NT3.50, NT3.51, NT4.0, 2000 and XP x86 and x64
B ResetSpy It enumerates tenant accounts and registered authentication methods through Microsoft's self-service password reset portal, a common information-gathering technique in red teaming and audits, packaged as a standalone script for convenience; useful for identity security assessments, but it could also be used to probe accounts.
Enumerate user accounts and registered authentication methods via the Microsoft Self-Service Password Reset (SSPR) portal
B closetscan Turns a walk-around-your-closet video into a searchable clothing catalog, with a local MCP server and agent skill, targeting a specific and rare scenario. But it depends on capture quality and a multi-model pipeline, and at this scale stability isn't proven.
Turn a closet walkthrough video into a searchable wardrobe catalogue, with a local MCP server and agent skill
B ninfer-v100 Deployment scripts that get models like Qwen3 running on V100 (sm_70), with systemd services and a comparison against llama.cpp. Anyone with a V100 has been missing this path. The limit is that it's a fork of someone else's project with narrow coverage.
NInfer on Tesla V100 (sm_70): Qwen3.8-27B deployment, build/convert scripts, systemd service, and cross-engine benchmark (vs llama.cpp). Fork of Neroued/ninfer with sm70 vision-encoding speedup.
B scan2mesh Aligns lidar scans with a prebuilt 3D mesh for localization and tracking; the goal is focused and the implementation is C++, making it real engineering. But it doesn't specify the middleware or data formats it depends on, so onboarding cost is unknown; those doing robot localization may want a look.
Scan2Mesh is a high-performance mesh-based LiDAR localization and tracking system that estimates LiDAR poses by aligning incoming scans with a pre-built 3D mesh.
B open-transformers Open-sources a model family's pretraining pipeline together with audit replay tools, emphasizing reproducible training and verifiable process, still rare in public pretraining. Useful for teams doing pretraining and compliance audits, though the bar is high and documentation quality remains to be verified.
Reproducible LLM pretraining and audit-replay harness behind the OPEN model collection
B AirPoise Uses AirPods motion sensors for posture reminders and head-gesture shortcuts, running entirely on-device without a camera; repurposing existing hardware is a clean idea worth ten minutes for Mac users. The limit is it only covers the AirPods ecosystem.
Mac menu-bar app: AirPods posture coaching and head-gesture shortcuts. On-device, no camera, no cloud.
B car-systemui-win98-pod It turns the Windows 98 window look into a system UI plugin for Android Automotive, with draggable, resizable windows that take over real app tasks; a rare little toy for car ROM tinkerers, but it requires an Automotive OS 17 environment, so almost no one can verify it firsthand.
Windows 98 window chrome for Android Automotive OS 17, built on Scalable UI. Drag, resize, minimize and maximize real app tasks. Ships as a CarSystemUI pod.
B EasyThumbnailGenerator A UE5 editor plugin that renders static and skeletal meshes into transparent PNG thumbnails with presets and batch mode. Artists and programmers building asset libraries and store images can use it directly; scope is restrained, though similar plugins already exist.
Unreal Engine 5.8 editor plugin for generating transparent PNG thumbnails from Static Mesh and Skeletal Mesh assets. Includes live preview, camera and lighting controls, reusable presets, batch genera…
B livediff A small command-line tool that shows git diff in real time, with its purpose stated in one sentence and no extra ambition; but lazygit, tig plus delta already cover most scenarios, so there is limited novelty.
A live Git diff view cli
B dev101x-pentest-lab A Docker-packaged pentest practice range chaining blind SQLi, hashcat, path traversal, command injection, internal pivoting, and SUID privilege escalation, with flag validation. Concrete and ready to run, but similar ranges like Juice Shop, VulnHub, and HTB are already crowded.
Self-contained Docker Pro Lab for pentest practice: boolean blind SQLi + hashcat cracking, web path traversal, command injection, internal network pivoting and SUID PATH-hijack privesc — with live fla…
B a-rhythm-trainer A browser-based rhythm trainer for tapping practice and rhythm dictation, with preset, random, and custom questions. The features are clear and the scope restrained, suitable for music learners; but rhythm-training web tools are already plentiful, so novelty is limited.
A rhythm training tool supporting tapping practice, rhythm dictation, and preset, random, and custom exercises
B kakao-headless Turns KakaoTalk into a local CLI with encrypted credentials and send confirmation, letting scripts read messages and batch-send. A rare automation entry point for Korean developers; but it relies on an unofficial client protocol, so account risk and terms-of-service issues are on you.
Read. Approve. Send. A local KakaoTalk CLI with a managed Aside integration, guarded sends and encrypted credentials.
B BareDroid Starts Linux natively on Android via Termux, bypassing chroot. Few have tried this direction, but it is self-labeled experimental and likely works only on certain models, so ordinary developers cannot install it today. More a tech demo; wait for specific models and images.
[EXPERIMENTAL] Full Linux environment on Android phones — native boot, no chroot, Termux, PRoot, containers, or Android compatibility layers.
B Moss-Transcribe-WebUI A local transcription WebUI built on transcribe.cpp with speaker diarization, and the localization angle holds up. But Whisper-based WebUIs are already crowded, and it doesn't clarify speed or accuracy differences, so watch for follow-up.
A local high-performance speech transcription WebUI based on transcribe.cpp + Moss-Transcribe-Diarize.
B resolvr-dns-dashboard A self-hosted monitoring panel for Technitium DNS with real-time query logs, cache and blocking stats, plus login and roles. Technitium users would want it, but the server's built-in interface already covers most of these features.
Self-hosted monitoring dashboard for Technitium DNS Server — live query logs, clients, cache, and blocking, with its own login and role-based access.
B seenix Draws the Nix store closure into the browser at one byte per pixel, with zoom and pan, to show visually what's eating disk space; the idea is fun and the implementation restrained. But only Nix users can use it, making it a playful diagnostic rather than a necessity.
Pan around a Nix closure in your browser, one byte per pixel
B xcode-project-format Claims to read and write Xcode's JSON project format, but Xcode projects are still pbxproj, so the description doesn't match reality, and being under the apple org makes source verification more necessary. Not recommended to rely on until clarified.
A Swift library for reading, writing, and manipulating Xcode's JSON-based project.xcproj format.
B qwen38-exl3-dflash2 Configuration notes for quantizing Qwen to 4bpw and running 262k context on a 24G 3090, with specific parameters that give reference values to people running local models on the same GPU; but it reads more like personal experiment notes and does not apply to other hardware.
Qwen3.8-27B EXL3 (4.00 bpw) + DFlash2 speculative decoding for ExLlamaV3, validated at 262k context on a 24 GB RTX 3090
B ComfyUI-FS_Audio_Suite Breaks YuE2 music generation into ComfyUI nodes for easy integration into existing audio workflows. Useful for ComfyUI music users, but node packs are already crowded and YuE2 itself is early, so its interface may change at any time.
FS_Audio Suite: modular YuE2 audio generation nodes for ComfyUI, by Make the Robot Do It
B frappe_config_pack A portable, versioned configuration package for the Frappe framework that addresses configuration drift during deployment, a pragmatic direction. The audience is limited to self-hosted Frappe teams, value is narrow, and full usage docs are still missing.
Portable, safe, versioned configuration packages for Frappe Framework.
B owasp-agentic-lab A weekend experiment designed around the OWASP Agentic Top 10: build an agent, then break it five ways and harden it. It has teaching value for agent security practitioners, but it's essentially course material with no reusable tool left behind.
Weekend lab for the OWASP Top 10 for Agentic Applications. Build an agent, break it five ways, then harden it.
B quietstart A control-recognition tool for HarmonyOS native pages that uses prompt learning to generate rules automatically, filling a gap in HarmonyOS automation; but the description is vague and lacks usage and accuracy data, making it hard to judge whether it works reliably.
A HarmonyOS native page control recognition assistant tool. Recognize and learn pages following prompts, and automatically generate rules.
B shanhui-ai-kefu An AI customer-service desktop app covering six platforms including Taobao and Douyin Shop, learning in-store product specs before answering and handing off to humans when unsure, a real need. But there's only a description, no code or deployment docs, and multi-platform adaptation costs raise doubts about maturity.
E-commerce AI customer service bot / AI customer service system: unified reception across six platforms and multiple stores — Qianniu (Taobao/Tmall), Xianyu, Douyin Store (Feige), JD (Jingmai), Kuaish…
B worky.nvim A git worktree switcher for Neovim, with a one-line description and clear boundaries. The problem is that worktree plugins and neogit already offer similar capabilities, and it covers only a tiny piece, so it is not worth installing separately unless you happen to lack this keybinding.
A git worktree switcher for neovim
B hotseat Shows quota usage for multiple Claude Code and Codex accounts in the terminal, then continues work interrupted by rate limits. The need is real and the scope restrained; but usage-panel tools have clustered recently, and 6 stars is not enough to prove it is more stable.
Terminal dashboard and CLI for Claude Code and Codex accounts, their quota, and work a usage limit stopped
B confset It uses Pkl to generate local config files for various tools, consolidating configuration into one verifiable source, with a small scope and restrained naming; but templates plus scripts already suffice, and it does not say which tools are supported, so the advantage is unclear.
Generate native tool configuration from Pkl
B glimmer A cross-platform Chinese input method written in Rust, explicitly stating it derives from Qingjian and follows GPL, which is honest; input method engines have a high barrier, and this is only an early derivative, far from replacing RIME day to day.
Weiming Glimmer: a cross-platform input method written in Rust, derived from Qingjian (GPL-3.0-or-later)
B uptime-pulse A lightweight self-hosted uptime monitor written purely in Rust, with a clear stack and low memory as its selling point. But Uptime Kuma and Gatus already cover this space thoroughly; unless you care deeply about resource use, there's no reason to migrate.
Ultra-lightweight self-hosted uptime monitoring system in pure Rust (Axum, Tokio, SQLite WAL, Rustls). Fast, low memory, and multi-arch ARM64/AMD64 ready.
B shiyi-ai-diet-assistant A live WeChat mini-program that uses multimodal recognition to log meals with nutritional guardrails, a more complete product idea than typical demos. But the diet-tracking space is crowded, and the repo doesn't clarify what's open-sourced or how to self-deploy.
Shiyi: a launched AI diet-logging WeChat Mini Program using multimodal recognition, Human-in-the-loop, and deterministic nutrition guardrails
B plexo Speeds up single-stream downloads via multi-connection parallelism, targeting slow large files and throttling, implemented in TypeScript for easy embedding; 56 stars in 0.6 days is decent growth. Monetization could be a download SDK or acceleration service, but mature free options like aria2 and axel are more stable, and server-side throttling weakens the payoff.
Speed up downloads by combining multiple network connections in parallel
B bg0 Runs local background removal in the browser with no upload and no usage limits, solving both privacy and API-cost pain and embeddable in any frontend; 84 stars in 1.3 days shows privacy demand is real. Monetization could be SDK licensing and a Pro batch-processing version, but native browser capabilities and imgly/remove.bg squeeze it, and model size slows first paint.
Private, unlimited background removal that runs in your browser.
B Show HN: Panel – A research workspace where the agent can build its own panes Lets agents generate interactive research panels themselves, upgrading output from chat bubbles to a workspace and hitting the pain that agent results are hard to consume; 47 upvotes show the interaction form resonates. Could become an agent frontend framework or hosted research platform, but the risk is each agent's built-in artifact features directly swallowing this scenario.
B Show HN: Ordewell – turn one goal into an ordered plan of coding-agent tasks Splits a goal into ordered coding-agent tasks, each with its own runner, model, and mode, then executes and verifies them, filling the planning and acceptance gap in multi-agent orchestration; 48 upvotes and 89 stars show real demand, but Conductor and Claude Code already occupy the space, making independent charging hard and absorption by IDEs or platforms more likely.
B video-generator-client Unifies video-generation APIs like Seedance, Kling, MiniMax, and Wan into an async Python client with CLI and local Web UI, solving multi-vendor switching and queueing costs; 340 stars in 1.3 days shows strong calling demand. But it is essentially a thin wrapper that official SDKs or ComfyUI plugins can replace anytime, so commercialization can only be an aggregated billing gateway.
Async Python wrapper for Seedance, Kling, MiniMax and Wan video generation. Supports CLI and a local web UI
B MCDMA Explores direct memory access between Metal and CUDA, trying to bypass copy overhead on unified-memory machines, meaningful for local training/inference hybrid stacks; 56 stars in 1.5 days shows interest in low-level GPU interop. It could become a cross-backend zero-copy library, but platform driver limits are strong and missing docs make validation and adoption hard.
Metal Cuda Direct Memory Access
B google-play-vitals-mcp Wraps Google Play's ANR, crash, and startup-time data into an MCP server so AI assistants can query Vitals directly, a concrete problem with a clear audience; the limits are needing developer credentials and the fast-growing crowd of similar MCP wrappers.
High-performance, token-efficient Model Context Protocol (MCP) server for Google Play Android Vitals - enables AI assistants to diagnose ANRs, crashes, and startup performance.
B vps-first-steps A bash script that hardens a fresh Ubuntu/Debian VPS into a usable state, aimed at people who just bought a machine and don't want to read ten tutorials. The scope is restrained and it has no dependencies, but hardening scripts like this are plentiful, and the only difference is doing it all in one command.
A single, readable bash script that takes a fresh Ubuntu/Debian VPS from default install to a hardened, ready-to-use server.
B pm5-force-logger Records Concept2 PM5 rower data over Bluetooth, with per-stroke power curves, a live dashboard, and Logbook upload. A ready tool for serious rowers in a space dominated by closed-source apps; limited to PM5 and maintained by one person.
Record a Concept2 PM5 over Bluetooth, force curves included, with a live dashboard and optional Concept2 Logbook upload
B dsh-computer-use_codex-style Recreates Codex-style desktop computer-use on DeepSeek Harness: window control, Chromium automation, visible overlays, and per-app permissions, MIT-licensed on Windows. Features are specific, but similar agents are crowded and maturity is unproven.
Codex-style Computer Use for DeepSeek Harness: the official window2 13-method desktop surface plus Chromium automation, a visible overlay with Esc interrupt, per-app approvals and bundled skills. Wind…
B turnloop An embeddable cross-platform event loop driver that unifies I/O into completion form across epoll, kqueue, IOCP, WASI, and the browser, a rare abstraction angle in the Rust async ecosystem; still too new, with no crates release or benchmark data seen.
Embeddable, cross-platform event-loop driver that the host turns: bounded turn(), completion-shaped I/O on epoll, kqueue, IOCP, WASI and the web.
B teatime-os A bare-metal OS for the ESP8266 without Espressif's binary blobs, a rare hard task in embedded work and valuable reading for anyone wanting to understand Wi-Fi chip internals. But it's only 0.7 days old, likely just serial blinking, far from real networking.
A bare-metal operating system for the ESP8266 with no vendor blobs
B doubao-say Doubao voice input for Wayland/Linux desktops, a category that's long been scarce on Linux. Clear use case, install and go; limited by dependence on the Doubao API and tested only on Wayland.
Doubao-powered voice input for Linux, Wayland and Omarchy
B ship-the-result A Claude Code hook that keeps temporary chat corrections out of commits, PRs, and test names, a real problem with a precise description. The limit is it serves one workflow, its effect depends on hook stability, and the alternative is adding constraints to your rules file yourself.
Claude Code skill + hook that keeps chat corrections out of commit messages, PR bodies, comments, test names and filenames
B hyprsplitter Turns Hyprland's tiled windows into a spaceship game arena, with each window as a level. It's pure fun but a novel idea, and one sentence makes clear what it is; practical value is near zero, the kind of project that earns a smile.
A spaceship game played with real Hyprland tiled windows
B claude-siri-ai Uses macOS 27 App Intents to delegate Siri requests to Claude Code, validating the new interface for system-level AI model replacement; 128 stars in 1.4 days shows developers are highly sensitive to Apple's AI openness. It has no independent business model and loses meaning once Apple ships native support, so it is only a signal for interface direction.
Experimental macOS 27 App Intents model delegation provider backed by Claude Code
B wormhole-display Turns an idle Meta Portal into a wireless AirPlay secondary display and speaker for Mac, iPhone, and iPad. A real use for people with obsolete hardware; only Portal owners benefit, and compatibility needs testing.
Give your Meta Portal a second life as a wireless display and speaker for Mac, iPhone and iPad (AirPlay-compatible receiver)
B protein_fighter Uses py2Dmol to animate two proteins fighting in real time, unfolding and coloring by AlphaFold pLDDT on hit. The author is a py3Dmol veteran and the fun and visualization are on point, but it's essentially a toy, not a tool to bookmark.
Street Fighter-style game with two humanoid proteins drawn live by py2Dmol; hits unfold them in AlphaFold pLDDT colours
B plumloom-autoeval-oss A CLI plus MCP server that evaluates AI apps and gates releases with result cards, with a restrained description that says what it does. But the space is crowded with promptfoo and langsmith, and it doesn't answer why it's easier than existing options.
Open-source CLI and MCP server for evaluating AI applications and gating releases with reliability evidence.
B cert-manager-scep-issuer Adds an SCEP external issuer to cert-manager so device certificates in MDM can be managed declaratively, a clearly missing small piece. Niche but immediately useful for teams doing device PKI; 8 stars show it's still early.
cert-manager SCEP External Issuer
B jellycanvas A plugin for theming Jellyfin 12, replacing hand-written CSS with live preview, so self-hosted media server users can tweak the web UI without hunting for selectors. Small scope and an honest description; the limits are serving only Jellyfin users, with custom CSS still the alternative, and a major Jellyfin upgrade may break it.
Theme designer plugin for Jellyfin 12 - customize the web UI without writing CSS, with a live preview
B turnrelay Tunnels UDP through WebRTC TURN relays disguised as call media, then routes it via sing-box, serving proxy needs on restricted networks. Novel and specific; limited by reliance on third-party relays like VK Calls, with stability and compliance out of your control.
Tunnel UDP through WebRTC TURN relays (VK Calls and others), disguised as call media, routed by sing-box
B shmup-deck Tap an arcade poster on your phone and MiSTer FPGA auto-loads the matching shooter, replacing the hassle of remembering ROM filenames. Small but complete; limited by its tie to MiSTer and a self-built poster wall, so the audience is narrow.
Flyer-wall shmup launcher for MiSTer FPGA: tap a flyer on your phone, the MiSTer loads the game
B rtx-5070ti-laptop-160w-power-limit For a specific Legion laptop with an RTX 5070 Ti, it studies raising the power limit from 140W to 160W, and the description honestly admits it's only research and a PoC; useful only to owners of this one machine, and changing power limits carries risk.
Research and proof-of-concept for raising the NVIDIA RTX 5070 Ti Laptop GPU power limit from 140W to 160W on Lenovo Legion Pro 5 Gen 10.
B r-doc An agent skill that has Claude Code/Codex continuously maintain AGENTS.md and the docs system, hitting the real pain of rotting agent-project docs. The idea is concrete, but similar skills abound and results depend on the host's diligence.
AI agent skill for maintaining project AGENTS.md and docs/ documentation systems.
B afmf-linux A Vulkan layer that brings AMD's AFMF frame generation to Linux, giving Radeon users a capability the driver lacks. Fills a gap; limited as experimental, possibly glitchy per game, and thinly documented.
Vulkan layer to enable AFMF (AMD Fluid Motion Frames) For linux
B robodiag-harness An LLM agent for diagnosing ROS 2 robots, emphasizing evidence chains, tool calls, and safety gates, far more concrete than generic AI ops assistants. Valuable reference for robot debugging teams; the barrier is needing a ROS 2 environment and matching log sources.
AI diagnostic agent for ROS 2 robots — evidence-based, tool-calling, safety-gated.
B omdrop-plugin Lets Omarchy (Linux) on Apple Silicon receive AirDrop from nearby Apple devices. The angle is specific and sidesteps the complexity of general AirDrop implementations; the limitation is that it covers only this one hardware and distro combination, so ordinary Linux users can't install it.
Receive files from nearby Apple devices over AirDrop, on Apple Silicon Macs running Omarchy
B modelnap A macOS menu bar utility that unloads idle Ollama models to give memory back to the system. The problem is specific and the approach restrained, and people running local models on a Mac can use it today; the alternative is setting keep_alive yourself, and its feature set is very narrow.
Let idle local LLMs sleep and get your Mac's memory back. A native macOS menu bar app to turn Ollama on and off and auto-free idle model RAM. Free and open source.
B NUEM Adds folding animations, hinge sounds, and lock-screen effects to a MacBook, mimicking the iPhone Duo, running smoothly at 120Hz with restrained scope. A charming little toy with no practical use beyond spectacle.
A fold animation for your MacBook, inspired by the iPhone Duo animation. With true customizable perspective, fold and snap style animations, hinge and snap sounds and vibrations, lock screen support, …
B AiGuard A local MITM gateway that intercepts only AI domains, redacting on the request side and streaming restoration on the response side, solving the problem of sending keys and business data straight to large models. Useful for those building enterprise AI access; but the specific rules for the "seven risk signals" aren't spelled out, and running your own proxy has a barrier.
AiGuard — a local-first AI traffic security gateway: a local MITM proxy intercepts only AI domains, redacts data on the request side, streams and restores responses, and detects seven categories of ri…
B VN_Sprite_Expression_Workflow Turns visual novel character sprite expression variants into a repeatable pipeline: local facial feature extraction, transparent PNGs, layered PSDs, plus pixel verification. Genuinely valuable for galgame artists producing sprites in bulk; the limit is being tied to fixed source images, so a change of art style or character means redoing the pipeline.
Visual novel sprite expression-difference workflow with fixed original art: local facial feature extraction, transparent PNGs, layered PSDs, and pixel verification. | Consistent visual novel sprite ex…
B fly-flappy Trains a fruit-fly brain connectome model to play Flappy Bird—a fun experiment combining neural simulation and a small game. Worth a look for computational neuroscience and bio-inspired agent fans; it's a toy, though, unlikely to yield reusable conclusions.
Train a fruit fly brain to play Flappy Bird.
B MuLaCover Controllable cover and remix generation using reference audio or MIDI, a direction already dominated by things like RVC. The description reads like a paper, with no inference cost or licensing notes, and "controllable" in music generation needs evidence.
MuLaCover: controllable AI cover-song and music-remix generation from reference audio or MIDI
B heimdall-pentest An autonomous agent framework for authorized penetration testing, and stressing "authorized" is honest; but Strix, PentestGPT, and others already hold this spot, and 39 stars in a day looks more like buzz, with no explanation of capability differences from existing frameworks.
An autonomous AI agent framework for authorized penetration testing
B uni-llm-bench A self-hosted lightweight LLM API benchmark platform focused on model fingerprint comparison plus quality evaluation, a specific direction. But similar evaluation tools are plentiful, the fingerprinting differentiator is unclear, and it's hard to see what installing it solves today.
A self-hosted lightweight LLM API benchmarking platform — model fingerprint comparison + model quality evaluation.
B UE2-C64U-Emulator Simulates the Ultimate 64 firmware environment so C64U firmware developers don't have to keep reflashing real hardware, plus an MCP server for agent control; a tiny audience, but a genuinely useful tool for retro computing and firmware work.
Emulator for Ultimate 64 Elite II / C64 Ultimate firmware (UI, services, TRX64 as C64) — develop and test firmware, including a mcp server
B wortkiste Zero-config German A1–B1 vocabulary training with the Goethe word list and Leitner spaced repetition, usable right in the browser. Cleanly built, but vocabulary apps are plentiful and this offers no obvious new approach.
Free German A1–B1 vocab trainer: Goethe word list built in, Leitner spaced repetition, zero setup.
B kalshi-trading-bot A Kalshi prediction-market bot with order book browsing, market making, paper trading, and risk limits, honestly labeled as educational. But forty stars in seven hours is fast, and similar templates abound; not recommending yet.
Kalshi trading bot for prediction markets — browse event contracts and watch the order book in real time, run market making and momentum/reversion strategies with paper trading, risk limits and positi…
B cloudkeep A Three.js sky-exploration creature-catching game with Blender models and procedural asset scripts, worth reading for anyone wanting the full web 3D game workflow. But it's still a personal game demo; playability can't be judged from the description, and it doesn't explain how it differs from existing Three.js templates.
A handcrafted 3D sky exploration and creature-capture game. Original prompt, complete Three.js source, Blender models, and procedural asset scripts.
B Fieldwatch A receive-only Wi-Fi and BLE observer on Android, focused on passive listening without active probing, a selling point for the privacy-minded; but Wi-Fi analyzer apps are plentiful, and the new repo shows no UI or limitation notes yet.
Receive-only Wi-Fi and Bluetooth LE observer for Android. MIT.
B spyde Declarative layout on top of PDFKit, aimed at the pain of its hard-to-maintain imperative API, useful for Node report generation. But react-pdf and pdfmake already own this space, the repo is very new, and there are no examples showing how declarative it really is.
Declarative PDF layout for Node, on top of PDFKit
B PocketOrca-LLM A local LLM server for Qualcomm chips, targeting the gap in on-device NPU inference; but the description is one sentence with no supported devices or quantization scheme, and llama.cpp's QNN backend is already doing the same thing.
A local LLM server built for Qualcomm chips
B wechat-group-report Reads local WeChat records on Windows to summarize group chats, outputting PNG/HTML/Markdown with sources. Directly usable for people who need to organize group chats, but export tools are plentiful, and it depends on WeChat's local database format, which breaks easily when the client updates.
WeChat group summary report Skill: reads local WeChat records on Windows and generates sourced PNG, HTML, and Markdown summaries.
B transductARC Uses a transduction-style autoregressive Transformer for ARC-AGI 1&2, a legitimate direction rather than a wrapper. But the repo only has training code, with no results, evaluation, or reproduction steps, so outsiders can't judge whether it scores.
Training transductive autoregressive transformers for ARC-AGI 1 & 2
B SpaceRangersHD_decomp A matching-style Delphi decompilation of Space Rangers HD—genuinely impressive work. But such projects are legally gray and serve only tiny modding and porting needs; not for broad recommendation.
A matching Delphi decompilation of Space Rangers HD: A War Apart - Космические рейнджеры HD: Революция
B AnvilCSG Brush-based CSG level design in Unity, recreating a Hammer-style workflow, supporting Built-in and URP. Handy for level designers, but Unity already has mature CSG options like ProBuilder and SabreCSG, and it calls itself a legacy preview, looking more like a shelved experiment.
Anvil CSG Beta 2: a free legacy preview of brush-based level design inside Unity. Hammer-inspired workflow, Built-in and URP.
B petex A desktop pet for macOS and Windows that supports importing Codex pet assets. It's lightweight, but desktop pets already have plenty of options like Shimeji, and the only difference is asset importing, so the novelty is limited.
Standalone macOS and Windows desktop pets with Codex pet imports.
B SmuggleMyPayload Turns HTML Smuggling into a templated generator for red-team payload testing; the technique isn't new and similar tools abound, while the README lists features without stating boundaries. As a reusable malicious delivery tool, risk outweighs value.
SmuggleMyPayload generates HTML pages that embed and reconstruct files client-side using JavaScript. The toolkit offers multiple HTML Smuggling techniques, payload encoding options, JavaScript variati…
B selflift-Avatar A SelfLift variant for ComfyUI that handles MiniMax audio-video masking while preserving the audio track. The problem is specific but the audience very narrow, and beyond one line there's no usage or effect comparison.
Experimental ComfyUI SelfLift variant for MiniMax H3 video/audio masks and audio preservation.
B beckon Voice control for Omarchy that executes commands as you speak via the Gemini Live API. An interesting idea, but tightly bound to one distro and one cloud API, so ordinary Linux desktops can't use it; voice assistants are plentiful, and it shows no advantage in convenience.
Voice control for Omarchy, powered by the Gemini Live API. Talk naturally; it acts while you're still speaking.
B kinesis Lets the Meta Neural Band read sEMG independently of the glasses and control macOS with finger gestures, solving the hardware being locked to the glasses scenario. 52 stars in 0.5 days plus 107 HN upvotes show developer appetite for new input devices; it could become the standard Mac driver layer for neural wristbands, but depends entirely on Meta's undocumented protocol and hardware stock.
native macos controls for the meta neural band
B Show HN: Kinesis – Control your Mac with the Meta Neural Band Lets the Meta Neural Band read sEMG independently of the glasses and control macOS with finger gestures, solving the hardware being locked to the glasses scenario. 107 upvotes and 52 stars in 0.5 days show developers are hungry for new input devices; it could become the standard Mac driver layer for neural wristbands, but depends on undocumented protocols and Meta hardware stock.
a couple hours ago i had astra take on a 10 month old repo of mine https://github.com/callbacked/neural-band-poc/ that involved finding a way to use the meta neural band independently from the glasses…
B Show HN: Neobrutalism.dev – Just added Base UI support and added new color theme It offers neobrutalist UI components and color themes so developers can build distinctive interfaces without hand-writing styles. 132 upvotes come from continuous iteration and Base UI adaptation, and it has built design-system brand recognition. Pro templates could be the monetization, with signs of becoming a standard frontend dependency; the risk is shadcn/Tailwind ecosystems building in similar styles and design trends cooling.
B gap-trap Presets rules and quality gates in the repo so AI-written code is automatically constrained before commit, addressing quality decay from unreviewed vibe coding. The author is ZoneMinder's creator, 39 stars in 1.1 days; it could become a CI-side code quality subscription layer, but Cursor and Claude Code are gradually building equivalent hooks and linting.
Turns vibe coding into high quality code. Sets up rules and gates in your repo so AI-written code stays correct without you reviewing every line.
B PhishDrill A zero-third-party-dependency phishing simulation platform that runs sending, tracking, de-faking, attribution and reporting from one intranet machine, plus an admin console and dashboard. Lighter than GoPhish-style multi-component setups and suited to internal security teams doing compliance drills; email deliverability and template ecosystem are on you.
PhishDrill is an open-source phishing drill platform with zero third-party dependencies: no C2 to set up and no commercial platform to buy — a single intranet machine (even a laptop) is enough to run …
B FlashREINFORCE Proposes critic-free, single-rollout asynchronous reinforcement learning for agentic language model training, aiming to cut both RL training cost and engineering complexity. 40 stars in 0.8 days show training-side sensitivity to new algorithms; if absorbed by mainstream RL frameworks it becomes a standard component, but GRPO variants are numerous and third-party reproduction is lacking.
FlashREINFORCE: Critic-Free, Single-Rollout, Asynchronous RL for Agentic Language Models
B ComfyUI-YuE2-Trainer A ComfyUI plugin for LoRA training on the open-source music model YuE that wraps the otherwise convoluted workflow into visual nodes. It is a directly usable small patch for existing ComfyUI music users; the limits are that it only serves the YuE ecosystem and documentation is still thin.
LoRa Trainer For YuE in ComfyUI
B jellyfin-plugin-update-notifier A Jellyfin plugin that reminds admins to restart after an update and lists what changed in the panel. The problem is specific and the implementation should be simple, a handy little patch for self-hosted Jellyfin users; limited value but honestly described.
A Jellyfin plugin that notifies admins when plugins have been updated and the server needs a restart, with a dashboard listing what changed.
B amt-tools A small toolkit for configuring and using Intel AMT: mutual TLS, power on/off, SOL serial, and KVM. vPro out-of-band management has always been a documentation nightmare, so scripting this workflow is genuinely useful for ops; the limit is covering only AMT machines, a narrow but precise audience.
Configure and use Intel AMT: mutual TLS, power control, SOL, and KVM
B omatune A terminal guitar tuner for the Omarchy desktop, supporting six- and twelve-string guitars, following theme colors and working with a microphone. Restrained scope and honest description make it a handy install for Arch users; it only serves this niche desktop and overlaps with tools like fmit.
A terminal guitar tuner for Omarchy — 6- and 12-string, theme-aware
B nv3d-vision-restoration Keeps discontinued NVIDIA 3D Vision glasses working on new Windows via the original USB emitter, bypassing the removed driver. Old-hardware-preservation reverse engineering with a small audience, but specific and honestly documented, worth a look.
NVIDIA 3D Vision glasses on modern Windows PCs through the original USB emitter, without the discontinued 3D Vision driver.
B fly-brain-atlas An interactive male fruit fly brain atlas with region and neuron reconstructions, aimed at neuroscience research and teaching. Compared with static atlases, in-browser 3D browsing is a real increment; the limit is that the description doesn't cover data sources or coordinate systems, so users must verify themselves.
Interactive MaleCNS fly brain atlas with native regional and neuron reconstructions
B rich-input A rich input box built from datalist, OpaqueRange, and the Custom Highlight API, offering keyword completion plus in-input highlighting. For frontend developers it's a fresh browser-native demo, far lighter than pulling in a heavy editor; the limit is that it's more a technical sample than an installable component.
A rich input field with keyword-based autocomplete and in-input highlighting powered by <datalist>, OpaqueRange, and the Custom Highlight API
B DevSweep A macOS menu bar tool that one-click clears build caches and stuck processes left by Flutter, iOS, Android, Node, and Go, with tech stacks defined in YAML. It addresses real, frequent disk anxiety and differs from DevCleaner in coverage—install and use.
macOS menu bar app that finds dev processes stuck in limbo and cleans the caches, builds and simulator data that Flutter, iOS, Android, Node and Go leave behind — one click, YAML-defined stacks
B mew A daemonless terminal git workspace panel written in Rust, focused on fast startup. Clear description, but the space is already filled by lazygit and gitui, with no differentiating design visible and 23 stars saying little.
fast, zero-daemon terminal companion and git workspace dashboard
B WeChat-H5-DevTools A debugging and reverse-engineering toolbox for WeChat's built-in browser and official-account H5 pages, using Frida hooks and a vConsole proxy, genuinely filling a painful gap. But it depends on the WMPFDebugger ecosystem, the description uses words like "full power," and maturity and compliance boundaries are unclear.
WMPFDebugger built-in browser and official account H5 full-featured debugging and reverse-engineering toolbox / The ultimate debugging & reverse-engineering toolkit for WMPFDebugger In-App Browser & H…
B FrameGen-Manager A graphical manager that one-click deploys DLSS frame generation (actually swapping in FSR3 frame generation) on RTX 20/30 series cards, scanning game libraries and auto-updating. It saves old-card users manual patching, with value mainly in integration and maintenance, while the core solution still comes from community mods.
One-click DLSS Frame Generation deployer for RTX 20 and 30 series GPUs, with game library scanning and auto-update. One-click DLSS Frame Generation deployer for RTX 20/30 series.
B inklet A hand-drawn-style Compose Multiplatform component library covering Android, iOS, desktop and WebAssembly, with an online gallery. Attractive to KMP teams wanting a non-standard look; the small component count and keeping up with Compose releases are the key limits.
Hand-drawn Compose Multiplatform components for Android, iOS, desktop and WebAssembly. Live gallery: https://blog.ggoggam.dev/inklet/
B GlassFan A SwiftUI press-and-drag ring menu using iOS 26's Liquid Glass material. The component is clean, but ring menus are already common, so treat it as a sample of the new system API.
Press-drag-release radial menu for SwiftUI, built on iOS 26 Liquid Glass
B codex-cc-orchestrator An orchestrator chaining Codex and Claude Code into layered collaboration, with task dispatch, session resumption, and independent review as concrete features. But multi-agent orchestrators have proliferated this half-year, the description is conceptual, and it's unclear what beats existing options.
Dianjiangtai: a portable Agent hierarchical collaboration orchestrator. Supports task dispatch, session resumption, independent review, background execution and usage logging. Currently works out of t…
B repro-lens Runs static reproducibility checks on ML code and coding agents, scanning without ML dependencies and offering experiment replay with before/after evidence. It lands right on the gap of proving an agent's edits didn't break things. Currently only prose, missing a real output example, and still small.
Static reproducibility checks, experiment replay and before/after evidence for ML code and coding agents — no ML dependencies for scans.
B hermes-my-skills A skill list, export, and safe-merge utility for Hermes Agent users, with clear scope and restrained wording. But it's fully tied to a niche agent platform, so it has no migration value if the platform doesn't take off, and it's little different from generic config sync.
Inventory, export, and safely merge learned skills across Hermes Agent profiles
B trace_search Turns regex search, line reading, and export over hundreds of GB of trace logs into an MCP service so agents can dig through big logs, hitting a real troubleshooting pain point. Only a Chinese intro is given, with no indexing strategy or memory figures for large files; benchmark it yourself before adopting.
An efficient MCP service for trace logs at the hundreds-of-GB scale, offering regex search, non-regex search, reading specified lines, exporting specified lines, and more.
B XiaoAi-LLM-Router Turns a dusty Xiaoai speaker into a voice assistant connected to DeepSeek or Ollama, claiming zero hardware modification—a practical reuse of old hardware. It requires hands-on skill and understanding of Xiaoai firmware, and the description omits failure modes and rollback, so be prepared before starting.
Upgrade an old XiaoAi speaker into a DeepSeek/Ollama smart assistant with one click, zero hardware mods. | Upgrade your XiaoAi speaker to a DeepSeek/Ollama AI assistant with one click. Zero hardware m…
B claude-code-routing Replicates Spotify's Claude Code model routing as a pure configuration approach, letting cheap models read files and write boilerplate, with tests on four tasks. The idea is practical, but the repo is HTML-written explanation plus conventions, so whether it applies depends on your project shape; it's closer to a blog post than a tool.
Spotify's 90% Claude Code routing setup, rebuilt for plain Claude Code and measured on 4 real tasks. A cheap model does the reading and the boilerplate.
B PapaSangre-Windows A faithful Windows port of the audio-only horror game Papa Sangre, preserving the original HRTF and supporting keyboard, gamepad, and screen readers. It's both game preservation and a rare accessibility-friendly case; the audience is narrow, but the scope and description are honest.
A faithful Windows port of Papa Sangre, the audio-only horror game by Somethin' Else. Keyboard, controller, screen reader, and the original's own HRTF.
B Sisu_DCP-3 Firmware written from scratch for a Sisu DCP-3 replacement motherboard, aiming to recreate the Nokia 3210 experience on modern hardware. Hardware replica plus firmware rewrite is genuinely rare, but only the few who own that board can actually use it, making it just a curiosity for most readers.
From-scratch firmware for the Sisu DCP-3 replacement mainboard, recreating the Nokia 3210 NSE-8 v6.00 experience on modern hardware.
B SOVA Official implementation of a GRPO calibration method for video reasoning, claiming to improve correctness without extra rollouts. Of reference value to researchers doing RL fine-tuning, but only title-level information is available and reproduction cost is unknown, so watch for now.
Official implementation of SOVA: Strict Outcome-Conditioned Virtual Advantages for Video Reasoning. Correctness-oriented calibration for TW-GRPO without extra rollouts.
B knowclip Automatically cuts highlight clips from livestreams, courses, and long talking-head videos, emphasizing local speech recognition so audio never leaves the machine—appealing to content editors. But tools like Opus Clip and open-source clippers already fill the space, and localization must be proven by actual clip quality.
Automatically cuts long videos such as livestreams, courses, talks and talking-head videos into highlight clips | local speech recognition, audio never leaves the machine; LLM picks the highlights, ba…
B uestc-thesis A LaTeX thesis template for University of Electronic Science and Technology of China, supporting Chinese and English for master's and doctoral students. It does save students typesetting hassle, but university thesis templates are everywhere and the project adds no technical increment, so non-students have no reason to look.
LaTeX template for University of Electronic Science and Technology of China degree theses, supporting Chinese and English, master's and doctoral theses.
B Sisu_SSE-9 A six-layer LTE replacement motherboard for the Nokia 3210, based on the RP2354B and Telit modules. It's an imaginative hardware-modding experiment, but the description stops at the board-level design with no progress or reproducibility, and the audience is only a handful of retro hardware enthusiasts.
Experimental six-layer LTE replacement mainboard for the Nokia 3210, built around RP2354B and Telit LE910C1-WWX.
B loomark A self-hosted bookmark manager focused on browser sync. The problem is that this slot already has mature projects like linkding and Hoarder, and the description doesn't explain the sync mechanism or differentiation, so yet another bookmark tool is hard to motivate migration for.
A self-hosted bookmark manager that syncs with your browser.
B mhi2-android-auto-video-vc An experiment hooking the factory GAL on Volkswagen MIB2.5 head units to achieve dual-screen Android Auto casting. It's a useful lead for car owners who tinker with head units, and the description honestly notes the tested model and personal use; but coverage is extremely narrow and firmware updates can break it anytime.
Experimental dual-screen Android Auto projection via MHI2 stock GAL hooking. For personal use; tested on VW MIB2.5 High (EU MU1367).
B topnodes A set of ComfyUI person-replacement nodes with SAM mask inpainting and full-video timing, aimed at AI face-swap workflows. Such node packs are already crowded, with no new method visible, just another workflow wrapper.
ComfyUI nodes and H3 workflows for single and dual character replacement, SAM mask repair, and full-clip timing
B daidocs Argues for storing AI long-term memory as plain-text .dai files on disk, with an MCP service and editor hooks—the direction is right. But the claimed LongMemEval scores include a nonexistent model name like 'Claude Fable 5,' undermining benchmark credibility, and several similar memory solutions already exist.
Open plain-text file format for AI memory. Your assistant's long-term memory as .dai files on your disk: readable by Claude, GPT, Gemini, Cursor, local models and grep (all LLM models work). MCP serve…
B BiliClean A Bilibili iOS client plugin that skips sponsors, filters ads, downloads videos, and tidies the interface. The feature list is clear and useful for jailbroken users, but it relies on private APIs and client modifications, easily breaking with version updates, and bypassing sponsors and ads raises compliance questions.
Bilibili iOS plugin: sponsor skip, ad filtering, full video download, CDN acceleration and UI cleanup
B asset-studio It chains Qwen-Image, Pixal3D and Blender into a fully local text-to-3D asset pipeline exposed to agents via MCP, CLI and FastAPI, directly addressing indie game art capacity and licensing; fully local means no copyright worries. Studio subscriptions or cloud hosting could monetize it, with a chance to become the standard asset-generation interface; the risk is fast upstream model iteration, no model moat, and reliance on high-end GPUs.
Fully local, agent-callable text-to-3D game asset pipeline: Qwen-Image-2512 -> Pixal3D (TRELLIS.2) -> Blender/meshoptimizer, FastAPI + CLI + MCP, Docker Desktop, RTX 5090
B docket Records for every AI-written commit what was tried, who verified it, and what went unreviewed, turning code provenance into checkable evidence. It hits a real problem for teams already using coding agents; the project is still small, so whether it can plug into existing CI remains to be seen.
Per-commit evidence records for agent-written code. Captures what the agent tried, what verified it, and what nobody reviewed
B ToneMappingAnalysis A tonemapping LUT analysis tool from runevision for people doing Unity rendering and color pipelines to quantitatively compare tonemap curves. A small, specific angle from an author with a reputation in rendering; the limitation is it's a pure utility with narrow use and sparse docs.
Tools for analysing tone mapping LUTs.
B adhan-mute A small CLI that automatically mutes system audio at each prayer time based on location, across macOS, Windows, and Linux. It addresses a real pain point for Muslim users who play audio while working; the limitation is its single function, and the restore-after-mute strategy is unclear.
CLI that mutes system audio (Spotify, YouTube, etc.) at each adhan time, based on your location. Cross-platform: macOS, Windows, Linux.
B Show HN: Makefaster.dev It runs an autoresearch loop for about $10k, distilling Lighthouse optimization lessons from the top 200 frontend repos into a leaderboard, then packages it as an iterative wrapper that takes over claude/cursor subscriptions, addressing the lack of checklists and execution in performance work; CI performance gates or team plans could monetize it. The risk is platforms building it in, or a single prompt replicating it.
I had a bunch of extra Fable credits, so I spent around $10k in api costs doing autoresearch loops on the top 200 github repos with frontends. The goal was to speed up the frontend / improve the light…
B adexview It browses and audits AD Explorer snapshots offline, filling in DNS, subnets, DFS, GPO links, LAPS and AD CS information BloodHound cannot see, a must-have for internal pentesting and domain admin troubleshooting. It spreads via security community word of mouth and pentest reports, with enterprise editions or audit services as monetization; the risk is the huge BloodHound and PingCastle ecosystems swallowing a point tool.
Browse, search and audit AD Explorer snapshots offline in your browser: decoded attributes, LDAP filters, and reports for what BloodHound doesn't show (DNS, subnets, DFS, GPO links, LAPS, AD CS).
B skynet-godot Ports two 1996 Bethesda XnGine shooters to Godot 4, explicitly requiring players to supply the original data, with a clearly stated scope. Useful for old-game preservation and Godot porting learners; the limitation is it's research-oriented and hard for ordinary players to pick up.
A Godot 4 port of Bethesda's 1996 XnGine shooters SkyNET and Future Shock - plays your own copy of the original game data.
B status-trio Combines Wi-Fi, battery, and volume into a single macOS menu bar icon, inspired by the iPhone's merged status bar. Small but complete with clear docs; unlike all-in-one monitors like Stats, doing one thing makes it lighter.
Status Trio — a native macOS menubar app that combines Wi-Fi, battery, and volume into one compact, configurable menu bar icon, inspired by the iPhone Duo combined status bar.
B omarchy-hotbar A Quickshell top-bar plugin for Omarchy/Hyprland, with fixed app slots, a file location, and a controlled run-window drawer. The scope is restrained and the description accurate, so Omarchy users can just install it; the limitation is that it only serves this one desktop setup.
Omarchy 4 bar plugin for Hyprland/Quickshell: fixed app slots, filesystem Places, one bounded Running drawer
B SoftCorr For the specific problem of targetless single-frame LiDAR-camera calibration, it provides the official paper implementation of differentiable local correlation, directly useful to researchers doing multi-sensor calibration for autonomous driving and robotics. The scope is clear and not overstated, but this is paper code without easy interfaces, data prep or run instructions; 7 stars and half a day old, its practical usability is unverified.
Official Implementation of SoftCorr: Differentiable Local Correlation for Single-Frame Targetless LiDAR-Camera Calibration
B sunshine-vdisplay Creates a virtual display for Sunshine and Moonlight on NVIDIA plus KDE Wayland, filling the common streaming black-screen gap. The approach is clear and scripted, but it may not apply beyond this driver combination.
NVIDIA + KDE Wayland virtual display for Sunshine/Moonlight streaming on Linux
B lookhome A long-term rental management system for small landlords, covering billing, meter reading, and deposits, self-hostable with hooks for a mini program and payments. The features are concrete, but CRUD platforms like this are common; the key is whether anyone actually uses it.
Lukjia - lightweight apartment management system (long-term rental management/billing/meter reading/deposits, WeChat Mini Program and payment reserved, supports private deployment)
B agent-skill-sync Scans SKILL.md files across toolchains and syncs them by category, solving the real hassle of keeping separate skill directories for Codex, Claude Code, and WorkBuddy. A zero-dependency CLI with a clear scope, though few people run multiple agents at once.
Scan, classify (A/B/C/D) and sync AI-agent SKILL.md files across toolchains (Codex, Claude Code, WorkBuddy). Zero-dependency Python CLI.
B linux-parental-controls A systemd timer that kicks the kid's session offline at a set time; the author plainly calls it "just a simple unit" without overclaiming. Enough for parents, but it also shows how far it is from real parental controls like ratings and whitelists.
Just a simple systemd unit to log off the kids at 22:00 until 06:30
B GlassProjection A fold/unfold glass animation for Xiaomi foldables built on Shizuku, working on both home and lock screens without root. It solves one small visual problem for a specific device with a restrained scope; the limitation is its tie to MIUI and Shizuku, so other devices cannot use it.
Glass Projection: real-time glass folding animation for Xiaomi foldables, supports home screen and lock screen, adjustable blur and screen-switch angle, based on Shizuku, no root required.
B MnemoBrain A dual-engine memory stack for AI agents, installed with one command and including self-checks. The idea is clear, but the memory layer is crowded with mature options like mem0 and Zep, and how the two engines divide work isn't explained.
Two-engine memory stack for AI agents: Mnemosyne (reflex memory) + GBrain (knowledge brain). Installer, launcher, doctor — one command, any host.
B agent-skill-remotion-motion-graphics A Claude skill that uses Remotion to make kinetic typography and brand films, outputting MP4 directly. It's a reusable starting point for programmatic motion designers, but it essentially wraps existing Remotion usage as a skill, and similar skill repos have proliferated lately.
Claude agent skill: build kinetic-typography / logo / brand promo motion graphics as code with Remotion (React) and ship an MP4 — 9:16 and 16:9 from one codebase
B melee-unlocked Unlocks the frame rate in Super Smash Bros. Melee as a native Windows version with Slippi online support, requiring your own ISO. Technically nontrivial and honestly labeled ALPHA, but the audience is limited to the Melee scene and it's legally gray.
Super Smash Bros. Melee with unlocked FPS as a native Windows build with Slippi online play. Current release is in ALPHA. Bring your own NTSC 1.02 ISO.
B nova-display-calibration SDR screen calibration for the Retroid Pocket Nova, with native config persistence and no background process, a very specific goal. Practical for owners of these handhelds, but the audience is narrow and it lacks measurement methods and comparison data; note it but don't rush to look.
Measured SDR display calibration for Retroid Pocket Nova, with native profile persistence and no background service.
B DengShell A Chinese-language SSH desktop workbench written in Go, bundling multi-session terminal, SFTP, and monitoring for Chinese ops users out of the box. But FinalShell and WindTerm already fill this niche, and long-term maintenance is the big question.
A Chinese SSH desktop workbench integrating multi-session terminals, SFTP file management, server monitoring, command history and personalized settings.
B luci-app-opengfw A graphical panel for openGFW on OpenWrt and iStoreOS, letting DPI detection and blocking be configured on a router with mouse clicks; convenient for those who need it, but it's just a shell over an existing core, and such capabilities are inherently sensitive.
OpenWrt & iStoreOS openGFW visual control panel, supporting deep feature identification and blocking of Shadowsocks/VMess/Trojan
B CC-Monitor Records and audits every action Claude Code executes on the machine, adding observability for those who let agents work freely; the need is real, but the one-line description doesn't say whether it blocks or just logs afterwards, nor how it differs from existing hooks approaches.
Claude Code Monitor : Monitor & audit every action Claude Code takes on your computer.
B mlx-AuK Ports a 1.5B speech foundation model to Apple Silicon MLX, covering TTS, voice editing, enhancement, and separation. It has practical value for developers running audio models locally on Mac; but the description piles on jargon, never names the model, and lacks examples and performance data.
AuK: Native Apple Silicon MLX Port for 1.5B Speech Foundation Model (TTS, Vocal/Content Edit, Enhancement & Separation)
B mindmeld-with-fly Connects a Muse EEG headband to a fruit-fly central nervous system simulation for a real-time "mind meld," which is genuinely unusual. But it's closer to a science-art installation with little practical value, and the reproduction requirements aren't explained at all.
Real-time Muse EEG driving one Drosophila MaleCNS reservoir (mind meld)
B Home-assistant-barcode-Scanner Uses an ESP32 plus a barcode scanner module to add products directly to a Home Assistant shopping list, a small but concrete scenario. It's a nice weekend project for HA tinkerers, but it's more of a hardware wiring guide with limited reuse value.
An ESP32 board with a barcode scanner for adding items to your home assistant shopping list
B Hangly A small physics-simulated widget that sits in the menu bar, a macOS toy written in Swift. Cleanly scoped and honestly described, worth a glance, but it's purely decorative and there's nothing more after that.
A tiny piece of motion for your desktop. A physics-simulated charm hanging from your menu bar.
B FlyBrain-HalfLife Drives DirectInput to play Half-Life using the FlyWire fruit-fly connectome plus LIF neurons, with very specific parameters; an amusing technical stunt, but not reusable elsewhere and more an experiment log than a tool, so ordinary developers can just watch.
Autonomous Drosophila melanogaster (MaleCNS v1.0 / FlyWire) connectome agent playing Half-Life via biophysical LIF neural dynamics, 60x60 retina transduction, and DirectInput.
B SynthIPTV A server plugin that adds IPTV live sources to Emby/Jellyfin, with a clear need and install-and-go setup. But similar plugins already exist, so innovation is limited; the main question is whether it can keep up with interface changes in both upstreams.
IPTV for your Emby/Jellyfin server
B CoTFly Lets a helmet-wearing fruit fly play tank battles in the browser using a real spiking neural network. The idea is rare and the implementation concrete, but it's purely a fun demo with nothing usable after the amazement.
A fruit fly in a tiny tank helmet plays Claude of Tanks. Live spiking neurons, a miniature 3D cockpit, and autonomous battles with 171 vehicles and 30 maps. Built with Three.js and TypeScript. Play in…
B bundlebleed Connects LLMs to JS asset reconnaissance and initial vulnerability triage, emphasizing authorization scope and passive-by-default behavior with restrained wording. But AI-assisted recon tools have clustered in the past six months, the repo is very new, and there are no cases proving effectiveness.
AI/LLM-assisted JS Recon and vulnerability triage for authorized bug bounty testing & pentesting — scope-gated, passive-by-default, human-verified
B pr-constellation Draws the blast radius of a PR change, hitting a blind spot in code review. But there's only a one-sentence concept description with no screenshots, install instructions, or platform notes, so it isn't yet a real project.
See how one change in a PR flows into the rest. Skip the noise.
B fadescii Adds a time dimension of color, glow and fade to ASCII characters in the terminal, positioned as a TUI rendering library, a rare angle; but the audience is narrow, the pitch leans on "expressiveness", and it remains to be seen whether anyone builds interfaces with it.
Fadescii — A native terminal text-rendering framework that brings temporal color, glow, and fade effects to ASCII characters. Built for smooth, dynamic, and expressive terminal visuals.
B birdview It scans and draws architecture diagrams before AI changes code, mitigating architecture erosion from agents' blind edits. 79 stars in 0.7 days reflect rising demand for agent context governance. It could be an IDE or MCP plugin with team subscriptions, or merge into code quality platforms; the risk is overlap with symbol-graph tools like Serena and graphify, and diagrams becoming one-off artifacts.
Stop letting AI code blind. Map the architecture before every change with Birdview.
B stampede It read-only browses wallet rotations following PONS v2 rules on Robinhood Chain, replaying samples without an API key to help on-chain players track fund movements. A new chain plus airdrop expectations brought 151 stars in 1.3 days. Subscription on-chain intelligence or alerts could monetize it; the risk is mature platforms like Nansen and Arkham covering it anytime, and new-chain hype fading.
Read-only explorer of observed wallet rotations on Robinhood Chain (PONS v2): terminal + web, replay of a recorded sample without API keys
B townies A multiplayer cooperative small-town life game built with Three.js and Cloudflare Durable Objects; the tech choices are clearly described and it should be playable directly in the browser. It is a useful reference for those learning edge room-state sync; the limitation is that the gameplay is still small and it remains essentially a demo.
A cooperative multiplayer town life game built with Three.js and Cloudflare Durable Objects.
B reprieve A regret pill for Omarchy users who hit Super+W by accident: the window is only hidden and comes back on a second press. Tiny scope, honest description, no extra promises, works once installed. Limited to Hyprland-based desktops, pointless elsewhere.
A safety net for Super+W on Omarchy: the window is hidden, not terminated, and one keystroke returns it exactly as it was.
B adexsnap Exports AD snapshots to AD Explorer .dat files on Linux/macOS, directly consumable by ADExplorerSnapshot.py and BOFHound. Immediately useful for internal network assessments, and matching mature tools' output is a plus; narrow scope, since Windows already does this.
Active Directory snapshots from Linux and macOS in the AD Explorer .dat format. Opens in AD Explorer, works with ADExplorerSnapshot.py and BOFHound.
B margin A writing review tool that writes annotations back into source files, supporting Markdown, Typst, and LaTeX, suited to revising papers and technical drafts with collaborators. It inverts the document-comment approach and is worth a look, but the description is brief and collaboration needs hands-on testing.
Source-first, comment-only writing review for Markdown, Typst, and LaTeX
B invoice-organizer-skill Organizes Chinese e-invoices locally into reimbursable archives: PDF field extraction, 23 expense categories, review, renaming and deduplication, as a Claude skill plus Python scripts. Reimbursement is genuinely painful and the boundaries are clear; it only covers Chinese invoice formats, so other types need rule changes.
Chinese e-invoice local organizer Skill: 23 expense categories guidance, PDF field extraction, classification review, renaming and deduplication.
B Tourmaline Reverse-engineering notes, IOCs, and deobfuscated source for a ClickFix Python RAT calling itself Tourmaline, aimed at incident response and threat intelligence people. The material is concrete and unhyped, a one-off but useful analysis artifact with no general tooling value.
Reverse engineering notes, IOCs, and deobfuscated source for a ClickFix Python RAT that calls itself Tourmaline.
B docstudio A native macOS widget workbench where you drag to preview Dock widgets and run them locally in real time, targeting the pain of fighting WidgetKit when building Mac utilities. The idea is concrete, but the one-line description has yet to prove it can bypass system limits; wait for a demo.
Native macOS widget studio with a drag-and-drop Dock preview and live local widgets.
B prism A differential fuzzer that uses parsing discrepancies across 54 HTTP servers to find request-smuggling primitives. Protocol security researchers will want to run it, and the approach is more solid than single-payload tools; results still need manual interpretation, and coverage decides success.
Differential fuzzer that finds HTTP parser discrepancies across 54 servers to uncover request-smuggling primitives.
B Flock-Noir An ESP32-S3 Sense camera with the IR filter removed, used at night to detect Flock license-plate cameras, log GPS, and save data to card in wigle format. The motivation and hardware requirements are clearly stated, suiting those concerned about street surveillance, though completeness remains unverified.
A IR Flock Detector & Wardriver with the Seeed Studio ESP32-S3 Sense. Using the OV3660 with no IR lens allows you to detect flock cameras at night through IR that will use the gps to log the location …
B snes9x-Z A performance fork of Snes9x 1.63 with dual-core PPU rendering, claiming bit-exact parity with upstream and a 61% average frame-time improvement, with the upstream commit noted. Directly valuable to emulator porters and low-end device players; bit-exactness on timing-sensitive games still needs independent verification.
Performance-optimized fork of Snes9x v1.63 (upstream commit 7a8878f1): bit-exact against upstream, dual-core PPU rendering, +61% mean frame-time gain.
B biquad-voxengo Biquad bell and bandpass filter design code released by the Voxengo author, giving exact coefficients rather than approximations. Audio DSP developers can take it and verify directly; the 'perfect' claim is a bit absolute and still needs testing against RBJ formulas.
Perfect Biquad Bell and Band-Pass Filter Design Code
B datadome-plv3-vm-disassembler Recovers the plv3 VM's opcode, primitive and string tables from DataDome's bundled code, outputting readable listings without hardcoded offsets. A practical analysis tool for anti-bot reverse engineers; it breaks as DataDome rotates, and it skirts adversarial boundaries.
Disassembler for DataDome's plv3 virtual machine, the one that produces the plv3 token on the interstitial and the captcha. It recovers the opcode table, the primitive table, the value tags and the st…
B strata A single-file, zero-dependency vulnerability lab that draws vulnerable apps as seven-layer cross-sections, with 17 exercises offering both attack and defense views. It is more intuitive for teaching and self-learners than payload lists, though coverage is hand-maintained and depth is limited.
See why a web vulnerability works, not just how to trigger it — a single-file, zero-dependency security lab that renders a vulnerable app as a 7-layer cross-section. 17 labs, attacker & defender modes…
B superstack NativePHP's official Laravel starter: Filament admin plus the SuperNative desktop shell and MCP, a ready path for packaging Laravel as a desktop app. Practical, but still just a scaffold with no new problems; many starters exist, differing only in official maintenance.
Open-source Laravel starter with Filament, NativePHP SuperNative, and MCP
B XKCDMakie.jl Adds a hand-drawn layer to Julia's Makie, turning charts into xkcd style in a few lines. Small scope, honest naming, works once installed, the kind of tool you skim over a coffee; only useful to Julia plotting users, nowhere else.
Makie.jl diagrams with a hand-drawn look and feel
B DeepSeekStatus A whale in the macOS menu bar showing whether DeepSeek API pricing is currently at peak or off-peak. It serves only a small group of token-paying users, but the need is real and the implementation clean, a typical instantly understandable small tool.
A macOS menu bar whale that shows whether DeepSeek is in peak or off-peak pricing.
B OpenTwit An open-source X/Twitter client for HarmonyOS, filling the gap for third-party social clients in the HarmonyOS ecosystem, with 30 stars in a day showing real demand. But the description is one line and does not cover login method or whether you need your own API key; after X tightened its API, such clients mostly do not last, so observe first.
Open source X/Twitter adaptation for HarmonyOS Devices
B openfly Uses the fruit fly connectome as a signal source for intraday NIFTY options trading, placing orders through OpenAlgo; it is a proof-of-concept joke project. Curious minds at the neuroscience-quant intersection may find it worth a look, but strategy validity cannot be verified and it is hardly a reusable tool.
A fruit fly connectome (MaleCNS v1.0) trading an intraday NIFTY straddle through OpenAlgo.
B wloc A revived and maintained self-hosted implementation of Apple's network location service, preserving the original author's history and license, usable for location query testing. The purpose is concrete but sensitive, and most of the value comes from the original project, with little new.
WLOC community-maintained edition: restored based on Yu9191/wloc, retains the original author's history and license, supports Apple network location testing and self-deployment.
B snake Mounts SFTP as a local directory in Finder and doubles as terminal file management, aimed at Mac users too lazy to open an FTP client. The scope is restrained and the description honest, but macFUSE+sshfs and Mountain Duck already own this path, the name snake does not match the function, and whether it actually works needs your own testing.
A minimal macOS terminal and SFTP tool that supports mounting remote SFTP in local Finder.
B lecture-deck-skill Generates animated, themed HTML handouts from a storyboard as a Claude skill plus Python suite with three built-in themes. Handy for frequent workshop hosts, but it overlaps with Marp, Slidev and many AI slide skills, differing only in being made for Claude.
Build animated, themed HTML lecture and workshop decks from a storyboard. A Claude skill plus a Python kit with three themes.
B NetraUploader A Telegram file upload bot running on Cloudflare Workers for people wanting zero-server hosted storage. The approach is clear and deployment cheap, but many ready-made bots of this kind already exist, so differentiation is weak.
Telegram file uploader bot powered by Cloudflare Workers
B lembah-kenanga A pixel village life game in pure Canvas 2D and ES modules, no engine or bundler, with chat and accounts running in a zero-dependency Python process, and clear boundaries. Gameplay is still thin, and it clashes in theme with townies from the same week.
A pixel-art village life game in the browser — farm, fish, and befriend the villagers. Canvas 2D + ES modules, no engine and no bundler; public chat and player accounts run on a single dependency-free…
B Show HN: Toast, a by default in-terminal IDE A terminal-oriented out-of-the-box IDE that aims to replace Neovim's configuration hell with a complete setup that looks good by default, immediately usable for TUI users. 187 stars in 0.3 days and 72 points as an HN hot post make it one of the strongest momentum items in this batch. Commercialization is hard, limited to sponsorships or an enterprise edition; becoming a standard dependency depends on the plugin ecosystem. The main competitors are Helix, Zed and LazyVim.
https://paradise-runner.github.io/toast/
B Show HN: Spanda – Sub-microsecond LLM epistemic uncertainty in Rust It does zero-cost epistemic uncertainty quantification and hallucination detection for LLM inference, with a Rust implementation claimed to be 90,000 times faster than Semantic Entropy, targeting the latency pain of production hallucination monitoring. But Show HN has only 12 points and the repo 17 stars, so momentum is near zero. The path is embedding into vLLM or inference gateways as an enterprise hallucination monitoring API; the biggest risk is that the speedup cannot be reproduced and probe-based approaches face fierce competition.
B flash.tmux Adds easymotion-style jump labels to tmux panes, written in Zig, single-purpose, and ready after install. Similar tmux-jump exists, but a lighter implementation is still worth trying; 13 stars mean it hasn't spread, and usage examples are missing.
Navigate terminal panes by jumping
B MotionClone Reconstructs a reference video into an editable motion graphics project file; the video-to-timeline path is rarely attempted and appeals to people making title templates. Currently only a local Windows app, so reconstruction fidelity and output project format are what most need validation first.
Rebuild reference videos as editable motion projects. Local Windows app.
B emibios A GBA BIOS rewrite in Zig to replace the original firmware, a genuinely foundational piece for emulator development and homebrew cartridges. From a well-known reverse engineer with a clear scope; the limitation is a tiny audience, and compatibility can only be proven on real hardware.
a GBA BIOS replacement
B esphome-airplay2 An ESPHome component that turns an ESP32 into an AirPlay 2 receiver, building on an existing port and saving DIY audio box makers from debugging from scratch. Note the non-commercial license; latency and stability still need real-world testing.
ESPHome external component that turns an ESP32 into an AirPlay 2 receiver. Non-Commercial licensed (port of rbouteiller/airplay-esp32).
B ClaudeStatsBar Shows current context usage and the approximate cost of the next turn in the Claude Code status bar, so you see cost before typing. Very practical for heavy users and restrained in scope; it relies on official usage figures, so model changes may skew it.
A Claude Code status line showing how big your session has got and what the next turn costs before you type
B local-workspace-mcp A self-hosted MCP service that exposes local files, terminal processes, and Office documents to models in one place, saving you from installing a separate server per resource type. The space is already crowded (filesystem, desktop-commander), so differentiation must come from permission boundaries, which remain unclear.
Self-hosted MCP for local files, terminal processes, Office documents and named computers
B eqbyear Uses swept sine tones so you can manually mark the resonant peaks of headphones or earbuds, then directly generates a parametric EQ. Very practical for audiophiles without measurement gear, works right in the browser; novelty is average, but it explains one thing clearly.
EQ headphones and IEMs by ear. Sine sweep, mark the peaks, get a parametric EQ.
B Devspeak A Claude Code plugin for Brazilian developers to practice technical English, role-playing a fictional international team with their own real code as material—a specific angle. But it's essentially prompt orchestration, the results depend entirely on the model, and it's hard to verify.
A Claude Code plugin that coaches Brazilian developers on technical English for international remote jobs through role-play with a fictional international team, practicing on your own real code.
B cubism-api-bridge An unofficial TS/Python bridge and local HTTP service for Cubism Editor's external API—tools like this are genuinely missing in the Live2D ecosystem, and the author honestly labels it experimental. The audience is very narrow, and interface stability can only be found by trying.
Unofficial TypeScript/Python bridge and local HTTP server for Cubism Editor External API, with StandRig integration. Experimental.
B MiniWatts It uses private iOS APIs to read real charging wattage, input current and thirty temperature sensors, making a professional gauge for iPhone battery and thermal management. 51 stars in 1.4 days show geeks have real demand for raw battery data. Commercialization is constrained because private APIs cannot pass App Store review, leaving only sideloading; the risk is OS version changes killing it, with official battery health and CoconutBattery as competitors.
An iPhone battery, charging and thermal instrument built on private iOS APIs — real charger watts, what actually reaches the cell, all 30 temperature sensors, and the USB-PD handshake.
B reelbench-skills A collection of study notes and tool skills around AI video organization, callable directly by agents like Claude Code to lower trial-and-error costs in video generation workflows. 162 stars in 0.8 days hits both the AI video and agent skill trends; it could evolve into a video generation benchmark or skill marketplace. The risk is that note-style repos are easily forked and split, lacking a maintainable engineering core.
Learning notes and tooling skills for AI video - AI 视频相关的学习与工具 skill.
B Show HN: Godot and Rust based multiplexer (terminal panes and more) A multi-PTY terminal workspace built with Godot and Rust, aiming to replace tmux while adding AI agent orchestration and media panels. 76 points and 37 stars shows the concept attracts attention, but using a game engine for a terminal is a heavy technical path. The commercial opportunity is an agent orchestration workbench rather than the terminal itself; the biggest risk is that tmux, Zellij and Wezterm ecosystems are mature, so rewriting the rendering stack yields insufficient returns.
I wanted to share my side-project: gPTY. Started off as an idea to combine Godot and Rust in a project (two stacks I wanted to use more to learn more). The base inspiration was tmux - simply allow spa…
B sloppy Static analysis for Laravel that specifically catches bad smells left by AI coding agents, running deterministically and locally without calling models. A fresher angle than another linter, honestly described; currently Laravel-only, so the audience is limited.
Laravel-aware static analysis for the code-quality patterns AI coding agents leave behind. Deterministic, local, no LLM.
B knot A self-hosted bot that stores LINE text, images, and files into your own Cloudflare D1/R2, genuinely useful for heavy LINE users with clear boundaries; it's just deeply tied to the full Cloudflare stack.
A self-hosted LINE memory bot for saving text, photos, videos, audio, and files to your own Cloudflare D1 and R2.
B playstunts Rebuilds the 1991 Stunts / 4D Sports Driving in the browser with TypeScript, requiring players to supply the original game files, a fairly restrained approach to copyright. Old racing fans can play right away; the novelty is mainly the porting work itself, since gameplay and assets come from the original.
Native browser reconstruction of Stunts / 4D Sports Driving, with tools to build from your own original game files.
B repkgs Mic92's experimental package set on native Nix: cheap evaluation, relocatable outputs, a nushell builder, and LLVM-only cross-compilation. Technically dense from a veteran author, worth reading for anyone pushing Nix's build model; documentation is thin, so you'll read the code.
A package set on stock Nix: cheap eval, relocatable outputs, nushell builders, LLVM-only cross, compile cache in the compiler driver
B worktree-import-guard Checks whether tests actually run against the source you think, since git worktrees plus editable installs easily read stale paths. A small but real idea, usable by adding one check; the audience is limited to people with multiple worktrees open.
Verify that your tests actually executed the source tree you think they did.
B duo-open Recreates the iPhone's frosted-glass crease effect on Android foldables, driven by real hinge angle and requiring no root—a clever idea. But it's just a visual easter egg, locked to specific models and OS versions, with limited practical value.
iPhone-Duo frosted-glass fold effect for Android foldables, driven by the real hinge angle. No root.
B NotchOTP A native macOS app that puts TOTP codes under the notch, summoned with a hotkey, signed and notarized so installation is painless. Menu-bar TOTP tools are plentiful; its only selling point is the notch position, but it's small and complete.
Native macOS TOTP authenticator: your two-factor codes in a panel below the notch, one shortcut away. Developer ID signed and notarized.
B web-cli Reuses a logged-in Chrome session for local browser automation, callable by AI assistants and scripts—a practical idea. But Playwright, browser-use, and similar abound, and it doesn't explain how it isolates login-state risk or handles concurrency.
Local browser automation for AI assistants and scripts. Reuse your logged-in Chrome session and run ordinary tasks in the background.
B xjvpn A Mihomo-based Android transparent proxy client with per-app routing and domain audit logs, somewhat useful for root users troubleshooting phone traffic. But mature Mihomo clients like ClashMetaForAndroid and FlClash already exist; this just adds an audit layer, requires root, and targets a narrow audience.
Root Android transparent proxy client powered by Mihomo, with per-app routing and domain auditing.
B SwiftDependencyGraphExplorer Pick a Swift type and it locally draws its type-level dependency graph, exporting Mermaid or SVG, with source never leaving the machine. Useful for people digging through large Swift codebases; general dependency tools exist, but type-level granularity is the differentiator.
A native macOS app for exploring type-level dependencies in Swift projects. Pick a folder, select a root type, and visualize its dependency graph with Mermaid — fully local, no source code ever leaves…
B DuoFlip A menu bar utility that switches desktops based on the Mac screen's lid angle, doing one thing and explaining it clearly. It is fun for people who like gesture-based control; the limits are obvious too, it only serves macOS, and the actual feel depends on the hinge sensor.
Native macOS menu bar app with lid-angle-driven desktop transitions.
B decky A DIY 7-inch ESP32-S3 touchscreen Stream Deck, with both firmware and Windows software open source. Hardware tinkerers can build it; several open-source decks already exist, but this one wins on screen and feel details.
Open-source DIY stream deck: touch firmware for the CrowPanel 7-inch ESP32-S3 and a Windows desktop app
B image-gen-open-source A local workbench for generating, editing, and comparing images, with your own key connecting to OpenAI and Google models. The feature mix is practical, but such wrapper frontends are already plentiful, and "open source" doesn't include the models themselves, so the differentiation is unclear.
A local image workspace for generating, editing, and comparing AI images with Codex, OpenAI, and Google. Bring your own accounts and API keys.
B better-context-engine Yet another codebase indexing and context retrieval engine for AI coding assistants, heavily overlapping with ACE, Augment, and similar. The description gives no retrieval-quality comparison or integration steps, so it's hard to tell what it does better than existing tools.
ACE-compatible code context retrieval engine: real-time codebase indexing and low-latency context for AI coding assistants
B nadir-core An open-source library that scores ADAS sensor drift, integrated in shadow mode under Apache license—it targets a real engineering problem. But it gives no metric definitions or sample data, so outsiders can hardly judge its accuracy.
Open-source ADAS sensor drift scoring (Pulse/Echo). Shadow mode. Apache-2.0.
B module-ballerina-a2a Ballerina's official implementation of the A2A protocol; the spec mapping is tidy, but the audience is limited to the few people writing agent services in Ballerina. A2A SDKs already exist for many languages, so this fills out an ecosystem rather than breaking new ground.
Ballerina Implementation of the A2A Specification
B sk-music A Rust music player built on the Soulseek network—wiring a P2P protocol into a player is interesting, but clients like Nicotine+ are already mature, and the "free music" positioning easily slides into copyright gray areas.
A free music player based on the Soulseek network.
B venue-studio A browser-based venue seating chart editor with built-in venue templates and clearly defined scope; but its target users are event and venue operators, so generality is weak, and it doesn't explain export formats or collaboration.
Professional browser-based venue seating map editor with built-in arena layouts
B WMRL A paper's companion project page and code arguing for scaling automated research agents with world models, a relatively new direction; but beyond the paper there are no reproducible details, so for now it's just reading material.
Scaling Automatic Research Agents via World Models (arXiv:2608.12564) — project page and code release
B deepseek-v41-flash-spark Runs MoE inference on a single DGX Spark using resident hot experts plus NVMe streaming loads, an idea with merit. But the repo calls itself WIP, the model name does not match any known release, and another repo on the same topic appeared the same day, so it cannot be verified for now.
DeepSeek-V4.1-Flash on a single DGX Spark (GB10): resident hot experts + NVMe streaming, DSpark, OpenAI API. Work in progress.
B life-recorder From the browser-use team with nearly 180 stars in a day, but the repo has no description, so the name only suggests an app that records personal operations. The key issue for this topic is precisely privacy boundaries and storage, which are currently impossible to judge.
B Travel-Plan-Page Turns travel plans into a nice-looking web page on your phone; the need is real but the bar is low, and similar static itinerary page tools are plentiful. Unless sharing, collaboration, or offline support is added, it's just a personal utility.
Quickly turn your travel plan into a travel webpage suitable for viewing and sharing on mobile.
B mjbatch It runs thousands of MuJoCo simulations in parallel on pure CPU, addressing the simulation throughput bottleneck in robot learning and RL training; the author has credibility in the MuJoCo community, and 227 stars shows it is a must-have tool with potential to become a standard experimental dependency. Monetization is cloud simulation or enterprise support, competing with GPU simulation approaches.
A Python library for running thousands of MuJoCo simulations in parallel on CPU
B omarchy-apple-dev Builds and deploys iOS SwiftUI apps on Arch-based Omarchy on Apple Silicon without Xcode, removing a hard blocker for Linux developers doing iOS. Its 92 stars show a small but high-pain audience, enabling toolchain subscriptions or CI integration, with the risk that Apple toolchain licensing and upstream changes make it fragile.
Build and deploy iOS SwiftUI apps from Omarchy Linux on Apple Silicon, no Xcode required
B routeVSCODE Lets VSCode Copilot Chat switch models without restarting via a local proxy, hitting the real pain of wanting to plug in your own keys and models, with 326 stars in a day confirming demand. But it relies heavily on Copilot internals and subscription compliance, breaks when Microsoft changes things, and viable monetization is proxy/gateway middleware.
⚡ Zero-reload dynamic AI model switcher & local proxy for VSCode Copilot Chat via 9Router
B Show HN: Compute polynomials twice as fast Halves the multiplication count for polynomial evaluation with a new construction and gives a complete Lean formal proof, directly affecting hashing, erasure coding and ZK proof performance. Its 128 likes show both academic and engineering interest, the path is adoption as a standard algorithm by crypto libraries, and the risk is unverified constant overhead and engineering deployment.
A few years ago my coauthor and I was wondering if we could reduce the number of multiplications used for hashing algorithms. We had a construction and a 100 page proof, but we were not 100% sure it w…
B 3dviz-pro-max An Agent Skill that turns a one-line idea directly into Three.js/Blender scenes, with 223 recipes and a knowledge base, solving slow from-scratch 3D visualization. 133 stars in a day show the Claude Code/Codex plugin ecosystem is hot, enabling paid skill packs and template markets, with the risk that skills are highly homogenized.
Agent skill for creative 3D visualization: turn an idea into a Three.js/Blender scene worth exploring. Claude Code + Codex plugin, 223 recipes, 440 knowledge records, 22 proved kits, 37 runnable studi…
B one-ip One-click lookup of IP purity, datacenter origin and proxy traces, serving IP profiling needs in risk control, scraping and multi-account scenarios. Its 57 stars show a niche with real demand, enabling a pay-per-use IP intelligence API, but competitors like ipinfo and Scamalytics are mature, and the risk lies in data-source quality and accuracy.
Instant insight into IP purity and risk rating. With just one click, physical coordinates, data center origins and proxy traces are all laid bare.
B pi-better-rewind-redo Adds multi-level file and conversation rollback plus recoverable checkpoints to AI coding agents like Pi, without relying on git, directly addressing the pain of agents breaking code with no way back. The positioning is clear, but it only serves the Pi ecosystem and is useless outside it.
Rewind files and conversations in Pi, with multi-level redo and recoverable checkpoints. Powered by filesnap. No Git required.
B teledrive A personal cloud drive over Telegram MTProto, shipped as a single static binary with a web panel, supporting chunked resumable uploads and 206 streaming playback, ready for self-hosters. The description is restrained and flags a ToS Safe Mode; novelty lies in completeness, since telegram-drive and others exist.
⚡ High-capacity personal cloud storage powered by Telegram MTProto. Single static binary with Google Drive-like Web Dashboard, resumable chunked uploads, HTTP 206 video streaming, and strict ToS Safe …
B paygateme An unofficial Go SDK for Shopee and GoPay merchant payments that saves Southeast Asian e-commerce developers from wrapping the APIs themselves; the author clearly states it's unofficial and keeps scope tight. Risks are platform API changes, plus missing tests and version commitments.
Unofficial SDK Go library for Shopee & GoPay merchant payments
B novel-agent-studio A local AI workbench for Chinese novels covering story planning, chapter writing, evidence-based review and story memory, solving consistency and revision pain in long-form creation. Its 43 stars show demand exists, enabling subscriptions or material markets, with the risk of squeeze from both general agents and web-novel platforms' in-house tools.
A local desktop AI workbench for Chinese novel writing: story planning, chapter-by-chapter writing, evidence-based review, local revision, and story memory.
B printfilm A Chinese AI science-video and comic-drama creation platform that batches the script-to-finished-video workflow, solving short-video creators' capacity problem. Its 183 stars show strong Chinese creation demand, but the AI video space is crowded, and without in-house models or distribution it's hard to retain users, with monetization mainly from compute resale and memberships.
PRINTFILM: an AI science video and comic drama creation platform
B dlss5-neural-amd A ReShade plugin that ports DLSS's neural rendering network to AMD cards via HIP, already tested in PCSX2 with an open call for forks; a rare piece of GPU-side reverse engineering, limited in that it only works for specific pipelines and is demanding to configure.
ReShade add-on running the DLSS-NR network on AMD GPUs via HIP, RenoDX-shaped. Tested on PCSX2. Published to be forked. Discord: https://discord.gg/wYhvS3JSHM
B Show HN: Syq – copy files between machines fast (better than rsync) Accelerates rsync-style file transfer with multiple parallel connections and direct encrypted TCP, also supporting remote-to-remote and persistent SSH sessions, solving slow large-file cross-machine copies. Its 37 likes and rapid growth show real interest, but rsync's mindshare and distro preinstallation are moats, and commercialization needs to target enterprise sync and backup.
I got frustrated with the slowness of rsync and made an alternative that works faster by using multiple parallel connections, direct encrypted TCP when available, and other optimizations. I also added…
B StandRig Turns layered PSDs into MCP-driven 2D rigging and playback, with a local service plus an embeddable runtime, aimed at animators editing character art with AI; the idea is novel, but setup is demanding and the ecosystem is small, with Live2D and Spine as alternatives.
AI-editable 2D modeling and playback via MCP. Parts-separated PSD input, standalone local service, and embeddable runtime.
B Show HN: What if the speed of light was 5 km/h? Scales the speed of light down to 5 km/h for relativity visualization, with 563 points of same-day viral interest, genuinely solving the teaching difficulty of relativity's counterintuitiveness. But it's a single-page demo rather than an integrable tool, with no API, reusable library or standard position, monetization only via content and education sponsorship, making it a showpiece signal.
I've always wanted to make a visualization where the speed of light was scaled down to human speeds, so that we could intuit relativistic effects with everyday objects. Here is the first version of it…
B Show HN: DOOM in the kernel, or fibers in eBPF Runs DOOM, Lua, llama2 and CPython inside eBPF, proving bounded execution and memory virtualization under verifier constraints, technically novel. At 33 likes interest is moderate, its practical value lies in programmable eBPF observability and XDP logic, but it's far from production-ready with almost no monetization path.
Some time ago my coworker who was working on first version of Perforator ( https://github.com/yandex/perforator) kept talking about eBPF, so I got curious about its ISA and restrictions. That was arou…
B EarGuard A small resident Windows tool that blocks momentary 100% volume spikes, clamping them back within 2ms to protect ears and headphones, with a concrete problem and solution. Its restrained scope is a plus; the risk lies in audio hook compatibility and permission requirements.
Hardware volume spike protector for Windows. Clamps rogue 100% audio bursts in <2ms to protect your hearing and IEMs.
B clawd-conduit Hooks Claude Code's 15 hook events into the desktop pet Clawd, so even permission confirmations can be clicked on the desktop — a very precise scope; but it's only useful to people using both, making it a fun, clean little tool.
Full-lifecycle wiring between Claude Code and the Clawd on Desk pet — all 15 hook events, with permission prompts answered straight from the desktop
B mist A zero-dependency front-end splash effect where a finger wipes fog and stirs water ripples, with an iOS integration example; the code is presumably small. Good for anyone wanting a ready-made interactive effect, though it's just a demo and similar CodePen snippets are common.
A frontend splash screen page where a finger swipe wipes away thin mist and triggers water ripple effects. Zero dependencies, with an iOS integration example.
B Audio-Finder Adds an activity view to macOS showing which tab or app is playing sound, with Chrome/Brave connectors; the need is real but small, and the system volume panel plus existing menu bar tools already cover most of it — an optional little toy.
Native macOS audio activity app and Chrome/Brave browser connector.
B Smartisan-Modern A port adapting Xposed modules on Smartisan phones to the newer libxposed API 102, a real need for those still tinkering with old Smartisan device modules. The audience is tiny and it's constrained by upstream API stability.
A third-party modern Xposed API adaptation of Smartisan's 锤锤, currently supporting libxposed API 102.
B deepseek-recipe A new Rust repository from the official deepseek account that gained over two hundred stars in a day, but with no description or README summary to judge the contents; it could be training/inference recipes or just internal tools released. Hot enough in stars, but too little information, so wait and see.
B jellyfin_inventory A Jellyfin plugin that builds a sortable, filterable master table for the media library, easing the weak native browsing. The C# plugin has a clear purpose but a single function, and similar report plugins already exist in the community.
A sortable, filterable table of everything in your libraries.
B Mi-Ripple Targets ripple artifacts in repeatedly edited AI-generated images, pointing at a specific problem few address. But the description is full of invented terms, with no before/after samples or metrics, making real effectiveness hard to judge.
MIYANG diagnosis-guided restoration for digital ripple artifacts in iteratively edited AI images
B Qpro-Enhanced-FT Enables tongue tracking and eye convergence on a rooted Quest Pro, a small unlock for VR face tracking. Useful to headset tinkerers, but the barrier is high, the audience narrow, and stability and rollback are unstated.
Enables tongue tracking and eye convergence on a rooted Quest Pro
B vivace A Rust rewrite of a Composer-compatible PHP dependency installer, with its positioning clear in one sentence, so PHP developers can try it today. The repo is only a day old, so real usage isn't visible yet; success hinges on handling composer.lock and private repositories.
Fast, Composer-compatible installer for PHP, written in Rust
B Future-Coast A visualization of sea-level rise flooding along China's coast using real elevation data, with a concrete problem and traceable data sources, usable by science communicators and the general public. The limitation is that it covers only China's coast as a single-page demo, so depth is limited.
See whether your coastal area will be submerged after global warming sea-level rise. Interactive sea-level rise visualization for China's coastline based on real elevation data | Interactive sea-level…
B kav Puts a full bus timetable into an app with no ads and no account, letting you plan routes without signal; the need is clear and the implementation restrained, and Android users can install it today. Limited by city coverage and timetable update frequency.
An ad-free, account-free transit app for Android. The whole timetable is on the phone, so journey planning works with no signal.
B DuoLikeAnimation A Swift recreation of the iPhone Duo fold animation that rode the hype name to 70+ stars, with no description or usage notes; by its size it's a single demo view. iOS motion designers can glance at it, but it isn't yet a project.
B microduck-replica-cad SolidWorks drawings and a 21-page assembly guide for an open-source robot dog replica project, useful to hardware hobbyists who want to build one. But it's companion material to a main repo, not a tool itself, so its general utility is limited.
Editable SolidWorks 3D drawings and a 21-page assembly and installation manual replicated from Microduck · Drawing author 机械行者Robo · Companion main repo fanhao375/microduck-replica
B ocean-simulation A real-time demo of FFT/JONSWAP ocean surfaces with foam, caustics, refraction, and weather, with detailed graphics work. But it's a personal showcase with no reusable library or parameter docs; for rendering, the official three.js examples are about as good.
An interactive real-time ocean simulation combining FFT/JONSWAP waves, depth-aware coastal behavior, foam, caustics, refraction and dynamic weather around a tropical island.
B elegantfin-jf12 A compatibility patch making the ElegantFin theme work with Jellyfin 12's new layout; the author honestly calls it a stopgap until upstream support arrives. But it's just a CSS override that breaks as soon as upstream updates.
Makes ElegantFin work on Jellyfin 12's Modern layout, until upstream supports it
B google-seo-mcp An MCP server bundling Search Console, GA4, PageSpeed, and WordPress, genuinely convenient for technical SEO practitioners, and this kind of data integration is uncommon. But 80 tools means a very wide maintenance surface, so first check whether it can reliably run GSC authorization.
SEO & GEO MCP server for Claude, Codex, Cursor and any MCP client: Google Search Console, Google Analytics 4, PageSpeed Insights, Core Web Vitals, structured data, llms.txt, AI crawler checks, WordPre…
B localmesh-engine Turns photos into textured meshes offline on a single 8GB GPU, built on TRELLIS.2 plus a four-view path, and the low VRAM threshold is a real selling point. Limited by dependence on the upstream model and no evaluation comparisons.
Photographs to a textured mesh, on one NVIDIA card with 8 GB, offline. Built on TRELLIS.2, with a four view path.
B justPostgres A lightweight entry point for running multiple Postgres instances on one machine, a practical idea, but docker compose, pgAdmin, and others already cover this heavily. The one-line description never explains isolation, ports, or backups.
A lightweight postgres hub to run multiple postgres database from a single machine
B fx-ui A GPUI desktop client for the fx coding agent, from an author with a Scira background; but fx and libfx themselves are obscure, the repo doesn't explain installation or usage, and its value depends entirely on whether anyone actually uses that agent.
A desktop app for the fx coding agent, built on libfx and GPUIX.
B anything2explainer Turns any topic into a generated explainer video with voiceover and subtitles, as a Claude Code/Codex skill. It gained 76 stars in 1.3 days, showing demand for automation in content creation; commercialization means a paid skill or template for vloggers and course makers, but video generation tools change fast and competition is fierce.
Topic in, narrated explainer video out. A Claude Code / Codex skill that turns any topic into a black-canvas motion-graphics explainer video with TTS voiceover, subtitles and a chapter progress bar. C…
B spotkit A Claude Code skill that converts feature descriptions into SVG illustrations for a unified design system. It gained 49 stars in 1.3 days, fitting the agent skill trend; monetization is paid design-skill templates or integration into design tools, with the risk that SVG illustration is limited and users may prefer direct image-generation AI.
Spotkit — a Claude Code skill that turns a feature description into minimal, abstract SVG product illustrations that all belong to one design system.
B haiming-app-monetization An AI skill helping developers commercialize apps, such as onboarding, paywall and in-app purchase design. It gained 79 stars in 0.6 days, showing the agent skill space is hot; monetization is a paid skill or integration into agent marketplaces, with the risk of being covered by general AI instructions or lacking systematization and thus limited value.
海明Dev: App monetization Skill | onboarding, paywall, in-app purchase options, competitor research
B INSLIB An inertial navigation algorithm library for embedded fields such as robotics and drones. 188 stars in 1.6 days shows real demand, but inertial navigation already has mature solutions and limited innovation. The commercialization path is as an underlying dependency for hardware and software companies or enterprise support; the risk is that existing commercial software or open-source libraries hold the market, and domain knowledge is required.
Open Source Inertial Navigation Library
B fast-rtxvsr Turns RTX video super-resolution into a standalone command-line single-pass pipeline: GPU decode, VSR, and encode in one go, no ComfyUI needed. RTX users wanting to batch-upscale old videos can run it today; limits are it's locked to Nvidia cards and offers no quality comparisons.
Standalone NVIDIA RTX Video Super Resolution CLI - GPU decode -> VSR -> encode in one pass, no ComfyUI
B rubric A TypeScript schema-first structured extraction library that pulls LLM output according to a given schema, with a one-line purpose and restrained interface. It is a lightweight option for developers who need reliable JSON output. Mature alternatives like instructor and outlines already exist, so whether it is worth switching depends on validation and error-handling details.
Schema-first structured extraction from LLMs.
B headroom A closed-loop agent system that judges audio mixes using DSP metrics instead of subjective feel, aimed at audio engineering. It gained 47 stars in 2.0 days; commercialization means becoming a production tool or plugin, but professional audio software already exists and automated mixing struggles to win expert approval.
A closed-loop agent system that mixes and masters audio toward a measured target. Orchestration graded by DSP measurement, not vibes.
B nibble A Go library for splitting long text into retrieval chunks, clearly scoped and restrained, usable today for Go-side RAG. The knock is that chunking has plenty of existing solutions, from langchaingo to various semantic splitters, and the repo doesn't explain how its splitting strategy differs.
A Go library for splitting long text into retrieval-sized chunks for RAG.
B omarchy-in-omarchy Spins up a self-installing Omarchy virtual machine with one QEMU/KVM command, skipping the wizard, password, and login, and discarding it after use. Perfect for people who want to try an Arch-based desktop without dirtying their main machine; the limit is it only makes sense for this one distro.
A disposable Omarchy machine in QEMU/KVM that installs itself: no wizard, no disk passphrase, no login prompt
B Surge A contribution sprint platform for scheduled funding in the Stellar ecosystem, where maintainers post tasks with points and contributors apply with proposals to earn USDC rewards. 93 stars in 1.3 days shows community interest, but the market is limited to Stellar. The commercialization path is becoming the ecosystem's standard incentive mechanism or expanding to other chains; the risk is competition from Gitcoin-like players and the ecosystem's limited size.
Funded, time-boxed contribution sprints for the Stellar ecosystem. Maintainers post scoped issues with a point value, contributors apply with a plan, and accepted work splits the wave pool in USDC. Fr…
B motion-web Builds motion-first creative websites with Three.js/WebGL as a Claude Agent skill, including a physics engine and automatic validation. It gained 43 stars in 1.8 days, meeting frontend creative demand; commercialization means a skill-market product or helping agents generate more motion, but competition is fierce and debugging complex.
Build motion-first, high-touch creative websites on vanilla Three.js / Canvas 2D / WebGL / CSS. A Claude Agent Skill with real physics and automated verification.
B naming-scout After naming a product, automatically checks domains, trademarks, GitHub, npm, app stores and search, as an agent skill for Claude Code and similar. Closer to the dev workflow than sites like namechk; trademark and store result accuracy still needs testing.
Names your startup, product or side project, then proves you can actually use it. Checks the domain registry, trademark, GitHub, npm, app stores and search. An agent skill for Claude Code, Cursor, Cod…
B agent-html A zero-dependency single-file HTML design system paired with a skill for LLMs, aiming to have models directly produce pages that open offline. The idea is practical, but the AI HTML template layer is already crowded, and it doesn't explain how its layout prototypes beat the others.
Zero-dependency, single-file HTML design system & agent skill inspired by shadcn/ui. 100% offline, zero npm, 6 layout archetypes & atomic arsenal for LLMs.
B ai-level-check It reads local agent logs and grades your AI usage from LV0 to LV5, explicitly saying it doesn't rank, with a restrained tone. It's a clean little toy that does one thing and explains it clearly; the limitation is that one look is enough, and you still have to judge the grading criteria yourself.
AI 幾級了 — What level is your AI usage? Reads local agent logs and writes an LV0–LV5 report. No ranking.
B URBER A reference implementation of a paper using a small-scale rule-based method to solve escape routing for biochip sample transport, narrow in scope, clear in goal, and not oversold. It's directly usable by researchers in EDA and microfluidics; the limitation is that the repo only provides code, so you still have to read the paper to get it running.
Code for URBER: Ultrafast Rule-Based Escape Routing Method for Large-Scale Sample Delivery Biochips
B fantasy-football-manager Wraps ESPN fantasy football drafting and lineup setting as an MCP service for clients like Codex, with login state, approval mode and per-team limits. A real need for people who both play fantasy and use agents; the risk is reliance on third-party site APIs that may break at any time.
ESPN draft and lineup tools for Codex and other MCP clients, with browser sign-in, approval modes, per-team limits, and server monitoring.
B whiteboard-explainer Turns handwritten whiteboard explanations into frame-aligned animation: shapes drawn stroke by stroke, text written stroke by stroke, locally at zero cost and re-rendered in seconds. Useful for teaching-video makers; the limit is it needs an agent to drive it, and complex layouts still need manual tuning.
Whiteboard explainer animation skill. Turns an explainer script into a "real hand-drawing process" video — shapes drawn stroke by stroke, text written stroke by stroke, drawing wherever the narration …
B runnel A small Go HTTP egress gateway with per-domain rate limiting, circuit breaking, queuing, caching and SSRF protection. A narrower option than Nginx for teams needing safe agent internet access; maturity and load-test data are not yet provided.
A small, secure HTTP egress gateway with per-domain rate limiting, circuit breaking, queueing, caching, and SSRF protection.
B minidauth Decentralized authentication that solves the need for a key no one holds and governance capability in existing auth systems. It gained 94 stars in 1.4 days, but the use case is niche; commercialization means an identity layer for distributed systems or an open standard, but many cryptographic alternatives exist, security audits are hard, and user trust is difficult to build.
Mini decentralised auth. Gives an existing auth system a key nobody holds, with governance over who can decrypt.
B memory-hub Local-first cross-agent working-memory retrieval on Windows, addressing a real pain point: multiple coding agents each remember only their own context. But mem0, basic-memory and many others already cover this space, and it doesn't explain its indexing approach or how it differs, so no advantage is visible yet.
Local-first cross-agent work memory search for Windows
B codex-glass It targets a specific pain point: on Windows you can't see Codex's token consumption and quota, so it builds a floating window with local SQLite history merging. The scope is clear and it works out of the box; similar usage dashboards are common, and its difference is being desktop-resident and Chinese-oriented.
A native Windows Codex usage floating widget and liquid glass dashboard: tokens, cost estimates, quotas, and SQLite history deduplication and merging.
B DiffExplorer A differential analysis tool for lab researchers who don't code, unzip-and-run, producing statistics and charts directly for group meetings, with a clear audience and goal. The worry is that no-code tools like this tend to be vague on statistical methods, so you need to see which experimental designs it supports.
Differential analysis tool for coding-free researchers – instant statistical results and charts, ready for your lab meeting. Just unzip and run! No-coding!
B learn-from-materials Converts PDFs, books and papers into interactive learning web pages with traceability, quizzes and notes. It gained 59 stars in 1.7 days in the education AI space; commercialization means a paid tool or cloud service for students and enterprises, but similar products exist and conversion quality is hard to guarantee.
Turn PDFs, books and papers into interactive learning webpages｜Turn complex materials into traceable, quizzable, note-taking learning webpages
B Show HN: EmbedFlow –> Upgrade embedding models without re-embedding your corpus It solves the problem of re-embedding the entire corpus when upgrading embedding models, enabling zero-downtime migration by reranking old index candidates in real time. It got 38 stars on release but only 5 points on HN, with no sign of users yet. The commercialization path is becoming a standard vector database migration tool or integrating into existing indexes; the risk is built-in migration in faiss, qdrant or pgvector, or embedding models offering reranking interfaces themselves and replacing it.
I was playing around with embedding models and I realized the moving between embedding models on a large corpus can cause a huge backfill, as all of the previous documents would have to be re embedded…
B ReUWP A compatibility layer that translates UWP Minecraft Bedrock into Win32, targeting only two old versions, 0.15.10 and 1.1.5. The technical substance is real and the scope clearly stated, making it a good sample for studying the UWP sandbox; but it's nearly useless to ordinary players, and playability is doubtful.
ReUWP is a UWP to Win32 translation layer for Minecraft Bedrock 0.15.10/1.1.5
B cli-tools Collects scattered Mac CLI tools into one directory and provides agent-friendly JSON output, useful for people who want local command-line tools callable by AI. But it doesn't explain how tools are discovered and indexed, and what the native app actually does is vague, so the value remains unproven.
One catalog for every CLI tool on your Mac: native app plus a JSON-friendly CLI for AI agents
B grove A Swift-written MCP server that lets AI assistants read and write macOS Reminders and Calendar directly, currently covering only these two native apps. Anyone using a Claude-style client on a Mac can install it immediately, and the native implementation is lighter than a Node bridge; the limitation is narrow scope, and several Apple calendar MCPs already exist.
Native Mac tools for AI assistants. A Swift MCP server for Apple apps, starting with Reminders and Calendar.
B supabase-multitenant This wraps self-hosted Supabase into a Supabase.com-style multi-tenant platform that runs on VPS, Coolify, or Dokploy, addressing the real need of giving each client an isolated instance. But the description is marketing-heavy, and whether isolation boundaries and upgrade paths are reliable can only be verified by reading code; for now it looks more like a set of orchestration scripts.
supabase-multitenant lets you self-host Supabase as your own Supabase.com-style cloud, running anywhere you choose—VPS, Coolify, Dokku, Dokploy, or your own infrastructure. Manage everything from one …
B boss-job-agent A self-hosted job-hunting agent that runs workflows on BOSS直聘 through visible browser automation and leaves reviewable receipts. Batch applicants have a real need, and "visible browser plus receipts" is more honest than a pure script. The limitations are platform risk controls and account risk, and configurability is unverified.
Self-hosted Codex job-search agent with configurable workflows, visible browser automation and durable receipts
B claude-code-statusline Adds a status line to Claude Code showing project, context usage, session duration, and quota, integrated with tmux. The need is real, but similar scripts are already plentiful, making this more a small config than a new tool.
Claude Code status line: project, context window, session time and usage limits, with tmux-vitals integration
B agent_room A CLI for multiple people to collaborate on the same Codex session and project, with participants needing host approval; the idea is concrete and rare. But with only 9 stars and under two days old, it doesn't explain how multi-user conflicts are handled or how it works with git, so for now it's just an interesting prototype.
CLI collaboration on one shared Codex session and project, with host-approved participants.
B chlorine-server A from-scratch inference engine rewrite for AMD Strix Halo that keeps the same wire protocol and OpenAI-compatible API, well-matched to the hardware and ambitious. But the upstream halogen 0.1.3 and the target models are hard to verify, and there's no evidence of it running, so it reads more like a statement for now.
Clean-room AGPL-3.0 reimplementation of the halogen 0.1.3 inference engine for Qwen3.8-27B on AMD Strix Halo (gfx1151) — same wire protocol and OpenAI-compatible API. Not affiliated with Peonist.
B swimmich A native SwiftUI iOS client for Immich, aiming for feature parity with the official Flutter version. Attractive to self-hosted photo users, but claiming full parity after a day of building is unrealistic, and long-term maintenance is doubtful.
Native iOS SwiftUI client for Immich — self-hosted photo & video management. Full feature parity with the Flutter mobile app.
B ai-memory-skillpack A "bounded" project memory skill pack for Codex CLI and Claude Code, pitched at limiting memory size so it does not blow up the context. The direction is right, but this space is extremely crowded lately, and it does not explain retrieval, eviction, or conflict-handling strategies, making it hard to judge how it beats existing options.
Bounded project-memory skills for Codex CLI and Claude Code.
B DuckDuckMove A Windows 11 avatar animation tool with a built-in duck GIF, support for custom animations, and one-click restoration of the original avatar; the need is small but clear, and the description is honest. The limitation is that it has only one feature and works by changing the system account picture, so whether it breaks after system updates is unknown.
Make your avatar move. A Windows 11 animated avatar tool supporting the built-in duck GIF, custom GIFs, and avatar restoration.
B DepthBenchCAD A three-tier evaluation for generative CAD systems, testing audit depth and evidence allocation, a new and narrow topic. The problem is it only matters to the small circle doing CAD generation research, and it doesn't say whether the eval set is public or how to reproduce it.
A three-level benchmark for evaluating audit depth, evidence allocation, and reliability in generative CAD systems.
B underplan A cable routing planner for Multiboard and Underware printed wall systems, a very narrow audience but a real need if it fits. The problem is it only serves this one hardware set, so anyone without these modules can't use it at all.
Interactive Underware & Multiboard Cable Management Planner
B tagtinker-adv A small toolset for tinkering with electronic shelf labels on the Cardputer ADV; the hardware modification angle is interesting. But the description doesn't elaborate on ESL, with no wiring or protocol details, making it hard for outsiders to reproduce.
A toolkit for Cardputer ADV to tinker with ESL's
B midi-m8 A small command-line tool that converts Dirtywave M8 projects into MIDI tracks, a real need for M8 users. But the audience is tiny, it doesn't say which tracks or firmware versions are supported, and the same author publishing three repos in one day also lowers credibility.
A command line tool for turning Dirtywave M8 songs into Midi tracks
B cfmeeting A multi-platform meeting demo built on Cloudflare Realtime, covering macOS, Windows, iOS, Android and Web; open-source examples of such realtime capability are still rare, so the implementation is worth a look. But it calls itself a demo and lacks deployment docs and signaling notes, so it is far from usable as a product.
An open-source multi-platform meeting app based on Cloudflare realtime-demo, supporting macOS, Windows, iOS, Android, web, and H5
B architecture-assessment Turns architecture review into an agent skill, bundled with Roslyn syntax analysis and a SQLite source directory. The author has standing in the .NET community, but what problem the three pieces solve together and how to run them aren't explained, making it look like a personal workflow dump that outsiders can't easily use.
Evidence-first architecture assessment and simplification skills, with Roslyn syntax analysis and a SQLite source catalog
B pc98-ai-localization-handbook A research handbook on localizing old PC-98 games, covering FM/SSG, MIDI and PCM verification flows, written in Traditional Chinese for easy AI reading. The material itself is scarce, but it's essentially a docs repo that can't be used directly, serving only the small retro-game translation crowd.
AI-readable PC-98 localization research handbook in Traditional Chinese, with FM/SSG, MIDI and PCM validation workflows and synthetic examples.
B viewport-ruler A Chrome/Edge extension that switches the viewport among five standard widths with one click, simulated in pure CSS and applied only to the current tab. Lightweight, but DevTools responsive mode already does the same thing; the only difference is fewer clicks.
Chrome/Edge responsive debugging extension: click to switch the page viewport width (pure CSS simulation, controls only the current tab), with five built-in standard presets and a real-device checklis…
B midori-companion-avatar An unofficial fan project for 明前奶绿: a local web AI companion avatar with audio lip-sync, dynamic character art, lyric visuals, and generic Agent integration. Its positioning is honest and boundaries clear, but completeness is more like a fan toy, and the actual lip-sync and Agent integration need to be tried firsthand.
Unofficial, non-commercial fan project for Mingqian Nailv: a local web AI Companion Avatar supporting audio lip-sync, dynamic character art, floating icons, lyric visuals, and generic Agent integratio…
B ClayGlobe A SwiftUI clay-style 3D globe component that plants a waving flag in each country you've visited, suitable as UI material for travel apps. The craft is appealing but the use is single-purpose, and ready-made globe components are plentiful.
A claymorphic 3D globe for SwiftUI that plants a waving flag on every country you visit
B mousiki A customizable TUI music player focused on being lightweight and fast. Mature terminal players like cmus and termusic already exist, and this offers no differentiating capability.
Highly customizable TUI music player designed to be fast and lightweight
B claude-agent-sdk-bridge Explains how to hook subscription quotas into a self-built chat frontend using the Claude Agent SDK, with parameters, permission callbacks and deployment notes. Practical for those wanting to save API fees, but it is a tutorial note and the approach risks violating terms of service.
Using the Claude Agent SDK to connect subscription quota and the full harness to a self-built chat frontend: differences from claude -p, query() parameters, permission callbacks, fixed prefix, prompt …
B hyalite--liquid-glass Real refractive liquid glass built with SDF and SVG, no WebGL required; 132 stars show frontend developer interest. Could be commercialized as a UI library or paid component. The main risk is browser compatibility, with Apple's liquid glass as a competitor.
Real refraction liquid glass for the web: SDF lens maps + SVG displacement + backdrop-filter. One file, no WebGL.
B linkedin-agent-skill Packaged as Claude Skills, it handles LinkedIn account operations such as content generation, comment replies and removing AI traces; 33 stars in 1.0 days show content creators have demand. It could commercialize as a social media automation subscription, but risks are platform anti-automation policies and competition from official AI tools.
Eleven free Claude skills that run a LinkedIn account: posts off 21 hook formulas, comments, replies, profile score, weekly plan, and a humanizer that strips the AI fingerprint and scores the draft be…
B glm-flash-offline-client An offline desktop client running GLM-5.3-Flash to protect privacy; 106 stars show demand for local AI. Commercialization could be a privacy-focused AI product, but model performance is limited. Risks are model updates, with Ollama as a competitor.
Offline desktop client for GLM-5.3-Flash — run private, local chat sessions without a server connection.
B ruraldte-engine Handles Chile's DTE electronic tax document compliance, including signing, CAF/TED and direct SII submission, a local essential infrastructure; 22 stars in 0.2 days but already in production use. Could become the standard Chilean DTE library or a commercial service. The main risk is its regional narrowness, with local Chilean vendors as competitors.
Motor open source de documentos tributarios electrónicos (DTE) chilenos: firma XMLDSig, TED/CAF, envío directo al SII y timbre PDF417. Los 12 tipos del catálogo, en producción.
B cloudflare-turnstile-bypass Provides dependency-free Turnstile token acquisition for CI and automation, removing captcha blockers; 367 stars in two days shows strong automation demand. Could be a commercial anti-bot service, but carries compliance risk. The main risk is Cloudflare updating its detection, with other captcha solvers as competitors.
Work with Cloudflare Turnstile in Python: find the sitekey, get a valid cf-turnstile-response token in seconds. Dependency-free library + CLI, built for CI and automation.
B whiteboard-animator Turns whiteboard images into hand-drawn reveal videos as the rendering engine behind Kinoslide; 73 stars show content creators need it. Could be a commercial video generation service. The main risk is AI generation quality, with VideoScribe as a competitor.
Turn a whiteboard-style image into a hand-drawn reveal video. The render engine behind Kinoslide.
B argo9s This is a K9s-style terminal UI for monitoring Argo CD, giving operators real-time views of resources, and 70 stars show DevOps needs it. Commercialization could be an open-source tool with enterprise support; risks include Argo CD API changes, and the competitor is k9s.
A K9s-inspired terminal UI for monitoring Argo CD resources in real-time
B Awesome-OKF Defines an open knowledge format so agents can search tools, plugins and skills. Its 102 stars show the AI agent ecosystem needs a standard; commercialization could be a knowledge base standard or hosted service, but standards compete and the rival is the MCP registry.
OKF (Open Knowledge Format) — curated catalog of tools, plugins, skills, proposals, and docs for agent-friendly knowledge. YAML-driven, agent-searchable, MCP-ready.
B AstraDraw Solves style control and editability in AI image/PPT generation by systematizing reference material, color and layout, and local refinement; 24 stars in 1.6 days show early users. It could become an enterprise presentation SaaS, but strong competitors include Canva, Gamma and Tome.
Let Astra not only draw editable PPTs, but also draw in the style you want. Reference materials, color and layout, then local refinement.
B To-Sheet-Music-Skill Automates turning song recordings into campus band scores, covering stem separation, transcription, arrangement and difficulty adjustment with MIDI/PDF/MuseScore output; 29 stars in 0.8 days shows educational music interest. Could be a music teaching SaaS or MuseScore plugin. Risks are transcription accuracy, copyright and competitors.
Automatically transforms mixed song recordings into playable five-piece school band arrangements through audio separation, music transcription, arrangement, and difficulty adjustment. Outputs include …
B turnstile-bypass A cross-platform Turnstile solver supporting macOS, Windows and Linux; 355 stars show multi-platform demand. Commercialization could be an API service, but compliance risk is high. The main risk is Cloudflare escalating its countermeasures, with similar libraries as competitors.
Cross-platform Cloudflare Turnstile solver (macOS, Windows, Linux)
B dlssg_for_sm86 This enables DLSS frame generation on RTX 30-series cards, fixing a hardware feature cut; 210 stars show gamers welcome it. Commercialization could be a paid mod, but NVIDIA may block it; risks include driver updates breaking it, and the competitor is the FSR3 mod.
Here is a dlssg for RTX30 Series GPU
B Show HN: Genotype: language that compiles to idiomatic TS, Rust and Python This solves type synchronization across TS, Rust and Python by generating idiomatic type code; after two years of development it has only 7 Hacker News points but a solid concept. Commercialization could be a standard tool for API client generation or a compiler service; risks include low language adoption, and competitors are protobuf and OpenAPI code generators.
Hey HN! I just shipped the first public version of a new programming language that I have worked on for the past two years. Its goal is to help synchronize types between TS, Rust, and Python (more to …
B TriAevum Natively recompiles Ocarina of Time 3D to PC, solving 3DS game porting; 130 stars show reverse-engineering community interest. Could be commercialized as a recompilation tool, but Nintendo copyright risk applies. The main risk is legal, with Ship of Harkinian as a competitor.
Experimental native PC recompilation runtime for Ocarina of Time 3D. Windows x64, NRI/Vulkan, single-screen UI and ROM-only setup.
B slider-captcha-lab Solves Alibaba Cloud slider captchas with YOLO plus trajectory generation at a 90.9% pass rate; 74 stars show automation demand. Could be a commercial anti-bot service, but compliance risk applies. The main risk is platform updates, with other captcha solvers as competitors.
Automatic Alibaba Cloud slider CAPTCHA solving: YOLO recognition + human-like trajectory generation + CDP injection, 90.9% single-attempt pass rate
B CUHK-X Provides a large-scale multimodal human motion recognition dataset and benchmark, filling a gap in action understanding and reasoning evaluation; 30 stars in 1.9 days shows academic interest. Could become a standard HAR benchmark, but commercialization is weak. Risks are dataset bias and competitors like NTU RGB+D and Kinetics.
[MobiSys 2026] A large-scale, multimodal dataset and benchmark for Human Action Recognition, Understanding and Reasoning
B Show HN: GET Together – A social network where you don't need POST to Post This uses HTTP methods as social actions, a novel idea, and 106 Hacker News points show developer interest. Commercialization could be a minimalist social product, but user growth is hard; the risk is that the concept outweighs practicality, and the competitor is traditional social networks.
B Show HN: HomeCat – Design your backyard office This designs backyard offices and automatically complies with building codes, solving the compliance challenge of DIY construction. With 39 Hacker News points momentum is modest; commercialization could be a subscription design service for the North American home-office market, but regulations are complex and competitors include Sheds Unlimited.
Hey HN! Ilya and Nikita here. We’re building HomeCat, a tool that lets you design a backyard office space - https://myhomecat.com We started out by helping people design sheds, and quickly realized th…
B SwiftMusic Lets Swift developers declaratively write music events, patterns and real-time state updates; 34 stars in 1.6 days show early interest in music programming circles. It could become a Swift audio dependency or join the AudioKit ecosystem, but commercialization needs existing community leverage; risks are the niche of Swift music development and competitors like AudioKit and MIDI frameworks.
Declarative music in Swift: patterns, musical events, and live update state.
B holo-card-studio It solves the problem of creative workers quickly generating editable 3D flashcards, but several similar skills already exist. Reaching 1,153 stars in two days shows heavy adoption by AI agent users. The commercialization path is a paid skill or template library in the Blender/Three.js ecosystem, but competition is fierce and official features could replace it; the risk is feature homogenization, with other holo-card skills as the strongest competitors.
Turn the user's description or uploaded reference into a finished, editable Blender card and an interactive Three.js page. Preserve the requested subject, style, typography and destination. This skill…
B 8086-xcheck-system An IP reputation investigation system that helps security analysts trace origins. Its 101 stars show demand; commercialization could be a SaaS security tool, but it needs data sources, and risks are data accuracy with VirusTotal as the competitor.
XCheck IP Reputation Investigation System
B UMR The official codebase for unified motion retargeting, solving animation transfer; 113 stars show research community interest. Could be a commercial animation tool, but needs academic transfer. The main risk is dataset dependence, with other retargeting algorithms as competitors.
The official codebase of Unified Motion Retargeting
B wechat-local-viewer This project makes local WeChat chat history browsable, searchable and exportable while running locally to preserve privacy, and 31 stars in 1.3 days suggest real demand. It could become a personal data management or enterprise audit tool, but risks include changes to WeChat's encrypted format, privacy compliance and an official ban.
WeChat local chat history viewer — conversation browsing/full-text search/multi-format export/optional LLM smart summaries, all running locally with zero external dependencies.
B note-bridge-desktop Addresses the pain of exporting and migrating from multiple note platforms to local/offline reading, supporting four formats and failure states; 24 stars in 0.9 days shows demand. Could be a knowledge management migration tool or enterprise data export service. Competitors include Obsidian's official importer and Joplin.
Windows note export and migration tool: seven-platform adaptation, four-format export, offline reading, with clearly defined verification scope and failure states.
B deckard Runs AI text detection locally in the browser so text never leaves the machine, directly addressing teachers' and editors' privacy concerns. The approach is more honest than wrapping another detection API, but it is currently a personal experiment with no public accuracy evaluation, so it cannot be compared with cloud services like GPTZero.
Local AI-text detection for your browser.
B nowinandroid-kmp This ports Google's Now in Android to Kotlin Multiplatform as a reference implementation, and 79 stars show KMP developers want it. Commercialization could be training or consulting, but it is itself a sample; risks include Google updates, and the competitor is the official KMP template.
📱 Now in Android KMP is a full Kotlin Multiplatform port of Google's Now in Android sample.
B refund-anything-ai-prompt This tool uses AI to generate refund letters covering 30+ jurisdictions, and 53 stars show consumers want it. Commercialization could be a refund service, but legal risk is high; the main risks are success rates, and the competitor is DoNotPay.
AI prompt that gets your money back — subscriptions, digital purchases, bookings. Polite letter first, legal escalation second. 30+ jurisdictions.
B awesome-remote-mcp-servers This project collects remote MCP servers so AI agents can connect easily, and 68 stars show the MCP ecosystem needs a directory. Commercialization could be a directory service, but maintenance costs are high; risks include dead links, and the competitor is the official registry.
A collection of remote MCP servers.
B robinhoodpools This is a Robinhood on-chain LP exploration, accounting and research terminal, and 50 stars show crypto users want it. Commercialization could be a DeFi tool, but it needs data sources; risks include Robinhood API limits, and the competitor is Dune Analytics.
Standalone Robinhood Chain LP explorer, accounting and research terminal
B Show HN: DriveSync – fast Git-styled Google Drive sync CLI This syncs Google Drive in a Git style, addressing cloud-drive version management pain points. With 17 Hacker News points and 29 stars momentum is modest; commercialization could be an enterprise cloud sync tool, but it must handle API quotas, the risk is Google's own tools, and the competitor is rclone.
B Show HN: LLM Attention Visualization This interactively visualizes LLM attention to help researchers understand models, and 147 Hacker News points show community interest. Commercialization could be a teaching tool, but the audience is narrow; the risk is existing similar tools, and competitors are Transformer explanation libraries.
B drivesync A Rust command-line Google Drive sync tool aimed at developers who want to script their backups. The topic is concrete and usage straightforward, but "Superfast" has no benchmark backing, rclone already covers the functionality more maturely, and it currently looks like an early lightweight alternative.
Superfast Google Drive CLI to sync files & folders
B Sung A Material 3 music player for Linux combining YouTube Music with local libraries and synced lyrics. The positioning is clear, but several similar players exist with no new approach, and the repo doesn't explain installation or account login, so there's no compelling reason to switch yet.
A native Material 3 music player for Linux. YouTube Music, local audio, synchronized lyrics, and Noctalia theming.
B odyssey A Quickshell desktop shell for Hyprland bundling system controls, launcher, notifications and wallpapers into one set. It looks fairly complete, but it's another desktop config heavily overlapping existing dotfiles projects, only useful to people who want to copy the config directly.
A contextual Quickshell desktop shell for Hyprland — system controls, launcher, notifications, media, theming, wallpapers, and more.
B ptx-optimization An NVIDIA PTX optimization skill for analysis, precise cubin patching, and verification, addressing low-level GPU performance tuning with direct value for AI training and inference. 10 stars is early but professional; risks are NVIDIA toolchain changes and competition.
Evidence-led NVIDIA PTX optimization skill: profiling, exact-cubin patching, validation, and measured tuning
B Show HN: Isle – managed application environments for computer-use agents Isle provides hosted application environments for computer-use agents, solving isolation and dependency issues during task execution. 12 points is early but promising; the path is agent infrastructure billed per environment, but the ecosystem is immature and it competes with container solutions.
B cck CCK is an account manager that batch-manages AI coding tool account quotas and launches with one click, solving developers' multi-account quota management, and 53 stars show demand. Commercialization could be a subscription service, but account sharing carries compliance risk; the risks are platform bans and grey areas.
CCK Account Manager — add custom accounts, batch-manage AI coding tool account quotas, and launch clients with one click
B chatgpt-model-meter Shows the 5-hour and 7-day quotas on chatgpt.com and compares your selected model against the model_slug actually executed in server-side telemetry, answering "was I downgraded?". The problem is real, the approach restrained, and it works once installed; limits are that it only applies to the web version and relies on internal page fields.
Chrome MV3 extension for chatgpt.com: shows 5h/7d quota, and checks the requested model against the model_slug the server's STE telemetry reports as actually executed.
B svg-trace-studio A single-file, offline bitmap-to-SVG tool that runs VTracer WASM in the browser and can export an animation of the tracing process, skipping the Python environment and dependency setup. Anyone vectorizing logos or line art can use it today; tracing quality is limited by VTracer and the project is still small.
Offline single-file tool: image to SVG (VTracer WASM browser-side tracing) + hand-drawn process animation export
B KeySniper It targets an unauthenticated account takeover in Keycloak's credential reset flow; if reproducible, it's a valuable security report directly relevant to teams doing identity authentication. The author is a known security figure, but the formatting is exaggerated, and the CVE number and actual impact still need independent verification.
**CVE-2026-18963** — unauthenticated Keycloak account takeover via the reset-credentials flow.
B declarative-attention A reproduction of the KAIST method letting language models control their own attention, by lucidrains, who tracks papers constantly and whose code usually runs directly. A low-cost entry for researchers trying new attention mechanisms; currently just a reference implementation with no training results backing it.
Implementation of the procedure in Language Models Can Control Their Own Attention, from Namgyu Ho et al. of KAIST AI
B uzbekistan-geocoder A self-hostable forward and reverse geocoding service for Uzbekistan built on Nominatim with Docker Compose and a Caddy reverse proxy, ready after install. Scope is clear and claims are modest; only data updates and performance notes are missing, and regional deployments like this are rare.
Self-hosted forward and reverse geocoding for Uzbekistan with Nominatim, Docker Compose, and an optional Caddy proxy.
B Pulse A backend-free PWA tracking GitHub release downloads and repository metrics, addressing developers' need for quick release analytics. 24 stars is early; the path is a developer dashboard competing with existing GitHub analytics, but features are basic and it depends on the GitHub API.
Pulse is a lightweight, responsive Progressive Web App designed to track, visualize, and analyze release downloads and repository metrics across GitHub. Built with vanilla JavaScript and styled with a…
B hyprquickpaper A keyboard-driven wallpaper picker for Hyprland written in Quickshell and QML, with a bottom card layout and custom fonts, restrained in scope. Wayland really lacks a handy wallpaper switcher; the downside is it only serves Hyprland users and has very few stars.
A fast, themeable, keyboard-driven Wayland wallpaper picker built with Quickshell and QML for Hyprland featuring a bottom-dock card layout and custom sci-fi typography.
B efficient-gaussian-appearance Adds a lightweight view-dependent appearance model to 3D Gaussian Splatting, with JIT compilation and a WebGL preview, claiming to match a paper on compact appearance modeling. The direction fits, but the repo is very new and whether code matches the paper still needs checking.
Faster-GS with JIT-compiled view-dependent appearance models and a WebGL viewer, described in the paper "Compact Neural Appearance Models for Efficient Gaussian Splatting"
B three-gtvbao Implements GT-VBAO ambient occlusion as three.js TSL nodes running on both WebGPU and WebGL2, a rare piece of graphics tech put into practice. Web 3D developers can borrow from it directly; it lacks performance comparisons and a demo page, so usability is one step short.
GT-VBAO ambient occlusion for three.js WebGPU and WebGL2 (TSL)
B STOP-X-AI A small tool that collapses AI-related posts in the X timeline, targeting the specific annoyance of information overload, with a single clear purpose that works once installed. The risk is that it is easily replaced by official filters or a self-built list; a ten-minute curiosity.
Collapses AI-related posts on X/Twitter
B pokemonlive A local Pokémon battle prototype that uses AI to generate battle cutscenes—an interesting idea, but still a toy experiment with no reusable tool form and no explanation of battle logic or generation pipeline. It hit 86 stars in just over a day; hype exceeds substance.
Local Pokemon battle prototype with AI-generated cinematics
B kouyu An English speaking practice skill with gentle correction and a local review workbench—a likable topic with restrained scope. But it depends on an existing agent platform, practice quality hinges on prompts and the voice pipeline, and the repo shows no evidence of results.
A guided English speaking practice Skill with gentle correction and a local review workbench
B imprint Copies one Omarchy machine's configuration verbatim to another, a lightweight approach to dotfiles sync, with restrained naming and a one-line description. The audience is limited to Omarchy users, and it overlaps functionally with tools like chezmoi.
Stamp one Omarchy machine onto another.
B gpui-omarchy A set of GPUI components for the Omarchy desktop style, from a well-known figure in the Ruby community, with restrained naming and scope. But it only serves Omarchy users and lacks complete examples, leaving very little reusable surface.
UI components for Omarchy system style
B factcircuit Aims at an auditable claim-evidence verification loop with tamper-proof provenance and staged retrieval, a direction that hits a real need in trustworthy AI. But the description is piled with jargon and shows no interface definitions or working examples, making it hard to tell a complete implementation from a concept draft.
FactCircuit: an auditable claim-and-evidence verification loop with immutable provenance, staged checks, bounded retrieval, and reproducible evaluation.
B real-api-pricing Calculates real API cost as subscription fee divided by usable tokens, with a Pareto chart, addressing LLM price comparison. 79 stars show developer cost concerns; it could be a SaaS comparison tool or data API, but competitors include OpenRouter and LMSYS, and prices need real-time updates.
Real API pricing: subscription fee ÷ usable tokens, with allowance, unit-price, and leaderboard Pareto charts.
B macos-sysdata Visualizes macOS system data usage and deletes items one by one, addressing storage anxiety; 200 stars per day show a real pain point. It could be a paid tool or system-level dependency, but competitors include DaisyDisk and OmniDiskSweeper, and accidental deletion of system files is a risk.
See what is really inside macOS System Data and delete it, item by item, from the menu bar.
B game-sensitivity-coach It turns FPS mouse sensitivity conversion and tuning into an Agent skill, enabling cross-game conversion and feel review in Claude Code and Codex. The audience is clear—people serious about aim training—and it gives steps rather than mysticism. The limitation is that results depend on the accuracy of built-in conversion tables; alternatives are various web eDPI calculators.
An agent skill for evidence-based mouse sensitivity tuning, cross-game conversion, and gameplay review. Works with Codex, Claude Code, and Agent Skills hosts.
B let-go-macos A small macOS utility that tells you directly which process is holding a file, folder, or disk, filling the gap between lsof and the system's "disk not ejected properly" prompt. The description is honest and the scope narrow, so ordinary users can just install and use it; the limitation is thin functionality, with similar paid software already available.
A lightweight, privacy-first macOS utility that shows what is holding a file, folder, or drive open.
B tailterm SSH in the browser, networked via Tailscale with encrypted credentials, plus tmux sessions and split layouts; handy for connecting home from another machine. Compared with ttyd and Wetty, the addition is Tailscale integration.
Browser SSH over Tailscale with encrypted credentials, tmux sessions, and tiled terminal groups
B Show HN: Mador – Make any DOM reactive with a tiny 80-line Proxy state tuple This implements DOM reactivity with an 80-line Proxy, solving native JS state-binding pain; HN's 58 points show demand for lightweight solutions. It could be packaged as an npm dependency embedded in frameworks, but competes with Vue, Svelte, and MobX, and its API ecosystem is unformed, so adoption is uncertain.
B dock-postprocess Standardized post-processing after molecular docking: importing poses, restrained minimization, conformation QC, and protein-ligand interaction analysis, also covering hard cases like PROTAC ternary complexes. Computational chemistry teams can run it directly; the limitation is an academic script style with no example data or installation instructions.
Open-source SBDD toolkit for standardized docking-pose import, restrained OpenMM minimization, pose QC, protein-ligand interaction analysis, ligand strain analysis, and integrated design prioritizatio…
B Routed A local Agent Skills router for environments like Claude Code and Cursor, solving the problem of each AI coding tool keeping its own copy of skills. The direction addresses a real current need, but there is only a one-sentence description and no config examples; whether it wins depends on how many formats it supports.
Universal local router for Agent Skills across AI coding environments
B cloudflare-turnstile-solver This claims to help developers integrate Turnstile into their own pages, but its name contains "solver" and it closely resembles another repo of the same name, suggesting automated star farming or verification bypass. If a legitimate integration library it is useful, but the risk is abuse and Cloudflare bans; the official SDK is the competitor, so be cautious.
A small Python library and CLI for working with Cloudflare Turnstile on your own pages - read the sitekey, build the token, verify the result.
B OMFAtlas Combines 643 BodyParts3D meshes with 77 self-added nerve, vessel, and gland structures into a 3D oral and maxillofacial atlas, clearly stating the data source and what was added. It is genuinely useful for dental teaching; the limitation is a narrow audience, and no anatomical review process is mentioned.
Interactive 3D atlas of oral and maxillofacial anatomy for dental education. 643 BodyParts3D meshes plus 77 schematic nerve, vessel, gland, sinus and joint structures the source dataset omits, fitted …
B SimdPaddleOCR A pure C# PP-OCRv6 inference library focused on SIMD acceleration and low memory, for offline OCR in .NET desktop and server apps. A real option for C# teams avoiding Python, though recognition accuracy still needs self-testing.
Pure C# PP-OCRv6 inference library: multi-platform SIMD optimization, low memory usage, high accuracy.
B t1bridge Brings Linux support for Apple T1 Mac keyboards, Touch ID and other hardware, a specific gap long left unfilled. The audience is the few installing Linux on old MacBooks, but once such drivers work they are the only option.
Linux support for Apple T1 hardware
B claude-pc-cleanup A Claude Code skill that scans macOS for reclaimable space, lists it by tier, and confirms each item before deletion. Useful for developers constantly short on disk who dare not clear caches blindly; its sense of restraint is clearly better than one-click cleanup scripts. The limitation is macOS-only, and it's essentially common cleanup commands wrapped in a reviewable flow.
Claude Code skill: scan a macOS machine for reclaimable disk space and clean it up safely, tier by tier, with explicit confirmation before deleting anything.
B cn-k12-math-knowledge-graph Organizes prerequisite relations in Chinese K12 math into an open graph with a 3D interactive browser, with both data structure and presentation clearly explained. Teachers and edtech products can use it directly; the limitation is that it covers only math, with no word on textbook version alignment or update method.
Open prerequisite graph and interactive 3D explorer for China K12 mathematics
B omavoice A voice recorder for Omarchy desktop users, wired to whisper.cpp for real-time transcription with GTK4 and PipeWire; small and clear in scope. Narrow audience, but it works once installed, more solid than most in this batch.
Voice recorder with live whisper.cpp transcription for Omarchy (GTK4, PipeWire)
B Show HN: Kadō – open-source habit tracker, with non-binary habit score, for iOS This is a privacy-first iOS habit tracker that replaces check-ins with non-binary ratings; HN's 66 points show the product design resonates. It could go paid or subscription on the App Store, but competes with Habitica and Streaks, and open-source monetization needs value-added services.
B pigeon An authorization credential layer defining what AI agents may do, moving permissions out of prompts. Timely direction, but only one sentence of explanation, no integration path or threat model, and its difference from existing agent authorization schemes is unverified.
Delegated authority for AI agents. A Pigeon Pass says what an agent may do.
B remedi Uses FDA data to find cheap prescription alternatives and patient assistance programs, plus income eligibility checks; a real pain point for uninsured Americans and a rare direction. Risks are data freshness and the boundary of medication advice.
Find affordable prescription alternatives and patient assistance programs in seconds using FDA data and income eligibility checks.
B Sol A native Swift macOS music player supporting local libraries and Subsonic/Navidrome, emphasizing bit-perfect output, plus podcasts and metadata editing. For those wanting both audio quality and a self-hosted streaming library, it's lighter than Electron-wrapper alternatives. The limitation is being a one-day-old repo with broad features, where Feishin and Supersonic already exist.
Native macOS music player for local libraries and Subsonic/Navidrome servers, with bit-perfect output, podcasts and a metadata editor.
B mador A tiny library that makes any DOM reactive, a clever idea aimed at people who want to add a little interactivity to existing pages; but the README is one line, with no API or size given, and Alpine and Solid already cover similar ground.
Make Any DOM Reactive
B ALIGN-Agentic-Loop-Image-GeneratioN Neither diffusion nor autoregressive: a coding agent repeatedly draws in p5.js and self-critiques to produce images, with Claude Code and Codex skills. Novel approach, but output quality and cost lack evidence; best viewed as a method experiment.
Agentic loop image generation without diffusion or autoregressive image models. Coding agents draw in p5.js, using iterative refinement and adversarial visual review for high-fidelity art and method f…
B PopSilicon It gets 32-bit PopCap classics like Peggle, Peggle Nights, and Bejeweled 3 running on Apple Silicon Macs, solving the real pain of "can't play old games on a new Mac." The scope is tiny but the goal clear. The limitation is reliance on a compatibility layer, and actual experience and legality need personal confirmation.
Play Peggle Deluxe, Peggle Nights, and Bejeweled 3 on Apple silicon based Macs!
B fileblade A customizable IDE-style file sidebar for the Omarchy desktop, hitting the pain point that tiling window users lack a file tree, with a concrete idea. But the audience is limited to Omarchy/Hyprland, and it does not explain how it coexists with existing file managers.
Customizable IDE-like sidebars for managing files and referencing information in Omarchy.
B RISCV_computer_emulator_running_Linux_in_1750_lines_of_Odin A 1750-line RISC-V emulator in Odin that can run Linux, with the author aiming for readable teaching code and neither the naming nor the description overselling. Worth reading for anyone wanting to understand RISC-V and Linux boot; the limitation is only 8 stars, with no tests or performance data.
A surprisingly fast and simple, yet really cool 64 bits project to learn from.
B evvdroid Ports the Eloquence speech engine to Android and Wear OS based on openevv; a rare choice for users needing old-style TTS or accessibility reading. Installation and licensing steps still require figuring out yourself.
Eloquence text-to-speech engine for Android and Wear OS, built on openevv
B qanat Organizes alpha factor construction and backtesting as a DAG and lets agents participate in generation, fitting a new style of quantitative research. But "agent-native workflow engine" is too broad, with no mention of data sources, execution cost, or how it relates to frameworks like qlib.
Agent-native workflow engine for building and backtesting alphas as DAGs.
B youjie-math-lab An interactive problem site for high school math olympiad, with three labs featuring step-by-step proofs and printable materials; the description is honest and the scope restrained. But it is essentially self-made teaching content overlapping Khan Academy and GeoGebra classroom materials, with no reuse value for non-target students—a small but complete personal teaching aid.
有解: High school olympiad math interactive lab with three lesson experiments, step-by-step proofs, exercises, and printable handouts.
B css-modules-content-mapper A small tool providing a TS content mapper for CSS Modules, selecting styles by content rather than class name; the author has some credibility in the type ecosystem. Very narrow audience, and without examples it is hard to judge fit.
TypeScript content mapper for CSS modules
B OpenWAM An open-source modular approach to world-action model pretraining, the paper's official implementation, aimed at researchers in embodied AI and video prediction. Publicly reproducible skeletons in this direction are still rare, so it is worth a look; the usual paper-repo caveat applies, as whether code and weights are complete and runnable still needs verification.
Official repository for "OpenWAM: An Open, Modular Exploration Towards Systematic World–Action Model Pretraining".
B EmbodiSteer Code accompanying a CoRL paper that uses joint-space guidance to deploy visuomotor policies zero-shot across embodiments, with a clear problem definition. But it stays at paper level, missing weights and simulation environment details; robotics researchers can read it, but engineering deployment is still far off.
[CoRL 2026] EmbodiSteer: Steering Embodiment-Agnostic Visuomotor Policies with Joint-Space Guidance for Zero-Shot Cross-Embodiment Deployment
B vistep Turns algorithm or code execution steps into interactive models with bilingual explanations and narration, a decent idea; but "visualize every step" is too vague, with no word on supported scenarios or generation quality. Forty-five stars in two days, lacking examples.
Visualize Every Step with AI — bilingual visual explanations, interactive models and synchronized narration.
B secant A notebook arguing that RoPE attention with full softmax is equivalent to one exact gradient step. If true, the claim would shed light on positional encoding, but for now there is only an assertion, with no organized proof or experimental validation, so reading it means checking the author's work.
RoPE attention is an exact forward-pass gradient step with softmax intact
B ExAt-Any-Ionogram Uses a vision-language agent to zero-shot extract F-layer traces from ionograms, a concrete problem with real value for ionospheric observatories. But it is research code with only 10 stars, and neither the dataset nor baselines are mentioned, so direct reuse is hard in the short term.
Zero-shot ionogram F-layer trace extraction with vision-language agents (ExAt)
B MACPulse-widget A native macOS widget showing CPU, memory, disk, power and battery; cleanly done but heavily overlapping Stats and iStat Menus. Unless you specifically want the widget form, there is no reason to switch.
Native macOS desktop widget for viewing CPU, memory, and disk usage plus power and battery info, built with SwiftUI and WidgetKit.
B unidl A terminal media browser and downloader, positioned close to yt-dlp plus a file-browsing interface. The need is real but similar tools are numerous, and the description does not say which sites are supported or how logins are handled, so the differentiator is unclear.
UniDL: a terminal-first media browser and native downloader
B ram0verflow A toy chain treating the README as a ledger and using subset sum rather than hashing for proof of work; the idea is genuinely uncommon. But it is a personal experiment with no usable interface or docs, so it is only a curiosity.
A small Bitcoin-like chain whose canonical ledger is my README. Proof of work is subset-sum, not hashing.
B launch A lightweight DOS boot menu written in C for people still tinkering with DOS games and FreeDOS; restrained scope and direct implementation. But the feature set is very narrow, and existing DOS menus suffice.
A lightweight and simple launcher menu for DOS systems
B ActivityMonitor A native macOS activity monitor covering CPU, memory, energy, disk and network with light and dark themes; decently complete. But it directly collides with the built-in monitor and Stats, lacking a must-use reason.
A native macOS activity monitor for CPU, memory, energy, disk and network, with light and dark themes.
B codenotch A macOS app that pins usage limits for Claude Code, Cursor, Codex and other AI tools to the screen edge, easing developer anxiety over AI quotas. 220 stars in 0.4 days show a new pain point; monetization is subscription or partnerships with AI vendors, with risks from macOS restrictions and built-in usage panels.
A macOS app that pins usage limits from Claude Code, Cursor, Codex, and Antigravity to a screen edge.
B actual-fingerprints Deterministically generates synthetic fingerprint data from seeds with zero-dependency Node, useful for creating reproducible test and demo samples that don't involve real users. But it doesn't specify whether "fingerprint" means browser, device, or biometric, and the audience is narrow. A clean little tool worth a glance.
Deterministic synthetic fingerprints from a seed. Zero dependencies, Node.
B lego-mac-compat It ports LEGO games that TT Games released only for 32-bit Mac to Apple Silicon, addressing the concrete problem of old games not running on new Macs; the technical approach and compatibility range still need verification, but it has clear value for nostalgic players.
Port of TT Games’ 32-bit-only LEGO Mac games to Apple Silicon
B human-atlas An open-source 3D human anatomy browser with 2,234 selectable meshes and layered systems, serving medical education and biology visualization. 333 stars in one day show strong visual appeal; monetization is education licensing or medical training SaaS, with risks in data licensing and WebGL performance against commercial platforms like BioDigital.
Open-source 3D anatomy explorer: 2,234 selectable BodyParts3D meshes, system layers, search, and exploded views.
B warp-masque-actions A GitHub Actions workflow that auto-generates Cloudflare WARP MASQUE keys and mihomo configs, automating proxy setup. 260 stars in 1.3 days show strong demand; monetization is as a companion tool for proxy providers, with risks from Cloudflare policy changes and compliance against one-click WARP scripts.
GitHub Actions automatically generates Cloudflare WARP (MASQUE) keys and mihomo configs, with artifacts stored only in artifacts
B zork-underground-empire It remakes Zork I as a first-person 3D dungeon in the browser, from AI researcher Mollick, clearly an experiment in "remaking classics with AI." Worth opening for those who want to play the old text adventure and see how far AI-assisted 3D can go; the limitation is that it is a demo with unknown depth, and Zork remakes are already plentiful.
A first-person 3D adaptation of Zork I. Explore the Great Underground Empire in your browser.
B short-video-generator-AI Automatically turns long YouTube videos into viral short clips with highlight detection, subtitles, translation and voiceover. 180 stars in 0.5 days show strong demand for content automation; monetization is creator subscriptions or API, with copyright risks and rivals like Opus Clip.
Free open-source project designed for turning youtube-viedos into viral short videos. Highlight detection, subtitles, translation, voiceover, all in one for your content.
B stop-stutter Automatically raises AWDL priority on macOS to smooth game streaming via Moonlight and GeForce NOW. 126 stars in 1.1 days show a real pain point; monetization is a game streaming optimization tool, with the risk that macOS updates may break it against other network optimizers.
Smoother game streaming on Mac. Automatic AWDL boost for Moonlight, GeForce NOW, Punktfunk, Parsec & Steam Link. Native SwiftUI + Liquid Glass. MIT.
B astra_piNas A six-layer CM5 Lite dual-NVMe carrier board with a PCIe switch, PoE, and USB-C PD, with design files public and an honest note that prototyping is paused. Hardware fans wanting to build a NAS or edge box can review and modify the board; the limitation is that it is an unverified design, and mature CM5 carrier boards already exist as ready options.
Six-layer CM5 Lite dual-NVMe carrier with PCIe switch, Ethernet, microSD, USB-C PD and PoE. Review design; fabrication hold documented.
B hermes-zvec-memory A local-first memory backend for the Hermes agent that runs hybrid BM25-plus-vector RRF retrieval over a Markdown library, with a clearly stated scope. A good fit for people who want data kept local and find vector databases too heavy; the downside is being tied to Hermes and zvec-grep, so switching frameworks means starting over.
Local-first Hermes memory provider on zvec-grep (hybrid BM25+vector RRF over Markdown vault)
B Get-ServerFullReport A PowerShell script that inventories Windows servers as an as-built list and runs health checks, outputting a single-file report; practical for ops with restrained scope. Not novel, but more honest than a pile of platform-type projects.
Inventario as-built y health check de servidores Windows en un solo .ps1
B storiesclient A third-party Android client for Meta Ray-Ban Stories smart glasses, focused on removing Meta from the process; clearly valuable to glasses owners and privacy-sensitive users. But it relies on private protocols, so stability and legal risks are on you.
Third party Android client for Meta/Facebook Ray-Ban Stories (stella) smart glasses. Get Meta out of your face!
B taierspeedtest A Linux client for the global network test / Taier speed test, filling the gap left by official mobile-only or web-only offerings; practical for Linux desktop users and network troubleshooting. The scope is narrow, but you need to confirm whether test nodes and protocols are public.
Global network test / 泰尔测速 Linux client
B mini-harness A readable Python agent harness with tool calls, context compression, retries, and a single-file TUI, positioned as a baseline. Suited to people who want to understand the agent main loop without wading through a big framework; honestly labeling itself small but complete is a plus. The limitation is that teaching harnesses are already plentiful, and 9 stars hardly show adoption.
A small, complete Python agent harness: tools, compaction, retries, and a single-file TUI. Built to learn and use as a baseline.
B NodePulse It turns multi-protocol node aggregation, subscription distribution and speed testing on the Pterodactyl panel into a two-file, zero-dependency Node service with Telegram control, so airport or personal node operators can get started in ten minutes. The idea is not new (subconverter and the like came earlier); its strength is restrained scope. The limitation is that it covers only a single panel ecosystem, and long-term maintenance rests on one author.
Build multi-protocol node aggregation, subscription distribution, and a real speed-test panel with the Pterodactyl panel. The whole project has only two files, `index.js` and `package.json`, with zero…
B deedchain This benchmarks whether browser agent reports truthfully reflect their behavior, addressing agent trustworthiness; 49 stars in 1.1 days. Monetization is an agent evaluation service, but benchmark design is hard and it competes with various agent evaluation frameworks.
Benchmark of report fidelity for browser agents: does the agent's final report tell the truth about what it actually did?
B HuntersRecomp It recompiles Metroid Prime Hunters into a native Android port for the dual-screen AYN Thor, with dual sticks and HD GLES rendering; the target device is very narrow, but the technical path is concrete and suits those following NDS and Android porting.
Metroid Prime Hunters recomp for the AYN Thor: native dual-screen Android port with twin-stick Prime controls and HD GLES rendering
B MCP-Server-Aster-DEX A read-only MCP service for Aster DEX that explicitly states no private keys, no trading, and unofficial status—boundaries clearer than most similar projects. But wrapping an exchange API in MCP is now routine, and the "points and airdrop eligibility" pitch makes first-day stars look more like airdrop hunters than real users.
MCP server for Aster DEX — a Model Context Protocol server that gives Claude, Cursor and Cline structured access to Aster DEX market data, volume, points and airdrop eligibility via natural language. …
B agent-skiller It builds, step by step and visually, skills an agent can follow, saving hand-written config. Aimed at teams that want to lock agents into workflows; but the skill-builder slot is crowded lately, and whether it wins on template quality and export format is unclear, with 30 stars a day no proof.
agent-skiller is an open-source visual builder for skills an AI agent can follow, step by step.
B YawNet Does full-circle head yaw regression with biternion and von Mises κ, plus DINOv3 teacher-student distillation on purely synthetic 42k data; the method is concrete with an experimental design, suitable for pose estimation researchers to reproduce, but the application is narrow and depends on synthetic data.
YawNet: Full-circle head-yaw regression via biternion (cos/sin) + von Mises κ, DINOv3 teacher-student distillation on a purely synthetic 42k dataset.
B omavoice A virtual microphone for Omarchy with noise reduction and echo cancellation, directly serving Zoom/Meet and podcasts. This need genuinely exists on Linux, but PipeWire filter chains and NoiseTorch already cover most of it; with 9 stars and ties to a single desktop distribution, it is worth waiting.
Virtual microphone for Omarchy: noise cancellation and echo cancel for Zoom, Meet, and podcasts
B Taara An RNA compiler built on MLIR that turns topological constraints into a compilation problem, an angle rarely seen in bioinformatics tooling. Worth ten minutes for RNA design work or anyone curious about MLIR outside traditional domains; the limitation is a one-sentence description, unknown maturity, and a long way from production pipelines.
topology-aware RNA compiler built on MLIR
B omarchy-widgets It provides a drag-and-drop widget grid on the wallpaper for the Omarchy desktop, a small and specific scope; it is a handy desktop enhancement for Omarchy users, but the audience is limited to that distribution or configuration, so it is not very general.
Desktop widgets for Omarchy: a themed grid on your wallpaper, arranged by drag and drop.
B ghunt-panel Wraps GHunt, a Google account OSINT tool, in a local Flask web interface so you don't have to remember CLI arguments; handy for account investigators, but it's essentially a thin wrapper that must track upstream interface changes, and it raises privacy ethics concerns.
A local, self-hosted web UI for GHunt — mxrch's OSINT framework for investigating Google accounts. ghunt-panel wraps the ghunt CLI in a small Flask app so you can run lookups, read the results, and ma…
B stylesmuggler-mitigation Emergency mitigation for an unpatched RCE in Magento/Adobe Commerce: web server rules, a CLI guard, and compromise scanning. E-commerce operators need such plug-and-play protection, but the cited dates and the 8-star scale are hard to verify, so confirm the vulnerability is real before deploying to production.
Emergency mitigation for StyleSmuggler, the unpatched Magento/Adobe Commerce RCE (Sansec, 2026-09-05). Web-server rules + CLI guard + compromise scanner.
B Forbidra A 401/403 bypass and endpoint reachability testing tool for bug bounty work. People doing authorized testing will want to try it, but bypass scripts are plentiful and it does not say which techniques it covers or how success is judged; 21 stars a day is not enough to prove effectiveness.
A bug bounty tool for bypassing 401/403 access restrictions and testing endpoint accessibility.
B OFXR-Bridge Uses optical flow for VR frame generation, a real need long covered only by vendor solutions like ASW, with few open-source implementations. 34 stars in a day shows people are waiting; the limit is a one-line description that doesn't say which headsets and runtimes are supported, so real usability is unknown.
Frame generation for VR using Optical Flow
B call14 It uses an ESP8266, MQTT, Home Assistant, and visual recognition to build a $20 local-first elevator call and tracking setup. DIY smart home fans will find it interesting, but it is tied to one's own building equipment, is not general, and involves property management and compliance; similar needs are usually met by generic HA integrations.
A $20 local-first smart elevator call and tracking system built with ESP8266, Home Assistant, MQTT and computer vision.
B psa It uses Token-2022 confidential transfers to hide balances on pump.fun tokens and requires deploying your own SPL Token Wrap, an unusual approach in the Solana space. But the audience is very narrow, it depends on self-deployed contracts, and the docs do not clarify keys or audit status, so for now it only suits developers studying confidential transfers.
Pump Shielded Assets: hidden balances for pump.fun coins via Token-2022 confidential transfers and a self-deployed, immutable build of SPL Token Wrap
B flight-over-the-world A browser geography game where you fly over real satellite terrain and guess locations, combining GeoGuessr with free flight. It appeals to geography game fans, but the one-line description does not clarify the data source or gameplay depth; GeoGuessr and its clones are already numerous, so 16 stars in a day is not a signal.
Fly over photorealistic Earth – guess the region, find your way home, or just explore.
B nexus-clipboard A local-first clipboard manager for Windows built with Rust and Tauri 2, emphasizing speed and privacy. The need is real, but this slot already holds Ditto, CopyQ, and the built-in history; the description does not explain its unique trade-offs, and 14 stars show no differentiation.
A blazing-fast, private, local-first clipboard manager for Windows built with Rust & Tauri 2.
B microduck-studio A local web workbench for the Microduck robot, for developing, debugging, visualizing, and safely orchestrating reinforcement learning pipelines. Teams doing small robots plus RL may benefit directly, provided you own this machine; generality and documentation depth are not yet visible.
Local web studio for developing, debugging, visualizing, and safely orchestrating Microduck robot and reinforcement learning workflows.
B contenthub Turns one piece of material into an article, carousel, and Reel at once, with a "fact-check" gate in the pipeline to stop copy from overreaching. That gate is more honest than similar content pipelines, but it's still a content-reprocessing tool lacking sample output and integration docs, so its value depends on your existing writing workflow.
Uma ideia vira artigo, carrossel e Reel — pela mesma esteira, com gates que impedem a peça de afirmar o que a apuração não sustenta.
B OsuFinder Helps osu! players download beatmaps more conveniently; the need is real and the scope narrow. But several such downloaders already exist, and the project doesn't explain how it differs from the official site or existing tools, nor mention API compliance or batch capability—usable but not urgent.
This is an application for easier downloading your beatmaps on osu!
B A-Label-Efficient-Model-for-Automatically-Extracting-Electron-Density-Profile-from-Ionograms A label-efficient model that automatically inverts electron density profiles from ionograms, a very specific geophysics ML task. Researchers doing ionospheric sounding who want to label less data may find it useful; but the repo name is just the paper title and data and training instructions are missing, so for now it is reference code only.
A Label-Efficient Model for Automatically Extracting Electron Density Profile from Ionograms
B sparkglm Records vLLM/EXL3 serving configuration and inference experiments for GLM-5.3-Flash on two DGX Spark machines; the goal is clear and reproducible. But the hardware bar is extremely high, only people with the same machines can use it directly, and it's more experiment notes than a general tool.
GLM-5.3-Flash on two DGX Sparks: opinionated vLLM/EXL3 serving, tinyGLM gates, and reproducible inference research.
B dualsense-controller Recreates the PS5 DualSense controller in the browser with Three.js, supporting button animations and real controller input; the work looks fun and suits interactive demos or teaching, but practical use is limited—more a polished toy.
Interactive PS5 DualSense recreation with Three.js, moving controls, gamepad input, and locally bundled assets.
B audio-locked-paper-theatre A narrated video pipeline that locks commentary to paper-cut collage visuals, with a deterministic quality-check gate. People making knowledge short videos and wanting batch output may be interested, but it is more like a personal workflow plus Codex skill configuration, with a high barrier for others to reuse.
Narration-locked paper-collage explainer workflow and Codex Skill with deterministic QA gates
B on-device-3d-scanner Generates Gaussian splatting 3D scans fully offline on iPhone from 8 photos in about 3 seconds, solving mobile 3D scanning's cloud dependency. 28 stars show AR creator interest; a consumer 3D scanning app could commercialize, but LiDAR scanning and Polycam compete, and accuracy and performance are the risks.
3D scanner running fully on-device: 8 photos to a spinnable gaussian-splat scan in ~3s on an iPhone, via Depth Anything 3 + react-native-executorch + TypeGPU
B plain-prose An Agent skill that removes AI writing traces from English, Russian and German, solving AI-generated text detection. 66 stars show content creator demand; a writing plugin or API could commercialize, but Grammarly and AI detectors compete, and language coverage is limited.
Agent skill that removes AI writing patterns from English, Russian and German prose. Merges stop-slop and avoid-ai-writing, adds a zero-dependency checker.
B Show HN: Open-Source eInk Bike Computer An open-source eInk bike computer using ESP32 to implement the ANT sensor protocol, solving cycling data collection hardware costs. HN 216 points show geek approval of its AI-assisted reverse engineering; hardware kits or ANT stack licensing could commercialize, but the open hardware market is small, and Garmin and mature bike computers compete.
Hey all, i just launched my Eink Bike computer project and think it is cool. Another tidbit, in the crazy things that AI has done... It has helped create a ANT (common sensor wireless protocol used in…
B WasmQuay Offline decodes WebAssembly components and classifies host capabilities, gating them by policy to solve WASM supply chain security. 36 stars in two days show security community interest; CI security scanning could commercialize, but wasm-tools and official tools compete and the ecosystem is immature.
Offline WebAssembly component inspection and capability-policy toolkit - decode .wasm binaries, classify host capabilities, gate against explicit policies. Rust + TypeScript, zero dependencies.
B DLSS-NR-on-AMD Runs NVIDIA DLSS 5 neural rendering on AMD GPUs by reverse engineering around hardware limits, solving DLSS unavailability for Radeon users. 444 stars in two days show strong demand; commercialization is nearly zero as a gray hack, and the biggest risks are NVIDIA legal action or driver updates breaking it.
Run DLSS 5 Neural Rendering on your AMD GPU. Magic!
B paddock A native inference server in Rust compatible with OpenAI and Anthropic APIs and multiple quantization formats, solving self-hosted model deployment. 35 stars show developer interest; enterprise private deployment could commercialize, but vLLM and Ollama compete, and ecosystem maturity is the risk.
Native Rust inference server for open models on NVIDIA GPUs. OpenAI- and Anthropic-compatible APIs, GGUF + safetensors, FP8/NVFP4/MXFP4/Q8/Q4, built-in Studio
B my-bad Beautifies dev server error pages with live updates, solving hard-to-read error messages during debugging. 58 stars come from a known author, so momentum is credible; a framework plugin could commercialize, but Vite and Next.js built-in error overlays compete, and ecosystem lock-in is the risk.
Beautiful dev-server error pages that live update. ✨
B kugiri Splits text into lines, words, and characters at the browser's actual line-break positions, solving the long-standing guesswork in subtitles and per-character typesetting, doing one thing with an accurate description. The downside is narrow applicability; only people doing text layout will think of it.
Splits text into lines, words and characters exactly where the browser already broke it.
B ps5_kernel_loader An IDA Pro 9.x loader for PS5 kernel dumps, ported from the PS4 tool of the same name, by a well-known scene figure who honestly notes Claude's involvement. The audience is extremely narrow but the thing is real; aside from console firmware researchers, basically no one can use it.
ps5 kernel loader for ida pro 9.x . created by myself and claude, based on socratic bliss's ps4 one
B Penelopa.ai Analyzes Codex and Claude Code sessions to extract repeated workflows and turn them into reusable skills, improving AI coding agent efficiency. 32 stars show team demand; an enterprise agent optimization platform could commercialize, but LangSmith and others compete, and data privacy is the risk.
Continuous improvement for AI coding agents: Penelopa analyzes real Codex and Claude Code sessions, finds repeated workflow patterns, and turns them into reusable skills, checks, prompts, and recommen…
B PromptLedger Provides version management, deterministic comparison snapshots and experiment reports for prompt templates, solving irreproducible prompt iteration. 42 stars show AI engineering demand; a team collaboration tool could commercialize, but LangSmith and Promptfoo compete, and ecosystem integration is the risk.
PromptLedger - version prompt templates, run deterministic comparison snapshots, score candidate responses, and ship diffable experiment reports. Python core plus a Go companion CLI.
B CacheCraft Replays request traces to test different semantic caching strategies and quantify real gains, solving hard-to-evaluate LLM cache effectiveness. 36 stars show AI cost optimization demand; a cache strategy library or SaaS could commercialize, but Redis and LangChain caching compete, and benchmark representativeness is the risk.
Semantic cache strategy bench: replay recorded request traces through candidate caching policies and measure which one actually pays off.
B unpeel A Swift terminal app plus a Rust multiplexing server, aiming to keep AI agent sessions in the terminal alive without dropping. The direction is concrete and more substantial than a pile of empty agent-orchestration scaffolds; but the repo is under a day old, usage docs are thin, and it goes head-to-head with tmux and claude-squad-style solutions.
A terminal app & rust multiplexer server for always-on terminal AI agents
B omarchy-workspace-layout Gives Hyprland per-workspace remembered tiling layout ratios, adjustable by dragging dividers, solving messy layouts across multiple monitors and workspaces. 99 stars show Linux user approval; commercialization is hard but it could become a standard Omarchy ecosystem plugin, with Hyprland interface changes as the risk.
Per-workspace tiling ratios for Hyprland, on Omarchy. Drag the dividers and your windows move — 25/50/25, master and stack, thirds, grid. Named layouts, per-workspace assignment, switchable profiles.
B packslip A Rust tool that signs manifests for vendored binary releases, by the mise maintainer, similar in approach to aqua's verification manifests. Useful for supply-chain verification in build and CI scenarios; still a prototype with thin documentation.
A signed release manifest for vendor binaries
B adaptive-frame-extractor A C++ tool for adaptive frame extraction in SfM, Gaussian splatting, and photogrammetry, picking keyframes by scene change to save manual image screening. The niche is specific and usable right away in 3D reconstruction pipelines; similar scripts abound, so it still needs effect comparisons.
Adaptive video frame extraction for SfM, Gaussian Splatting, and photogrammetry.
B pii Detects and classifies PII locally with a single neural program, solving data anonymization and privacy compliance. 33 stars show privacy computing interest; an SDK or compliance tool could commercialize, but Microsoft Presidio competes, and accuracy and coverage are the risks.
Detect and type PII locally with one ProgramAsWeights neural program.
B Qwen3.8-Flash-Next-Single-DGX-Spark Runs Qwen3.8-Flash-Next with TP=1 on a single DGX Spark, exploring the limits of running large models on small hardware. 67 stars in two days show strong edge AI interest; monetization could be a hardware reference solution, but the audience is narrow against NVIDIA's official optimization stack.
Qwen3.8-Flash-Next on ONE DGX Spark (TP=1)
B paper-engine Fixes a rigid B5 layout for AI-generated books, one concept per page, sharing the same CSS for screen and print. The idea is concrete with clear boundaries, suited to those who really want a printed book; but the readership is small, and versus general typesetting it just hardens the constraints.
Write one page at a time. Get a real book. A strict B5 canvas for AI-authored books: one concept per page, the same CSS on screen and in print, diagrams written as SVG.
B EdgeMosaic A small evaluation tool for edge AI: validating manifests, running benchmark plans, aggregating latency/memory/temperature, and producing reports. It saves effort for mobile or embedded inference work with a clear niche; the limitation is an immature ecosystem and non-standardized report format.
Rapid on-device AI evaluation lab: manifest validation, benchmark plans, latency/memory/thermal aggregation, JSON+Markdown reports.
B gitgui A pixel-rendered Git GUI for the kitty graphics terminal, a single Rust binary that works over SSH, with a narrow niche and an honest description. Great for people living in remote terminals who find plain TUIs hard to read; the hard limit is requiring kitty.
A Git source-control GUI rendered as pixels inside kitty-graphics terminals. One small Rust binary, no browser engine, works over SSH. cmux+pi+gitgui = 😍
B fermats-last-theorem A Lean repo under the official Anthropic account, its name pointing straight at formalizing Fermat's Last Theorem. If real, it's big news for the proof-assistant community; but beyond the name there's almost no explanation, so completeness is unknowable and it's only worth bookmarking for now.
B bg-gone Fully local background removal that runs in browser and CLI without uploading images, with a crisp description. But mature options like rembg and imgly already exist, and it doesn't explain model provenance or accuracy differences, making it a replaceable small tool.
Remove image backgrounds locally, in the browser or from the terminal. Nothing is uploaded.
B zen-sidebar-bookmarks A CSS module that tucks the bookmarks toolbar into the Zen browser sidebar—single purpose and honestly described. For Zen users it's a five-minute small fix, but it's just a style patch, and similar userChrome snippets already exist in the community.
A CSS module that allows you to display the Bookmarks Toolbar in the Zen Browser sidebar.
B rivet-durable-streams A native Rust implementation of the Durable Streams protocol for Rivet Actors, from the protocol's own authors, valuable for those building long-connection streaming services. But the ecosystem is very new, so only teams already using Rivet will find it useful.
A native implementation of the Durable Streams protocol for Rivet Actors
B uno A research repo claiming lossless acceleration of LLM inference via discrete diffusion. The idea has indeed been rare these past two years and is worth watching; but it only offers conclusory description without experimental setup or baseline comparisons, so treat it as a preprint for now.
Unlocking Lossless Speedups in LLMs via Discrete Diffusion
B ccbp-pytorch lucidrains' PyTorch reproduction of Continuous Continual BackProp, a convenient starting point for those tracking continual learning papers. It's his usual pipeline output with no extra engineering or evaluation, only worth a casual look for researchers.
Implementation of the proposed CCBP (Continuous Continual BackProp) in Pytorch
B NtfyChat A Kotlin client that forces a chat system onto the ntfy push channel; the idea is interesting but essentially uses a notification service as a message bus. Demo value outweighs practical use, and with the official ntfy client and mature IM apps for daily chat, it is hard to convince people to migrate.
A chat system based on ntfy
B PKMN-ED Extracts Gen 6/7 Pokémon models, textures, and animations from 3DS ROMs and converts them to .bch for SPICA, a very specific niche. Useful for the modding community, but built on pk3DS.Core with a narrow audience, so general developers can skip it.
A lightweight C# utility built on top of pk3DS.Core to seamlessly extract Gen 6 and Gen 7 Pokémon 3D models, textures, and animations directly from Nintendo 3DS ROMs. Designed specifically to generate…
B collections2mo2 A script that converts Nexus Collections into a form usable by Mod Organizer 2, directly useful to Skyrim/Fallout modding players with a clear niche. The downside is tight coupling to Nexus and MO2 version changes, high maintenance cost, and an audience limited to the modding scene.
Tools that convert Nexus Collections to Mod Organizer 2
B taskbar-fluent-media-player-Netease-Twilight Embeds NetEase Cloud Music playback controls, lyrics and cover art into the Windows 11 taskbar, with support for a specific skin, very vertical. A must-have for people using that exact combo, but narrow in coverage and easily broken by upstream client changes.
A Windows 11 taskbar player for NetEase Cloud Music and Twilight Echo, supporting synced lyrics, playback controls, cover art, audio visualization, and heart favorites.
B search-mcp A free web search and scraping MCP for agents, hosted and requiring no API key, with a low barrier to entry. But Tavily, Exa, and Brave already have many implementations here, and it doesn't clarify index sources or rate limits.
Free web search and page fetching for agents. Hosted MCP, no API key required.
B FalconFlank This repository discloses a Crowdstrike Falcon privilege escalation 0day, extremely valuable for security research. 409 stars in three days show very strong momentum; integrating into attack-defense toolchains or selling to vendors could commercialize, but legal and ethical issues are risks, and Crowdstrike may patch quickly.
Crowdstrike Falcon 0day Privilege Escalation Vulnerability
B Show HN: The cheapest GPU cloud – H100s at $2.04/HR, H200s at $3/HR Offers low-cost H100/H200 cloud GPUs, directly addressing AI training and inference cost pain points, and drew exposure through high-scoring HN discussion. Monetization is obvious, selling compute or acting as an aggregated price comparison platform, but risks are price wars when supply is ample and cost cuts from major cloud providers.
B fable-orchestrator Orchestrates multiple LLM models with Shell to complete tasks together, addressing the limits of any single model, and gained 320 stars in three days with strong momentum. Monetization could be agent infrastructure, but the name references a possibly unreleased GPT-5.6, risking unstable model APIs or being a pure stunt.
Fable 5.1 orchestrates. GPT-5.6 Luna and DeepSeek V4 Flash implement.
B ai-evaluation-framework Benchmarks AI solutions on accuracy, latency and cost with per-field scoring, making it highly practical, and gained 112 stars in four days. Monetization could be paid evaluation services or enterprise embedding, but it competes with LangSmith and DeepEval, with risks around test set construction and standardization.
Accuracy, latency p95, and cost benchmarking for model-based solutions, with per-field ground-truth scoring.
B image-to-3d-pipeline Integrates several open-source image-to-3D models with automatic scoring, lowering the barrier to 3D content production, and gained 242 stars in four days with good momentum. Monetization could be SaaS or a game asset pipeline, but it competes with TripoSR and ULIP, and 3D quality and copyright risks are challenges.
Reconstruct 3D meshes from images with several open-source models and score which did it best.
B polyledger Indexes Polymarket market metadata and on-chain trades for SQL queries, valuable to quantitative traders, and gained 231 stars in three days. Monetization could be a paid data source or research tool, but it competes with Dune and others, with risks around market data compliance and API changes.
Resumable Polymarket indexer: CLOB market metadata plus on-chain trades from Polygon, in one DuckDB file you can query with SQL
B camera-to-blender Quickly imports real objects into Blender from phone photos, simplifying the modeling workflow, and gained 101 stars in three days. Monetization could be a Blender plugin or asset service, but it competes with 3D scanning apps, with risks around model quality and copyright.
Take a photo of real objects, and paste them in Blender
B spafox It specifically scans React/Next.js single-page app bundles for secrets, targeting the common oversight of SPA configs ending up in the bundle. One sentence states the scope clearly, and security engineers can run it on their own frontend immediately; the limit is frontend-only coverage, where general secret scanners already abound.
A React/Next.js-aware secret hunter for single-page applications.
B kitter A local-first AI skill manager that isolates dependencies per project, improving agent skill management, and gained 126 stars in three days. Monetization could be an enterprise edition or cloud sync, but it competes with official skill directories and plugin markets, and skill formats are not yet standardized.
A local-first Skill manager: one library, only what each project needs.
B Show HN: I built my first MCP to manage Google Ads Addresses the pain point that Google Ads is read-only by letting AI agents directly manage ads, gaining HN attention in two days though upvotes were modest. Monetization could be SaaS subscriptions or agency services, but it competes with official MCP and existing marketing automation platforms, with risks around data security and API limits.
My co-founder and I tried running Google Ads for our Shopify store, but we could not get a positive ROAS (we spent 150USD on a single conversion). As we also could not afford an agency, we turned to A…
B visio-com-skill A Codex skill that generates editable Visio .vsdx files through Windows COM automation, solving the pain that AI diagrams only come out as images. The approach is clean and the scope clear; the limit is that it requires Visio and Windows locally, so others can only borrow the idea.
A clean-room Codex skill for generating editable Microsoft Visio .vsdx diagrams through local Windows COM automation.
B unreel Builds a personal AI video streaming service that may aggregate or generate content, drawing high interest with 103 stars in two days. Monetization could be subscription or ads, but it competes with YouTube and Netflix, with risks around copyright and generation costs.
Your personal AI video streaming service
B Xiaomi-SuperIsland-Playground An experimental project that takes over Xiaomi's 'Super Island' without any elevated permissions, an approach with reference value for MIUI/HyperOS customization fans. The repo is still too early to show generality; those interested in Android notification-bar rework are worth a few minutes to skim the implementation.
A Xiaomi Super Island takeover system that requires no advanced permissions
B heyflare A self-hosted HEY-style mail client on Cloudflare Workers, copying Screener, Imbox and other concepts, supporting Gmail and custom domains. Attractive to those who want to leave Gmail but keep HEY's interaction; the limit is that sending and receiving still rely on external services, and with only 9 stars in a day, real deployment is unverified.
Self-hosted HEY mail style email client on Cloudflare Workers — Screener, Imbox, Feed, Paper Trail, Bundles, Gmail + custom domains
B bough A single binary that reads local Claude Code history and summarizes what you actually changed, entirely offline. Handy for Agent users who cannot say what they did all week; limited to Claude Code, and summary quality depends on the session logs themselves.
Reads your Claude Code history and shows you the work you actually did. Local-first, single binary.
B LongGapsBetweenPrimes A Lean formalization of a theorem on prime gap lower bounds, a checkable copy of research output. Valuable to people working on formal math and theorem proving; extremely narrow, and useless to anyone who does not read Lean.
A Lean formalization of a bound concerning long gaps between primes
B tspu-check A dependency-free Perl script that uses TTL-limited SYN tracing to judge whether an IP is DPI-blocked by Russia's TSPU. Practical for Russian networks and clearly explained; limited by needing raw socket privileges, with conclusions depending on probe location.
Detection of IP address and subnet blocking on TSPU (DPI) using the cheburcheck/probe method — SYN tracing with limited TTL. A single Perl script with no dependencies.
B balrogg A lossless Vorbis/Opus recompressor written in C that squeezes up to 15% off file size by tuning encoding parameters, close in spirit to tools like precomp. Useful for music library archivists; the limit is that every file must be re-encoded, with real verification and compatibility costs.
demonically compacting (up to 15%) lossless Vorbis/Opus recompressor
B llm-inference-capacity-handbook It organizes how many QPS a given GPU can handle into a methodology, and the queueing-theory plus open-loop load-testing part is genuinely useful for capacity planning. But it is essentially a notes repo with nothing runnable, and accuracy rests on the author's own claims.
A desk reference for LLM inference (open-source edition): given GPU compute, how many QPS can a model handle? Three-layer model × three walls × queueing theory × open-loop load testing
B SideWire-KernelSU It opens a native, non-ADB channel on rooted Android to control the device from a desktop. A practical addition for mobile security and automation testing; limited by requiring KernelSU, with latency and stability unaddressed.
SideWire is a native bridge for controlling rooted Android devices from Windows, macOS, or Linux. It does not rely on ADB.
B herdr-telegram-agents It uses Telegram topics as session windows for each coding agent, with icons showing live status and inline buttons for choices. A handy remote control for people already running Agents long-term; many similar bridges exist, and it only wins on status display.
Drive your coding agents from Telegram like from the terminal. A topic per agent, live status in the topic icon, two-way chat with inline buttons for choices.
B claude-code-in-wechat It bridges terminal Claude Code into WeChat via official channels and iLink QR login, and clearly documents the risk-control case where ret=0 but no message arrives. Useful for people wanting a coding Agent on their phone; the risk is WeChat tightening at any time.
Connect Claude Code in the terminal to WeChat: official channels + Tencent iLink Bot API QR-code access; plus the silent risk control where ret=0 but nothing is received — the criterion (message_id), …
B vanguard Writing a compiler in Scheme with offline partial evaluation, directly targeting the Futamura projections, a topic rarely touched in recent years; but it's far from usable, with no example output or working instructions, more like a personal research draft. Compiler enthusiasts might take a glance; everyone else can skip it.
Compiler and offline partial evaluator aimed at the 3 futamura projections
B MFGAdaUnlock-RenoDx A memory patch that unlocks 3x/4x/6x multi-frame generation on RTX 40-series cards and fixes the temporal midpoint, explicitly stated to be pure memory operations with no file changes, which is fairly candid. But several similar unlock solutions already exist, it depends on driver versions, and you must weigh the risk-reward yourself.
DLSS MFG on Ada — unlocks 3x/4x/6x frame generation on RTX 40-series cards and corrects the temporal midpoint. In-memory only.
B knbs-cbk-pipeline It lands Kenya's KNBS/CBK public indicators in PostgreSQL via dbt, then exposes query APIs with signed cursor pagination, ETag caching, and rate limiting, with engineering detail more solid than most data projects. Unfortunately the audience is limited to those following Kenyan macro data; everyone else can only read it as an architecture example.
Ingests KNBS and CBK public indicators, standardizes and transforms them with dbt, stores marts in PostgreSQL, and exposes query + CSV APIs with signed cursor pagination, ETag caching (Redis), and rat…
B Malleon It automatically runs HTTP/HTTPS simulations against Cobalt Strike Malleable C2 profiles to verify traffic masquerading works as intended. Red teams have long hand-built C2 profile tests, and this fills a gap; sensitive use, so the audience is naturally security circles.
Automated HTTP/HTTPS simulation for Cobalt Strike Malleable C2 profiles
B greek-postal-code-db It embeds Greek postal code mappings to municipalities and communities in an offline Go app, with coverage and gaps clearly stated. It saves anyone doing Greek address validation a data-cleaning pass; limited in that update mechanisms and street-level coverage are left to you.
Embed a postcode database into your (offline) application. Regions, municipalities, communities, and partial street coverage included.
B FactReach A multi-platform search skill for agents covering 23 channels with zero-auth scraping; such aggregators break easily as platforms change, so breadth costs reliability. The two or three channels like GitHub and Reddit work, but overall it is more shell than engine.
Internet search skill for AI agents: covers 23 channels (web, X, Xiaohongshu, Douyin, Weibo, Bilibili, Reddit, YouTube, GitHub, WeChat official accounts, etc.), zero-auth first. Internet search skill …
B Mako A minimalist external display controller for jailbroken Apple Vision Pro. The description is clear and the scope restrained, but the prerequisite is a jailbroken headset, so very few can actually use it, and for most readers it is just an interesting curiosity.
A minimal external-display controller for jailbroken Apple Vision Pro.
B starcil A terminal workspace for AI coding agents with splits, tabs, workspaces, and an agent-drivable CLI. The direction is right, but terminal orchestration tools are already crowded, and its stars and usage don't yet show which point it gets right.
Terminal workspace for AI coding agents: panes, tabs, workspaces and a CLI your agents can drive. By Xtarify.
B Go-Live It hands developers anycast subdomains under *.go-live.me, an idea that is not new (sslip.io and ngrok came first), and the repo looks more like a deployment template for its own service, with real usability depending on that domain and infrastructure; the Instant and Globally-Distributed wording outweighs the information.
Instant, Globally-Distributed Anycast Subdomains for Developers (*.go-live.me) — 1-Click Vercel Deploy & Auto TLS 1.3
B cesium_river An implementation of river flow water surface effects in Cesium, a need common among digital twin and GIS visualization people. But the repo is more an effect snippet, its description ends with a plea for stars, and how much is reusable depends on code organization, with no docs to support it.
High-quality river flow-direction water surface for Cesium; if you find it useful, give it a Star ⭐ for support.
B icekarim A Go tool that renders Iranian gold prices and USD rates as Telegram WebP stickers in real time. Small but cleanly explained, very useful for local trading groups and meaningless outside that market.
High-performance Telegram bot for rendering live Iranian gold and USD prices as WebP stickers
B shiro-neko An agentic coding CLI in the terminal that always prompts for approval before any change, a restraint worth praising; but Claude Code, aider, and Codex CLI already occupy the same spot, the project doesn't explain its differentiation, and 10 stars don't yet show real usage.
An agentic coding CLI. It reads your code, edits it, runs your tests, and asks when the request is ambiguous — in a terminal UI, with every mutating action gated behind an approval prompt.
B FACET A foundation model for contact-rich fine manipulation, with an arXiv number, in a currently hot direction. But the repo shows no weights, data, or inference entry point, so it reads more like a paper landing page and offers limited help to engineering readers until it can actually be reproduced.
Facet-0: A Robotic Foundation Model for Contact-Rich Precise Manipulation (arXiv 2609.01596)
B matou A multi-agent workbench for Claude Code on macOS, with session management, DAG visualization, and Git Worktree all included. The features are comprehensive, but Claude Code frontends have proliferated in recent months, and it doesn't say how it beats existing ones.
Matou — a multi-agent desktop workbench for Claude Code on macOS: session management, cross-card natural-language collaboration, DAG visualization, tiered notifications, Agent HUD and Git Worktree
B MotionVault 200 web animation examples with copy-pasteable AI prompts, plus reverse-engineering any website's motion into a prompt—the latter half is somewhat interesting; but the main body is still an example-plus-prompt collection, with inflated marketing, essentially a material library for frontend inspiration.
200 web animation examples with copy-ready AI prompts + MotionLens: turn any site's animation into a prompt. React, TypeScript, Framer Motion, Three.js
B Wisp A native macOS assistant that reads the current screen and browser context when needed before answering, an intuitive interaction direction; but screen-aware assistants are already numerous, the project doesn't clarify privacy boundaries or which models it uses, and 8 stars make completeness hard to judge.
A native macOS assistant that reads the current screen and browser context when you need it.
B kubernetes-the-hard-way-epub-builder Converts the latest branch of Kubernetes The Hard Way into a readable EPUB, with an honest name and description, doing one thing well; but the audience is very narrow, serving only those who want to read this tutorial offline, the value is one-time, and 9 stars in two days shows limited interest.
Build a secure, reader-friendly EPUB from the current Kubernetes The Hard Way default branch
B korean-layout-compositions 277 Korean layout composition cards, breaking composition principles into one per page, a rare resource for designers working on Korean typography and publishing; but it's a content compilation, not a tool, and the barrier is Korean, so developers basically won't use it.
Korean-style layout composition encyclopedia, 277 entries — composition, visual principles, publishing, Hangul typesetting, traditional Korean composition. One principle per card.
B Show HN: I built a version of Omarchy that runs on Apple Silicon Lets Omarchy users run directly on Apple Silicon without a full install, using QEMU and HVF for hardware acceleration. The technical work is solid and 1384 stars show real demand; monetization is limited but it could join the Omarchy ecosystem, though Omarchy itself is niche and QEMU performance loss is noticeable.
I wanted to try out Omarchy before doing a full install, but I only have an M series MacBook so I would have to find a laptop and there's no official M support yet. So I decided to create something th…
B BirdWatch A solar-powered smart bird feeder that identifies birds via ESP32-CAM and AI, combining environmental appeal with fun, and its 50 stars show hardware hobbyist interest. Monetization could be a smart hardware crowdfunding campaign, but the market is limited, and risks are high hardware costs and AI recognition accuracy depending on lighting, with no core developer need.
BirdWatch is a solar-powered smart bird feeder with an ESP32-CAM. It captures visiting birds and uses AI to identify their species. A web interface lets you view photos, detected species, visitor hist…
B mvl A lightweight MVVM support library for multitasking systems and LVGL 8.x that simplifies embedded UI development. Its 40 stars show embedded-community approval, and it could build reputation as open source before offering custom services, but LVGL ecosystem updates and other UI frameworks compete.
A lightweight MVVM support library for multitasking systems + LVGL 8.x.
B Show HN: Kit. Claude Code but Concise Kit is a lightweight coding-agent runtime that is faster and cheaper than Claude Code, offering a terminal and ACP server. Its single binary and multi-platform compatibility are innovative but not clearly differentiated; it must compete with mainstream platforms, has few stars, and faces weak early momentum and high developer migration costs.
Kit is a coding agent runtime. It gives the model one tool for building and running programs. It provides a terminal client, and Agent Client Protocol (ACP) server, and A2A endpoint and a subagent orc…
B image-story-video-wizard A confirmation-gated skill for Codex and WorkBuddy that automates audio-first image-story videos, with 63 stars showing content-creator interest. It could sell as a paid skill pack, but the market is smaller than general video tools, and AI video quality is unstable amid heavy competition.
A confirmation-gated Codex and WorkBuddy skill for audio-first image-story video production.
B sayszz Local Whisper dictation on macOS with Metal acceleration, a one-key hotkey and history stats, explicitly benchmarked against Wispr Flow but fully offline. The scope is restrained and the promise clear, and it can replace paid dictation once installed; the challenge is that several local tools already exist, so it must win on ease of use rather than features.
Your voice, at your cursor, in milliseconds. Free, open-source, 100% on-device dictation for macOS — Metal-accelerated Whisper, one-key hotkey, local history with stats. A Wispr Flow alternative that …
B ductgen Parametrically generates a ducted-fan drone frame with build123d, and the output can be opened in SolidWorks for further edits. The topic is narrow but the problem is real and stated in one sentence; the limit is that it only serves ducted-fan hardware hobbyists, and it lacks example images and parameter documentation.
ducted drone frame generator for solidworks / build123d
B ndi-viewer-android A minimal NDI receiver for Android that spells out the SDK version, mDNS discovery, RGBA/UYVY handling and AAudio output, with implementation-level precision that shows it addresses real on-site monitoring needs. Suited to live streaming and studio users; the catch is you must supply the NDI SDK and compile it yourself.
NDI viewer Android - minimal standalone NDI receiver (SDK v6.3.2, mDNS, RGBA/UYVY, AAudio)
B oc-english It turns vocabulary memorization into a raising-style mini-game for Chinese learners of English who struggle to keep going, and passing a hundred stars in a day shows it hits a real emotion. Layering gameplay onto study is a valid path, but the description is one sentence with no word-list source and no indication of web or client, so retention depends on future updates.
English learning through a raising sim game
B zodchiy An architecture audit spec for coding agents: only findings whose cost can be measured from git history count, paired with a stdlib-only CLI. Treating measurability as a gate is rare in agent code review and worth a look; the limitation is more philosophy than empirical data.
An architectural audit where a finding is admissible only once its cost is measured from git history. A doctrine for coding agents plus a stdlib-only measurement CLI.
B pons-launch-engine An open-source engine for token launches and sniping that directly serves speculators. It gained 75 stars from short-term hype, but the project is somewhat gray and may violate platform terms; monetization depends on the crypto market, with legal and compliance risk.
An open-source campaign engine for launching and sniping your own tokens on Pons
B zenspark-browser A gamepad-first browser for ten-foot TV interfaces, a direction that is nearly empty in open source and genuinely needed by set-top box and HTPC builders. But the repo is only half a day old, with no rendering engine choice or remote-control testing, so usability must wait for a demo.
An open-source controller-first web browser designed for TVs and 10-foot interfaces.
B Academic-AI-Literature-Reviewer-Ollama A literature review tool driven by local Ollama, whose key design is verbatim citation verification so the model cannot fabricate quotes, outputting APA7 directly. Practical for grad students writing reviews; many AI review products exist, and its only difference is citation checking.
Automated literature reviews using local LLMs via Ollama. Searches academic databases, acquires full-text papers, and writes APA 7th reviews where the model is structurally prevented from fabricating …
B KorvaTTS A Vietnamese-first on-device TTS supporting code-switching, running on ONNX Runtime. The language coverage is a genuine gap and the description is restrained; but the audience is narrow, with no model size, quality comparisons or samples, so short-term it looks like a personal product.
Vietnamese-first, on-device text-to-speech with code-switching (ONNX Runtime)
B ui-atlas-slicer A pure front-end tool for splitting packed UI sprite sheets back into individual images, useful for interface recreation and game UI work. It runs in the browser and stays narrowly scoped, but TexturePacker and several online tools already cover sprite slicing; its only edge is needing no install, not a new approach.
A pure front-end UI image atlas splitting tool.
B threejs-procedural-rocks-cliffs Generates seamless, watertight rocks and cliffs in Three.js, with an interactive demo and a Codex skill. It saves work for procedural-terrain developers; however, it is closer to sample code, lacking generation parameter documentation and an export pipeline.
Watertight procedural rock fields and cliffs in Three.js, with an interactive demo and Codex skill.
B omaroll A C++ image and video browser that organizes local folders into a media library, positioned like an image viewer with video added. C++ brings performance advantages for local media browsing and a dozen stars in a day shows the pain point is real, but it names no platform, gives no install method, and its difference from ImageGlass or qView is a single slogan.
A fast, beautiful image and video viewer that turns your media folders into a library.
B openscad-skill Wraps OpenSCAD modeling, rendering, and export into an agent skill, a handy supplement for those using coding agents for parametric 3D printed parts, with restrained scope and an honest description. The limitation is that it's just prompts plus script wrapping, with capability capped by the model.
ModelRift OpenSCAD skill for designing, rendering, inspecting, and exporting printable parametric parts.
B subpool A self-hosted AI subscription pool for teams, essentially sharing subscription quotas. The idea is practical but runs into each provider's terms of service, and it doesn't explain account isolation, quotas or auditing, so the deployment risk outweighs the benefit.
A lightweight, self-hosted AI subscription pool for teams.
B ryza-ai-revive An offline Ryza fan AI companion with a bundled model and TTS integration, where the author honestly labels it fan work and BYO-model. Technically it packages an LLM plus voice into a local desktop pet, with configuration still the barrier; similar character companion projects are numerous, and it depends on the original IP, so long-term maintenance is doubtful.
Offline fan-made Ryza AI companion. Bring your own LLM/TTS. 离线同人莱莎 AI 陪伴，自填大模型与语音。
B research-paper-daily-push A Codex Skill chaining daily arXiv search, evidence review, scoring, Chinese summaries and WeChat push, a complete pipeline. But paper-digest tools are plentiful, value depends on scoring quality, and it heavily depends on personal WeChat configuration.
A configurable Codex Skill: daily retrieval, evidence review, scoring, Chinese summary, WeChat push, deduplication and research memory.
B TubeFlow A Manifest V3 extension that blocks YouTube ads, skips Shorts and prevents pause, with concrete features that work on install. But its lifespan depends on fighting the platform, similar open-source options are plentiful, and long-term maintenance is unproven.
A powerful Manifest V3 Chrome extension for YouTube & YT Music featuring dynamic adblocking, Shorts auto-skipping, Anti-Pause, and real-time stats in a clean uBlock-style UI.
B agent-git Version control for agent sessions is an interesting idea—right now an agent run leaves only a chat log. But the description is just a joking slogan, with no indication of snapshot granularity, its relationship to git, or how to integrate into existing workflows. Concept first, tool not yet formed.
Code is cheap, show me the talk: version control for agent sessions
B binance-go An unofficial Go SDK for Binance REST/WebSocket, filling a real market gap, as 31 stars in a day suggests. But the production-ready claim is too much for a one-day repo: rate limits, reconnects and error codes are unaddressed, so read the code before integrating.
Production-ready unofficial Binance REST and WebSocket SDK for Go
B BinancePayTool A personal payment gateway using Binance Pay, implemented in Go, for small merchants or indie developers who need to accept cryptocurrency. The direction is specific but the scenario is narrow, and it depends on Binance merchant credentials; similar self-hosted gateways are plentiful, and it doesn't explain its differences or how it handles compliance.
A personal payment gateway based on Binance Pay
B Co-Scientist-Ideation A research ideation skill for Codex that drives hypothesis evolution through independent review and rebuttal, with restrained wording that emphasizes human decisions. But it is still a prompt-flow package whose results depend entirely on the underlying model, and such AI co-scientist skills have clustered lately with no verifiable output differences.
Evidence-grounded Codex Skill for bounded AI co-scientist ideation: distinct hypotheses, independent review, objection-driven evolution, and human-led decisions.
B react-native-continued-task Unifies iOS 26 BGContinuedProcessingTask and Android WorkManager to fix React Native background task pain points. With 117 stars in 0.5 days, developer demand is strong, and it could be adopted in production with paid support. Risk: the new iOS 26 API brings compatibility risk.
Background tasks that keep running after your app is backgrounded, for React Native and Expo. Wraps iOS 26 BGContinuedProcessingTask (Live Activity) and Android WorkManager foreground services behind …
B course2md Turns video courses into Markdown notes with slides, useful to students and developers. Nearly 190 stars in two days shows broad reach, and it could become a SaaS or learning-platform component. Risk: output quality depends on the slides, and existing note tools compete.
Turn YouTube, Bilibili, or local course/meeting recordings into slide-illustrated Markdown and HTML lecture notes.
B morpho-stock-vault Lets Robinhood stock tokens be used as collateral to borrow USDG on Morpho Blue, blending traditional finance with DeFi. At 43 stars it's niche but frontier, with high protocol potential. Risk: regulatory uncertainty and tokenization risk.
morpho-stock-vault : Morpho Blue USDG Loans Against Robinhood Stock Tokens
B DLSS5oneclick Enables the leaked DLSS 5 in one click across nearly all DX11/12 games, greatly improving the experience. Its 166 stars show players urgently want it; commercialization is legally constrained and NVIDIA will block it, while using leaked binaries carries stability and legal risk and competitors abound.
One-click setup of the leaked DLSS 5 neural-rendering build for any DX11/DX12 game on RTX 20–50, with or without DLSS. ReShade + RenoDX add-on (or OptiScaler engine); DLSS5-Feeder + LumeniteFX for gam…
B Show HN: Weedout – Safari extension that hides YouTube AI-labeled videos Uses YouTube's official tags to hide AI videos, addressing content filtering pain. At 33 HN points it sells directly with a clear model, but the market is niche and depends on platform tags and App Store review. Risk: YouTube changing tags or policy, and competitors copying it freely.
I'm the developer. I built this because AI-generated conspiracy videos were getting out of control in my own YouTube feed. Weedout is a $1.99 Safari extension for macOS that removes videos YouTube lab…
B cn Merges Tailwind class names fast, replacing tailwind-merge and clsx with a claimed 30x speedup. At 34 stars and from shadcn, it's credible, with a clear path as a standard dependency. Risk: weak first-mover advantage and reliance on community adoption.
cn is a new engine for Tailwind class merging and conflict resolution. It replaces tailwind-merge and clsx. Same APIs. Full parity. And it is 30× faster.
B MediaToPose A Blender plugin that converts images or video directly into skeletal poses and animation, saving manual alignment for 3D users doing previs and reference animation. Boundaries are clearly stated. It does not say which pose estimation it uses, its accuracy, or dependencies, and paid options like Rokoko and Plask already exist.
Blender plugin for converting any images and videos to poses and animations.
B abacuscopilot An interactive CLI assistant for ABACUS first-principles calculations, helping materials and physics users write inputs and do post-processing. The domain is narrow but the audience is clear, and the description does not overstate. It does not say which model backs it or how errors are handled, and general LLMs can partly do the same.
Interactive command-line pre/post-processing copilot for the ABACUS DFT software (GPL-3.0)
B ComfyUI-H3VideoOutpaint A ComfyUI node for MiniMax H3 video outpainting, using sliding windows for prompt-free sequential extension, with a restrained description and clear boundaries. It saves people doing long video extension from stitching windows themselves, but only serves H3 and ComfyUI users, and 9 stars is no usage evidence.
Prompt-free sequential sliding-window MiniMax H3 video outpainting for ComfyUI
B Hyper-Liquid-Trade-Copy-Verifier Uses public APIs to assess Hyperliquid trader risk and avoid copy-trading traps. With 39 stars in 0.2 days, the need is clear, and it could become a paid query service. Risk: API limits and emerging competitors.
Assess a Hyperliquid trader before copy trading — independently analyze ROI, win rate, and PnL using the public Hyperliquid API, detect martingale/grid recovery strategies, and assess maximum drawdown…
B m4-prefill-engine Metal kernels targeting the LLM prefill bottleneck on Apple M4, claiming about 3.7x speedup, a clear pain point for Mac users running long contexts locally. But the repo is one day old with no reproduction steps, so "True GPU Speedup" needs verifying yourself.
Hardware-saturated Metal inference engine targeting the LLM prefill bottleneck on Apple M4 (~3.7x True GPU Speedup).
B bma8 A bare-metal Atari 8-bit emulator on Raspberry Pi that skips the OS for instant boot, following BMC64's approach. Restrained scope and an honest description make it a clean little toy for retro and Pi fans, limited by a narrow audience and few stars.
Baremetal Atari 8-bit emulator for Raspberry Pi (400/800/XL/XE/5200). Instant boot, no OS. Inspired by BMC64.
B ggwave A React Native module wrapping ggwave's sound-based data transfer, adding the audio layer the original library lacks. Two phones can send bytes without network or pairing, a concrete use case, though the very low bitrate limits it to pairing handshakes or small config snippets.
Data over sound for React Native. Two phones pass bytes through the air with no network, no pairing and no account - a Nitro module wrapping ggwave, with the audio layer it doesn't have.
B agentic-resource-discovery Offers federated search and a registry index for MCP tools to solve discovery. At 50 stars the ecosystem is still early, and it could sell data services or a paid API. Risk: the MCP standard evolves fast and new competitors appear.
Neuronto Agentic Resource Discovery (ARD) Index. Federated search across every public ARD registry, plus a verified tool index read from each MCP server's own tools/list, hybrid lexical and semantic r…
B constellation-dice A nebula die rendered in Unity URP with a single-pass raymarch, each of its six faces carrying a constellation; a clean small sample for shader learners, but only a visual demo, not really a tool.
Astral dice in Unity URP: nebula, stars and per-face constellations in a single raymarch pass on the default cube
B ts-refinement A small library adding refinement types to TypeScript so types carry value constraints, worth a look for anyone doing type-level validation; but there is only a one-line description, no usage examples, and no clarity on how it differs from zod or effect.
Refinement types for TypeScript!
B NeSLE Runs thousands of NES environments in parallel on a single CUDA GPU to train RL agents, making the emulator GPU-native, a concrete and novel idea. Useful for researchers building RL benchmarks and game agents, though it only covers NES and depends on specific hardware.
GPU-native NES emulator + RL stack: thousands of parallel NES envs on one CUDA GPU, training agents on Super Mario Bros.
B FeedKit One-click install and removal of DLSS5-Feeder so games without DLSS get neural rendering. It's essentially an installer wrapped around an upstream tool with no technology of its own, and you bear the risk of modifying game render pipelines.
One-click installer/uninstaller for DLSS5-Feeder: brings DLSS 5 neural rendering to games without DLSS. Fetches everything fresh from upstream at install time.
B SlopTV This bot scrapes YouTube comments to endlessly generate AI slop short videos, a piece of performance-art satire on AI slop; the idea is fun and the name self-deprecating, but once installed it does nothing beyond watching it amuse itself.
SlopTV: an infinite AI slop generator from youtube comments
B manaca-1b-base An open Brazilian Portuguese 1B base model from a national lab, emphasizing reproducibility, a free foundation for pt-BR application teams. But Sabiá and Tucano already exist at this scale, and the repo gives no comparative evaluation, with unclear weight and data openness.
LLM aberto e reprodutível para o Português do Brasil An Open, Reproducible Brazilian-Portuguese Language Model
B hmg-ipmap An open-source geolocation platform resolving IPv4 addresses to city and country, hosted under the Hyundai Motor Group organization with corporate backing. A realistic alternative for anyone self-hosting IP geolocation, though accuracy and data update frequency still need comparison with commercial libraries like MaxMind.
Open-source IP geolocation platform that resolves IPv4 addresses to city, country, and continent
B LocalAI-Advisor A Windows hardware detector plus local LLM recommender that answers the concrete question of how large a model a machine can run, and can plug directly into Ollama. Convenient for local-model users, though recommendation quality depends on the GPU database keeping up with new hardware.
Windows Hardware Diagnostic & Local AI Model Advisor. Fast in-process hardware detection & intelligent local LLM recommender for Ollama.
B bside Tests whether AI browsers can really replace script automation using five Korean web tasks, a refreshingly practical direction. But it's a benchmark for Aside's own product with only five samples, so discount the conclusions.
Benchmarks for Aside — do AI browsers actually beat existing automation on the real web? Five Korean-web errands, three setups, everything published.
B pebble-watchface-builder A drag-and-drop Pebble watchface editor in the browser that exports Pebble C SDK code for CloudPebble; aimed at the recently revived Pebble hardware crowd, the scope is clear but the audience is tiny and the ecosystem just starting.
Browser-based drag-and-drop watchface builder for the Pebble Time 2, Core 2 Duo and Pebble Round 2 that emits Pebble C SDK code for CloudPebble.
B anti-ai-defensive-writing Turns ASD-STE100 Simplified Technical English rules into a Claude Code skill, specifically stripping AI-sounding disclaimers and fabricated metrics from academic writing. The direction is more specific than generic de-AI prompts, but effectiveness depends entirely on the model and is hard to verify; it is still a prompt rule set.
Remove unrequested caveats, faux rigor, invented metrics, and AI formatting residue from academic writing.
B CAER.code Causal action-effect reweighting in world model training, precisely titled and legitimate research output. But there are no install instructions or reproduction path, so ordinary developers can't use it today; only world model researchers would read the code.
Causal Action Effect Reweighting for World Model Training
B tourcn A React onboarding component published as a shadcn registry, built on Base UI and Floating UI, installable directly for product tours. But mature options like driver.js and react-joyride already exist; here the main difference is distribution.
Product tours for React, delivered as a shadcn registry. Built on Base UI and Floating UI.
B YMM4-LiquidGlass A liquid glass panel effect plugin for YukkuriMovieMaker4, redrawing Apple's material in Unity. The audience is narrow, but the scope is restrained and the description honest, and YMM4 users will find it directly useful.
A video effect plugin for YukkuriMovieMaker4 that renders Apple's Liquid Glass-style glass panels
B pons-sniper A sniping bot for Pons token launches that open-sources its front-running logic for fellow players; there is nothing technically new, and the profit window for such tools is measured in seconds, so seeing it two days late means it has expired.
Open-source bot for buying tokens launched on Pons
B omarchy-airplay A QML plugin that adds AirPlay receiving to the Omarchy desktop, solving the small pain of receiving screen casts on Linux, with a restrained scope. But there is only one line of description, no dependencies or known issues, and 8 stars in 0.2 days says nothing about usability, so keep it on watch.
Omarchy Airplay Plugin
B audit-mind An agent for regulatory compliance and auditing, listing knowledge management, rule extraction, document auditing and evidence Q&A all at once. The direction is real but the scope is too broad, with no deployment or evaluation, more a proposal than a working system.
AuditMind is an AI Agent system for regulatory compliance and audit scenarios, supporting regulatory knowledge management, traceable rule extraction, document compliance auditing, audit task managemen…
B osu-gd A userscript that lists pending and graveyard guest difficulties on osu! profile pages, helping mappers and modders check their own work. The target is extremely narrow but cleanly done. It is meaningless to anyone outside osu mapping, and 9 stars in a day suggests little maintenance.
osu!のプロフィールページにゲスト難易度のPending、Graveyardの譜面を表示します。
B fuyue A local-first AI companion PWA with a self-hosted relay and portable memory, useful for people who want to host character chats themselves. But one sentence packs in PWA, Android shell, relay, memory, and UI package, a scattered scope, and mature options like SillyTavern already exist.
Local-first AI companion PWA with optional Android shell, self-hosted relay, portable memory, and reusable UI packs.
B simplified-technical-english Turns ASD-STE100 Simplified Technical English rules from aviation manuals into a Claude Code skill that helps writers compress wordy sentences. The rules are mature and checkable, more reliable than homemade "de-AI-ify" prompts, though it only constrains English technical writing, a narrow scope.
Kill AI slop with the writing standard built for flight manuals. A Claude Code skill applying ASD-STE100 Simplified Technical English principles.
B Descent-Android A continuation fork porting DOS-era Descent to Android, of interest to retro shooter fans; but the description is a single line about an "updated port," with no build steps and no word on how it differs from the upstream port, leaving the effort required unclear.
updated port of descent ms dos game for android
B cizgile A zero-dependency URL slug engine supporting Unicode and IRI conversion, solving multilingual URL normalization pain points. 63 stars in 0.7 days, with clear developer demand. It could become a mainstream slug library, monetized through maintenance sponsorship. The risk is mature competitors like slugify, so it must prove performance and script coverage advantages.
cizgile — Zero-dependency URL slug engine. RFC 3986/3987 slugs, transliteration for 7 scripts, Unicode slugs, IRI ↔ URI, percent-encoding. Pure TypeScript, works everywhere.
B gameops-investigator An incident investigation agent for game operations, using read-only MCP tools with evidence first. It addresses reliance on manual experience in game incident troubleshooting. 52 stars in 1.3 days, with attention from the game community. It can be commercialized for game companies. The risk is competition from general AIOps and long procurement cycles in the game industry.
Evidence-first game operations incident investigation agent with read-only MCP tools and reproducible evaluations.
B Show HN: SlideOps – slides from a repo that flag when they drift from the code It lets slides carry code provenance and detects code drift to prevent stale documentation. It requires no model calls and is implemented fast with the Python standard library. As a Claude Code plugin or Agent Skill it has real use cases and could commercialize via plugin sales or subscription; the risk is competition from documentation generators like Mintlify and dependence on the AI Agent ecosystem.
I kept generating slide decks about my codebases with an agent, and weeks later they would go stale. Updating them was quite costly (time & tokens), as the agent would have to re-scan the whole repo t…
B Show HN: Floe – an open-source plugin for sample libraries – CLAP/VST3/AU An open-source sample library audio plugin supporting CLAP/VST3/AU, offering musicians a free host. 18 points on HN; the author's main business is selling sample libraries, and open source can expand the audience. The commercialization path is ecosystem traffic and paid libraries. The risk is many DAW plugin competitors and the need to continuously maintain format compatibility.
I'm Sam, I make sample libraries (as FrozenPlain). These sample libraries run inside Floe, my audio plugin for Linux, macOS and Windows. It's designed for musicians, composers and producers - typicall…
B osintPanel An open-source OSINT panel combining WHOIS, IP, email and SSL lookups with a desktop UI. With 37 stars in 0.1 days the need is real, and it could become a standard security analyst tool with a paid pro version. Risk: mature rivals like Maltego and the need for constant feature updates.
Open-source OSINT panel for WHOIS, IP geolocation, email, and SSL certificate lookups — with a bundled desktop UI.
B shader-for-interfaces It provides GPU shader design and debugging skills for AI Agents, lowering the barrier to UI effects development. 51 stars in 1.6 days, part of the Agent Skill ecosystem. It can be commercialized through a skill marketplace or enterprise licensing. The risk is that shaders are niche, and competitors are traditional design tools.
Agent Skill for designing, building, debugging, and validating focused GPU effects in product interfaces.
B r20-quantum-trader An LLM-driven self-evolving quantitative trading terminal with a six-asset AI brain. With 36 stars in 1.4 days it appeals to quant enthusiasts, but financial risk is high and it competes with professional platforms like QuantConnect.
⚡ LLM-Native Self-Evolving Quantitative Trading Terminal & Execution Engine (Bloomberg-Style Dark Terminal + 6-Asset AI Brain)
B codex-bridge-chatgpt Builds a reasoning handoff bridge between Codex and ChatGPT that preserves local edits and test evidence. With 40 stars in 1.7 days it meets AI development integration needs and could become a standard agent collaboration component, though commercialization is limited. Risk: OpenAI API changes and competing multi-agent frameworks.
Verified Codex-to-ChatGPT Web reasoning handoffs with local edits, tests, and receipts.
B agent-notion-template-docs Lets AI generate Notion-style documents with locked structure and styling for consistency. With 50 stars in one day the need is real, and it could become a document template standard with subscriptions or a template marketplace. Risk: dependence on the Notion ecosystem and competing document AI tools.
Make your AI agent write documents like Notion — Claude Code skill that locks document structure & Notion visual style
B academic-defensive-writing-auditor Detects and rewrites academic defensive writing while preserving scientific hedging. With 32 stars in 1.7 days it meets a research writing need and could be sold as a plugin or service, but demands strong corpora and academic norms. Risk: competition from Grammarly and the need for professional accuracy.
An open-source LLM skill for detecting and rewriting defensive academic writing while preserving necessary scientific caveats and evidence boundaries.
B codex-task-pointer Addressing how Codex drifts after automatic context compression on long tasks, it uses a pre-compression hook plus a CAS-protected task pointer so the agent remembers where it was; the problem is real and the scope small, a rare approach, though it's tied to a single toolchain.
Pre-compaction hooks + a CAS-guarded task pointer that keep Codex agents on-goal across context auto-compaction — no mainline drift, no duplicate work, no conflicting progress in long runs.
B elephpantdb A SQL database implemented in pure PHP by the developer behind fselect and carbon, embeddable in PHP apps without installing extensions, a direction almost nobody pursues; it's experimental for now, so don't expect production use, but its implementation approach is worth a look.
A pure PHP SQL database.
B cratera A bare-metal microVM execution engine in Rust touting snapshot restore within 5ms, NIC-free isolation, and 30 language toolchains, aimed at AI code sandboxes and evaluation machines; the metrics are specific, but it faces Firecracker and microsandbox, and 14 stars in a day doesn't yet show it can land.
Bare-metal microVM execution engine in Rust. Sub-5ms snapshot restores, Zero-NIC network isolation, and 30 built-in language toolchains for AI code sandboxes & online judges.
B LoopArena Benchmarks Loop engineering with LLMs as runtime controllers, aimed at ML research. With 70 stars in 0.9 days it has academic attention and could become a standard benchmark, but commercialization is limited. Risk: a new concept with narrow applications, competing with other agent benchmarks.
Benchmarking models as runtime Controllers for Loop Engineering
B jcss A build-free Tailwind-style utility-class runtime that adds a layer of scoped component styles, a small and clear positioning; but the browser build of tailwindcss and UnoCSS already cover most scenarios, and at 15 stars it's unclear which specific pain point it solves, so just note the name.
A lightweight, build-free runtime for Tailwind-like utility classes and scoped component styles
B pai-voice A self-hosted, model-agnostic realtime voice layer for both PWAs and terminal agents, with a concrete entry point. The problem is that Pipecat and LiveKit Agents already cover similar needs, and it does not explain key differences like latency or interruption handling.
A self-hostable, model-agnostic realtime voice layer for personal PWA and terminal agents.
B stampstudio An online open-source seal/stamp generator with a real but very niche need, the kind of project that does one small thing. Scope looks restrained, but the description never explains export formats or use cases, and 25 stars means no usage feedback yet.
An online, open source stamp creator.
B tempmail-telegram-bot Generates temporary email addresses on demand inside Telegram, a plain idea that does work; the problem is its dependence on third-party temp-mail services for stability and legality, and with many similar bots already around, there's little reason to switch.
A Telegram bot for generating and managing temporary email inboxes on demand.