Worth watching384
2026-10
A florilegium Procedurally draws Delft-blue floral page margins in LuaLaTeX. Extremely narrow but tidy, and its fast star growth shows demand for small typographic craft; only useful if you already write in LaTeX.
Procedurally generated Delft-blue floral page margins for LuaLaTeX, drawn in Lua
A dontallow Auto-dismisses macOS's local-network discovery prompt by pressing Don't Allow, idling at zero CPU and surviving reboots. Tiny scope, honest description, one thing done well; only relevant to privacy-minded Mac users.
Auto-dismiss the macOS 'Allow X to find devices on local networks?' prompt. Presses Don't Allow for you, zero CPU idle, survives reboots.
A Radiance A focused Windows utility folding NVIDIA digital vibrance, DDC/CI monitor control, color-vision overlays and crosshairs into one place. Modest, specific claims and clear scope; limited to Windows with a discrete GPU.
Lightweight, bloat-free Windows display utility for NVIDIA digital vibrance, DDC/CI hardware monitor control, Direct3D 11 color vision overlays, and custom crosshairs.
A ps2-native-recomp Statically recompiles a PS2 disc image into a native PC binary, crediting its MIT upstream, marked WIP, bring-your-own-disc. Honest scoping and a genuinely uncommon approach; the most code-worthy repo of the day.
Turns a PS2 disc image into a native PC program by static recompilation; builds on ps2-web-recomp (MIT). Work in progress. Bring your own disc.
A narratarr An *arr-style audiobook manager with a built-in player, wiring discovery, fetching and listening into one pipeline. A credible option for self-hosters beyond Audiobookshelf, though one day old with unproven source reliability.
Your audiobooks: found, fetched, and playing. An *arr-style audiobook manager with a built-in player.
A c3rb3r-rescue Forensics and recovery for ransomware-hit Synology NAS: image preservation, RAM key recovery, copy decryption and GitLab validation. Reads like a runbook a victim could follow; scoped to one family and DSM.
Evidence-based DSM ransomware recovery: preservation, RAM key recovery, copy decryption, and GitLab validation.
A instantTrack Sets up whole-body motion tracking for a Unitree G1 in 15 minutes on one GPU, combining a world model with a simulation loop and naming its frameworks. A reproducible starting point for robotics researchers with the hardware.
Whole-body motion tracking for the Unitree G1 in 15 minutes on one GPU (SuperMimic-style world model + InstantMimic-style GPU loop, mjlab)
A pcaplat Computes one-way latency, tail statistics and microbursts from two packet captures; narrow, honest and immediately runnable for network troubleshooters, though offline-only and without visuals.
One-way latency, tail statistics and microburst analysis from two packet captures
A Nook A Go terminal markdown editor built around a 'focus line' that keeps raw text centered while the rest scrolls rendered; a concrete interface idea in a crowded editor space.
Nook is a terminal markdown note editor written in Go with the concept of Focus Line: the line you are typing always stays in the middle of the screen as raw markdown, while the rest of the document s…
A gen5-RearDashcam Records the factory backup-camera feed to a USB stick on a specific head unit and adds in-car playback; a small, honest job well scoped, held back by a very narrow vehicle audience.
Records the backup camera feed to a flash drive plugged into the head unit and adds an app for playback on the car's screen.
A Perfect-Dark-XBLA-PC-Recomp A native Windows recompilation of the Perfect Dark XBLA remaster via ReXGlue, honest about needing your own copy; mostly build glue, but genuinely playable for retro fans.
Native Windows PC recompilation of the Perfect Dark Xbox 360/XBLA remaster using ReXGlue. Requires your own copy.
A roboArm A complete open robotic arm: CAD, Arduino firmware and a PySide6 control app under MIT; well scoped for teaching, though payload and precision are unvalidated.
Open-source robotic arm: SolidWorks CAD models, Arduino UNO firmware and Python/PySide6 desktop control. MIT licensed.
A retrack-tracker A from-scratch four-track looper firmware for the original Polyend Tracker; real engineering, honestly described, but flashing is risky and the audience is tiny.
Retrack for Polyend Tracker: a from-scratch 4-track looper firmware for the original Polyend Tracker
A Ultra-Water A real-time FFT ocean for three.js with shore breakers, foam, caustics, underwater rendering and buoyancy above 200 fps. Narrowly scoped as a rendering module rather than another engine — worth ten minutes for WebGL graphics people.
Real-time FFT ocean renderer for three.js (WebGL2): shore breakers, foam, caustics, underwater rendering and buoyancy at 200+ fps
A duper-cves Localizes patched CVEs by diffing iOS 24A435 against 24B115 binaries; 36 entries so far. Genuinely useful to firmware-diff and iOS security researchers, but it's a list, not a tool or PoC.
cves localised in the 24A435->24B115 binaries (iPhone17,5), currently 36
A leancast A minimal ad-free DLNA/UPnP cast receiver for Android TV. Most receivers on that platform are ad-laden, so doing only this one thing honestly fills a real gap; the technology is simple and long-term upkeep is the open question.
Ad-free DLNA / UPnP cast receiver for Android TV · 极简无广告投屏接收器
A shoulder A TUI showing Claude Code, Codex, Gemini CLI and opencode working in your project live — finally a place to watch several agents at once. Small scope, honest naming; the risk is vendors shipping this themselves.
Watch over your agents' shoulder. A TUI that shows Claude Code, Codex, Gemini CLI and opencode work in your project, live.
A bf-private-server A revival server for the shut-down mobile game Brave Frontier: C++ game server plus Cloudflare Workers tooling, code only and honest about it. Real value for its community, but you need the original client assets.
Unofficial Brave Frontier revival server: C++ game server, Cloudflare CDN workers and tooling (code only)
A ai-phone A SIP phone daemon running local speech and language models on Apple Silicon — an unusual pairing of telephony and on-device inference. Concrete for voice-agent prototyping, though locked to macOS hardware with models you supply.
A SIP phone daemon with local speech and language models on Apple Silicon.
A omux A native macOS terminal that runs several Claude Code sessions side by side with projects, live agent status, resumable conversations and file previews. Aimed squarely at heavy agent users; tmux and Electron shells are the alternatives, this bets on native feel.
Native macOS terminal for running multiple Claude Code sessions side by side. AI coding agent manager with projects, live agent status, resumable conversations and file previews. Built with SwiftUI.
A scrubshot Paste a screenshot and it detects phone numbers, ID numbers, card numbers, tokens and addresses, then blurs them, entirely in the browser and offline. Genuinely useful for tickets and docs; accuracy is unproven and it faces many existing redaction tools.
Paste a screenshot and it finds phone numbers, ID cards, bank cards, tokens and addresses, then redacts them in one click. Images stay in the browser the whole time — no upload, no network, works offl…
A velo A terminal speed test in Go with latency, up, down and bufferbloat in a cockpit-style TUI. Terminal speed tests exist, but quantifying bufferbloat is rare, and it works the moment you install it.
A speed test for the terminal with a cockpit dashboard: latency, download, upload, and bufferbloat.
A portal2-recomp Portal 2 statically recompiled from its Linux x86 binaries to ARM64, running on macOS and iPhone with your own copy. A rare 3A example of the static-recompilation wave: real engineering, but playability and legality hinge on your own files.
Portal 2 statically recompiled from its Linux x86 binaries to native ARM64: playable on macOS and iPhone. Bring your own copy of the game.
A hdUE An in-process Unreal Engine Hydra render delegate with Lumen, path tracing and Houdini Solaris integration. Anyone wiring UE into a USD pipeline gets it instantly; the build barrier is high and docs thin, so it stays early-stage.
A same-process Unreal Engine Hydra render delegate with Lumen, Path Tracer, and Houdini Solaris integration
A html A fast HTML parser with source offsets and CSS selectors from a proven author. Small, clear API; its edge over parse5 or htmlparser2 is position tracking plus speed, useful for scrapers and editors, though docs are still thin.
A fast HTML parser with source offsets and CSS selectors
A scx_pgtier A two-tier sched_ext/eBPF CPU scheduler aimed at PostgreSQL servers mixing short transactions with long queries, openly implementing an arXiv idea. Narrow, honest and technical; valuable if you tune Postgres on new kernels.
Two-tier, selectively unfair sched_ext (eBPF) CPU scheduler for PostgreSQL servers that run short transactions next to long queries. Open-source implementation of the UFS idea from arXiv:2605.02377.
A anymove Fits a rig from two views inside one AI video and exports GLB, aimed at indie animators without mocap gear. Clear output format and tight scope; accuracy is still capped by monocular video.
Motion capture for any creature from one AI video: two cameras in one frame, a DINOv2 joint reader, the rig fitted in both views, a GLB out.
A kanshi-gtk4 A GTK4 front end that arranges displays on Wayland and saves the layout as a kanshi profile. A small, sharp tool for multi-monitor Linux users who switch layouts often; the scope is one task, tied to the kanshi ecosystem.
Arrange displays on Wayland and save the layout as a kanshi profile
A vita-turbo-storage A taiHEN plugin that speeds up SD2Vita and internal storage on PS Vita firmware 3.60–3.74. Narrowly scoped and clearly described, immediately useful to the remaining Vita homebrew crowd — and to nobody else.
Faster SD2Vita and internal storage for the PS Vita (taiHEN plugin, firmware 3.60-3.74)
A vram-switcher Schedules a persona LLM, IndexTTS voice and ComfyUI image jobs to take turns on one 12GB GPU without a proxy. A very real pain point for local-AI hobbyists; the catch is that it is tied to the AstrBot/ComfyUI stack.
GPU coordination for local AI companions on AstrBot / DeepSeek Harness: persona LLM, IndexTTS voice and ComfyUI selfies take turns on one 12 GB card. No proxy.
A toyxyz_RGBMocap Native Windows video mocap chaining ExMR, FlowHMR and GVHMR with contact correction, retargeting and FBX/VMD export. Practical for MMD and VTuber work; the cost is heavy model dependencies and fiddly setup.
Native Windows video motion capture with ExMR, FlowHMR and GVHMR, contact correction, humanoid retargeting, and FBX/VMD export.
A eo-overpass-predictor Predicts Landsat 8/9 and Sentinel-2 overpasses for any ROI using SGP4 orbits and swath coverage, cross-checked against ESA acquisition plans. Concrete and honestly documented for remote-sensing teams; limited to a few optical satellites and a notebook workflow.
Swath-aware Landsat 8/9 and Sentinel-2 overpass prediction for any ROI: SGP4 orbits from CelesTrak TLEs, % ROI coverage, ESA acquisition-plan cross-check, DST-aware UTC/local-time CSV reports and an n…
A GRAB A standalone SysEx configurator for ROLI MPE controllers that removes dependence on ROLI Dashboard and its servers. Narrow and honest, aimed at owners of orphaned hardware; useless unless you own a Seaboard, Rise or Block.
Open, standalone config tool for ROLI MPE devices (Grand, Rise, Airwave, Block) over MIDI SysEx — no ROLI Dashboard/servers needed.
A esp32-usb-to-airprint Turns an old USB printer into an AirPrint/AirScan network printer with a cheap ESP32-S3, from an active llama.cpp contributor. Concrete, low-cost and immediately useful to Mac/iOS households; limited to USB printers.
Add AirPrint / AirScan support to your old printer, using just a cheap ESP32-S3
A pngdeer A Rust PNG lossy compressor with both CLI and browser use, honest enough to cite a checkable '8-12% smaller than TinyPNG at equal quality'. Good for bulk image shrinking; limited to PNG with automatic quality choice.
Smaller PNGs, same picture. Lossy PNG compression that picks the quality for you: 8–12% smaller than TinyPNG at the same quality. Command line and in the browser.
A trompe Textures a greybox game level with image models by planning cameras, painting, projecting and baking to glTF with real shadows. A rare angle that saves level artists real work; at 1.8 days old the pipeline still needs hands-on testing.
Texture a whole game level from a gray blockout with image models: plan cameras, paint, project, bake. Real shadows, glTF out.
A HermesOBD A native SwiftUI OBD-II monitor for macOS speaking to ELM327 over BLE, serial or Wi-Fi, with logging, trouble codes and a simulator. Concrete and well-scoped, though it lives or dies by the hardware and has no user feedback yet.
Live OBD-II monitor for macOS (SwiftUI): ELM327 over BLE/serial/Wi-Fi, CSV logging, trouble codes, simulator, optional Hermes analysis
A BBTree A C# port of bbtree, a B+-tree with SIMD inline-key search giving .NET fast ordered maps. One job, narrowly and honestly described; it matters only if you are pushing lookup performance in .NET.
Fast ordered maps for .NET: a C# port of bbtree, a B+-tree with SIMD inline-key search
A GEOInflow Self-hosted analytics tracking AI crawler hits and AI-search referral traffic, aimed at people doing generative-engine optimization. The scope is clear and timely, though the co-publisher framing and fast star growth need a real deployment check.
Self-hosted web and AI crawler analytics for Generative Engine Optimization (GEO). Track visitors, AI search referrals, crawl logs, content coverage, and conversions. Jointly published by 猎看科技 · 北京日报社…
A tv-fresh A root-free Xiaomi TV cleanup skill: ad removal, bloatware pruning, launcher swap, plus firmware reverse-engineering notes. Concrete and honest about reversibility, though it only covers MIUI TV and model quirks are on you.
tv-fresh TV refresh: an Agent Skill for Xiaomi TV (MIUI TV) to remove ads / clean preinstalled apps / change launcher / install proxy and media repos — root-free and reversible, with firmware reverse-…
A ex-ussr-kinship-research An evidence-first genealogy skill for tracing former-USSR church, military and deportation records, replying in the user's language. Unusually specific for family historians, though narrow and dependent on how well the archives are actually wired in.
Evidence-first genealogy skill for AI agents (Codex, Claude Code): church books, civil, military, deportation and repression records across the former USSR. Replies in your language.
A secbrief Deterministic OSV/KEV matching with the LLM only explaining, giving agent-managed machines an auditable daily vuln brief. Useful if you already run automated ops; coverage still depends on upstream feeds.
A daily vulnerability briefing for AI-agent-managed machines — deterministic matching against OSV + CISA KEV; the LLM only explains. MIT
A fp4Attn FP4 attention forward and backward plus analysis of stochastic rounding and softmax gradients. Concrete and checkable for low-precision training researchers; still far from drop-in production kernels.
CroissAttn: FP4 attention forward and backward, with analysis of stochastic rounding, zero-sum softmax gradients, and sampled centering.
A FireFrame An ad-free launcher for Fire TV Stick plus one adb command to strip Amazon ads, tracking and bloatware. Practical for fed-up owners, though Wolf Launcher exists and the tacked-on Claude Code line is padding.
A clean, ad-free home screen for the Amazon Fire TV Stick. No root. Custom launcher with a glass app dock, plus one command that removes Fire TV ads, tracking and bloatware over adb. Works with Claude…
A OpenTV-Cast Brings an AirPlay receiver to Android TV and Google TV so iPhones and Macs can mirror. A real need with few options and non-trivial protocol work; stability and latency remain unproven.
Open-source AirPlay receiver for Android TV and Google TV. Mirror from iPhone, iPad or Mac.
A vps-forge An agent skill that hardens a Debian/Ubuntu VPS — SSH, firewall, fail2ban, Docker, TLS, backups, alerts — claiming it will not lock you out, benchmarked over 120 VM runs. Convenient, but still a script bundle.
Agent Skill that lets Claude Code, Codex, Cursor, Copilot and other AI agents set up and harden a Debian/Ubuntu VPS without locking you out: SSH, firewall, fail2ban, updates, Docker, TLS, backups, pho…
A wh1000xm4-anc-optimizer Parses WH-1000XM4 firmware partitions and biquad coefficients to calibrate ANC, with a dry-run simulator and flasher. Impressively specific for tinkerers; real bricking risk, one headphone only.
Sony WH-1000XM4 ANC calibration tool: Airoha NVDM Partition 6 parser, CXD90050 Q28 biquad IIR synthesizer, dry-run simulator, and RACE/ATCI flasher
A samshush Roots a Samsung TV and disables mics, ads and other questionable services — small, clearly scoped, from a known author. Real privacy value, at bricking and warranty risk on specific models.
Root your Samsung TV and turn off the microphones, ads & other questionable services.
A tinysd A dependency-free C library doing only SD cards and SDIO on bare metal, with clear scope and limits. A handy building block for driver authors; the constraint is that FatFs-based stacks already cover this ground.
Dependency-free C library for SD cards and SDIO on embedded and bare-metal systems.
A gdirect Turns Google Drive share links into aria2 direct downloads, useful for anyone pulling large files without browser throttling. Pinning libaria2 shows awareness of the compatibility pitfalls, but gdown already covers much of this; quota handling is unproven.
Rust Google Drive direct downloader using pinned libaria2
A quantpicker Measures rather than estimates: real speed and KL-divergence quality loss for each GGUF quant on your GPU, then says which to run. Immediately useful to local-model users; limited to GGUF and single-machine setups.
Measures which GGUF quant of a model to run on your GPU: real speed and quality loss (KL divergence), not estimates.
A unbloat-android-tv A one-click Android TV slimmer that strips ads and telemetry and swaps in a minimal Apple TV-style launcher, explicitly reversible. Genuinely useful for streaming-box owners; limits are ADB dependency and TV-only scope.
Slim down any Android TV (ads, telemetry) + minimalist Apple TV-style home screen. Reversible, one double-click. 🇫🇷 Allégez n'importe quelle TV Android (pubs, télémétrie) + écran d'accueil minimaliste…
A broll-mcp An MCP server that searches, previews and cuts public-domain b-roll from the Internet Archive with provenance per clip; narrow, honest scope and immediately useful to editors already driving Claude.
MCP server to find, preview and cut public-domain b-roll from the Internet Archive, with a provenance record for every clip
A open-iphone-use Lets AI agents drive a real iPhone over macOS Bluetooth HID with local screen sharing and an MCP layer, sidestepping simulators and jailbreaks. A fresh, clearly described route for mobile agents; limited by macOS dependency and setup effort.
Give AI agents a real iPhone through macOS Bluetooth HID, local screen sharing, and MCP.
A xss-audit Packages XSS static checks as an agent skill with 14 rule layers for React/TS/JS projects, useful as a quick pre-CI pass over frontend render paths. More granular than generic scanners, but the star burst is suspicious and false positives need testing.
Static XSS audit skill with 14 layers of checks for React, TypeScript, and JavaScript projects.
A whatsmyagentdoing Renders every Claude Code and Codex session on your machine as a tiny live office, answering 'who is doing what' when running several agents at once. Charming and lightweight, but purely visual: no scheduling or log search.
See what every Claude Code and Codex agent on your machine is doing — live, in a little office.
A caffeinate-plus A tiny macOS CLI that keeps the laptop awake after the lid closes, covering a case the built-in caffeinate does not. Narrow scope, honest description, immediately usable; several alternatives exist, but this one stays minimal.
Small macOS command-line utility to keep your MacBook awake after you close the lid.
A sam3-segmentation Wraps Meta's SAM 3 into an agent skill for outlining, masking and counting objects on your own NVIDIA GPU, with review and export. Saves work for batch image annotation; limited by needing the GPU, weights, and the upstream model.
Agent skill: outline, mask and count objects in photos with Meta's SAM 3 on your own NVIDIA GPU, with review, exports and a viewer. Works with Claude Code, Codex, Cursor, Pi and other Agent Skills cli…
A openMapper A Rust map editor for OMSI 2 modders: draw roads, place objects, sculpt terrain and test-drive the result. Narrow, honestly stated scope; the catch is you must own the original game and the audience is tiny.
A simple map editor for openOMSI: draw roads, place objects, shape the ground, drive the map in openOMSI (requires the original OMSI 2)
A old-phone-tailnet A concrete recipe and scripts to turn an unrooted old Android phone into an always-on node in a Tailscale network via Termux, key-only SSH and wireless ADB. Practical for self-hosters; fragile against OS updates and Termux quirks.
Connect an old Android phone without root to a Tailscale private network: Termux + userspace Tailscale + public-key-only SSH + on-device wireless ADB + Chrome debugging port. Includes prompts and a Cl…
A gpsp-3d A 3DS GBA emulator based on gpSP that renders layered stereoscopic 3D and runs at full speed even on older 3DS models. A rare treat for 3DS homebrew users, with the base and performance goals clearly stated. The limit is serving only a small niche.
GBA emulator for Nintendo 3DS with per-layer stereoscopic 3D. Full speed on Old 3DS. Based on gpSP.
A astro-conversion-kit A migration tool packaged as an Agent Skill that converts WordPress or any live site into an Astro static site, emphasizing URL, SEO, content, and pixel-level matching, callable from Claude Code and Codex. Useful for site owners leaving WordPress; but pixel-level matching is hard to verify and quality depends on site complexity.
Agent Skills that convert a WordPress site (or any live site) into a static Astro site with URL, SEO, content and pixel parity, then take it through launch. Works in Claude Code, Codex and any agent t…
A lantern A zero-dependency TypeScript library that renders code execution as a step-through playable SVG, suited to people writing technical articles and teaching docs. Output is static and runs in browsers and Workers, with a restrained scope; the young ecosystem is the main uncertainty.
Animated, steppable code explainers. Zero-dependency TypeScript, SVG output, browser and Cloudflare Workers.
A codex-quota-bar A macOS menu bar app showing remaining Codex quota, with an optional desktop overlay and bilingual docs. For people watching their quota while coding, it installs and works with a small scope. Several quota monitors exist; this one wins on being light and hassle-free.
macOS Codex quota menu bar with an optional desktop overlay. 中文 / English guides.
A xinliqiu-sdk An unofficial SDK for OPPO's Heart Ball that pairs with any computer, syncs and decrypts your own recordings, and ships an MCP server for agents. For people who bought the hardware but are stuck with the official app it is the only way out, with a clear scope; the limit is reliance on an unpublished protocol that a firmware change could break.
Unofficial SDK for the OPPO AI 心力球: pair from any computer, sync and decrypt your own recordings, optional ciphertext-only storage, MCP server for agents.
A tellyscope Manage an Android TV from your phone: uninstall preinstalled apps, install apps, transfer files, under 3MB, no root, no tracking. The description is restrained and honest, a TV tool that does one small thing cleanly; the limit is Android TV only, same LAN required, and adb commands can do the same.
Debloat your Android TV, install apps, and browse and transfer files on it, all from your phone. Under 3 MB, no tracking, no root, no PC, no AI.
A diy-fido-key A FIDO2 hardware key built on ESP32-S3 with an ST7735 status display, firmware based on pico-fido, aimed at hardware hobbyists who want to solder their own and see key state changes; wiring and flashing steps are cleanly written. Several similar open-source options exist, and the only difference is that screen.
DIY USB FIDO2 security key on ESP32-S3 (pico-fido + ST7735 status display)
A me-uitk-particles Adds ParticleSystem support to Unity UI Toolkit. UI Toolkit has long lacked particle effects, a real small gap, and the scope is restrained; Unity developers doing UI animation with UI Toolkit should take a look, at the cost of wiring the render pipeline themselves.
ParticleSystem for Unity UI Toolkit
A mac2jb Uses an Apple Silicon Mac as a USB HID keyboard to send the ns2jb bootloader to a Switch 2 and automatically run follow-up commands. A practical piece for the young Switch 2 homebrew scene, with a concrete, clearly explained method; the limit is serving only this one chain and needing the right hardware.
Use your Apple Silicon Mac as a USB HID keyboard to transfer the ns2jb bootstrap and automate post-bootstrap commands on Nintendo Switch 2.
A AARE An open-source PoC in Kotlin replicating the Android Auto phone-side protocol, for head-unit modders and reverse engineers; before this there were only closed-source apps and scattered protocol notes. The limit is that it is a PoC with little feature coverage, so do not expect it to replace the original.
A proof-of-concept opensource implementation of Android Auto's client side (phone side)
A RADWinUI A framework bringing WinUI 3 to Delphi, with an IDE plugin and control panel. For the few still maintaining Object Pascal desktop projects, a path almost nobody takes; scope is restrained and naming is honest. The risk is that the Delphi ecosystem itself is shrinking.
A WinUI-3 framework for Delphi + IDE Plugin and Tool Palette components.
A assembly-nist An independent AlphaNIST reproduction of assembly policy learning for a Franka arm in MuJoCo, with PPO training code and pretrained weights. It saves reproduction effort for researchers in fine robotic manipulation, and the description is restrained rather than hyped, but it covers only a single task in simulation with no real-robot validation, and the community is still small.
[Assembly NIST] Assembly policy learning for a Franka arm in MuJoCo. Independent implementation of AlphaNIST with PPO, training code, and pretrained policies.
A mcd-diet-guard Translates dietary restrictions into real-time McDonald's menu options—what you can order, what works with modifications, and how to swap combos—and even prices and places the order. A real need for people with allergies or religious restrictions, not another chat wrapper; the limitation is tight coupling to domestic store APIs, useless abroad.
Helps friends with dietary restrictions order safely at McDonald's. No pork, vegetarian, seafood allergy, lactose intolerance, no spicy, no pickles… say one sentence and it checks the entire store's r…
A lglow Replaces the ad-filled home screen on rooted 2019 LG TVs with a clean one, with an install script and bilingual English/Spanish instructions. A must-have little tool for owners of these TVs, and the scope is restrained; similar projects include GlassHouse and today's lightkeeper, and this one wins on more complete docs.
LGlow: bring your old LG TV back to life. A clean, fast, ad-free home screen for rooted 2019 LG TVs (webOS 4.5). Installer, guides in English and Spanish, and before/after.
A redivivus Function-by-function reconstruction of Diablo II 1.14d's Game.exe into compilable C++, replicating the original binary rather than rewriting it, which is valuable reference for old-game reverse engineering and porting research. It needs your own game files and has blurry legal boundaries, so short term it suits research, not play.
Diablo II 1.14d recovered as C++ source that builds. Not a reimplementation: the shipped Game.exe, function by function.
A manga-generator A comic pipeline for Codex that goes from original script to storyboard images, layout and PDF export, breaking one concrete task down fairly clearly. It suits authors who want to batch-produce short or fan comics, and 60 stars in two days shows people are trying it, but it depends heavily on model image quality and a Japanese workflow, and layout still needs manual checking.
A manga production tool for Codex that goes from original work to script, image generation, typesetting, and PDF output
A Vesalius A codebase graph for Claude Code: a local code graph plus an MCP server, CLI, Mac app, and VS Code extension. Shipping four clients in 1.8 days makes it hard to do any well; code-graph MCP alternatives like Serena exist, and 64 stars look more like concept hype.
A map of your codebase for Claude Code: an on-device code graph with an MCP server, a CLI, a Mac app and a VS Code extension.
A dafeiyu-cleaner A browser extension that uses offline rules plus the DeepSeek API to identify and bulk-clean AI bot accounts in Bilibili comments. The problem is real and the approach specific, more focused than generic content filters; the limit is unverifiable accuracy and a real risk of flagging human comments, and the copy is marketing-heavy.
DeepSeek Blue Big Fat Fish Elimination Plan: a Chrome / Edge / Firefox browser extension that identifies and cleans up bot accounts (big fat fish) driven by the DeepSeek large model and AI Agents in B…
A oss-verify Registration and build instructions for formal verification of open-source fuzz targets, applying toolchains like Kani and CBMC to specific repos. It suits developers maintaining C/C++ parsers who want to add a layer of proof to fuzzing, but it's mostly process documentation with narrow coverage, and results depend on upstream tool maturity.
OSS Verify: enrollment and build instructions for formally verifying open-source fuzz targets
A php2shell-scanner A high-fidelity scanner for the two php2shell CVEs from the Assetnote team, whose detection logic is more credible than the flood of bandwagon PoCs. CVE hype plus vendor backing brought 287 stars in two days; it can drop straight into ASM or vuln-scanning product lines, making it the most commercializable, though demand will decay fast once official patches ship.
High fidelity scanner for php2shell (CVE-2026-109618 and CVE-2026-109619)
A php2shell-bypass-aslr-exp This is a full exploit for the PHP session.upload_progress UAF (CVE-2026-109618) that needs no official-version changes, bypasses ASLR and reaches practical, directly working PoC level. Nearly a thousand stars in a day show security researchers grabbing a reproducible exploit; monetization means scanner, honeypot and red-team integration, with value dropping once the official patch lands.
A php2shell exploit that requires no modifications to the official version, and does not require display_errors or limit_memory.
A hatchzip It deletes already-extracted archive parts as it unpacks, so you only need space for the output file, supporting zip, 7z, tar and split volumes. Aimed at people with tight disks unpacking hundreds of GB, it saves more space than typical extractors. The catch: once sources are deleted, errors can't be retried, and maturity needs real testing.
Extract huge archives in place: the archive shrinks as your files hatch, so you only need room for the output. ZIP, 7z, tar, split archives.
A limitbar A macOS menu bar app showing usage quotas for Claude Code, Codex, Gemini, and Grok, explicitly read-only on local state and never touching credentials. For AI coding users constantly hitting rate limits, it's a small tool you can install immediately, with restrained scope.
AI coding limits in the macOS menu bar — Claude Code, Codex, Gemini (Antigravity), Grok. Never touches your credentials.
A rekey Re-signs an API key with one command and syncs updates to 1Password and references in templates, for developers who rotate keys often. Scope is restrained and the description honest, exactly the small and specific kind of tool; the limitation is it only covers the 1Password plus template path.
Reissue an API key with one command and update it everywhere it is used (1Password + templates)
A php2shell-scanner A detector for the newly disclosed PHP2Shell (CVE-2026-109618/109619) covering URL black-box, local device, and SSH paths, from an author with a track record in security tooling. It's directly usable for incident triage; the limitation is it only covers this one CVE.
Scanner for PHP2Shell (CVE-2026-109618/109619): URL blackbox, local device and SSH.
A aidbg A Windows x86/x64 command-line debugger built for AI calls, with JSONL output, an MCP server, anti-anti-debugging, snapshot rollback and PDB types. The feature list is concrete and well-scoped, filling a gap for automated reverse-engineering agents; the downside is barely a day of history, unverified stability, and x64dbg's command-line mode as the nearest comparison.
AI-first Windows x86/x64 command-line debugger | JSONL output, MCP server, anti-anti-debug (ScyllaHide-style), snapshot rollback, IAT rebuild, PDB types
A CinderVM A deterministic bytecode VM for agent execution, with image verification, snapshots and hash-chain replay logs, aimed at infra developers building agent sandboxes and reproducible execution. The angle is novel but still early, and whether it handles real workloads is unverified; wasm sandbox approaches exist in a similar direction.
Deterministic bytecode VM for agent execution - verified images, snapshots and a hash-chained replay journal
A JoyOff A Switch homebrew that returns original Joy-Con to factory low-power shipping mode and disconnects them, fixing battery drain and pairing confusion from long docked sessions. Clear goal and small codebase make it a clean little tool for homebrew users, though ordinary users won't need it.
A simple Nintendo Switch homebrew app that puts original Joy-Con controllers into shipment low-power mode and disconnects them.
A mcd-advisor Uses McDonald's China official pricing data to find cheaper ways to order the same meal combo, packaged as a zero-dependency CLI plus MCP and Agent Skill. Small scope and honest description make it directly useful for regulars, but it's locked to China-region APIs and useless abroad.
Mai Advisor: keep the same cravings, use McDonald's official pricing to find a better way to buy this meal (McDonald's China MCP · Agent Skill · zero-dependency CLI)
A custom-tv A custom desktop for Samsung Tizen TVs: weather, scenery wallpapers that change with weather, app shortcuts, plus steps for taking over the Home key. People who hate TV ads and the stock launcher will like it; the limitation is Tizen-only, requiring developer mode sideloading.
A calm custom Home screen for Samsung Tizen TVs: weather, weather-matched landscapes and your apps, with guides to make it the Home button.
A Looking-Ahead A paper repository from Tencent ARC researching future-evolution modeling in autoregressive video generation, offering reference value for teams working on video generation. It has institutional backing and a restrained description; the limitation is that such repos often hold only a paper and a few scripts, so reproduction must wait for full code.
Looking Ahead: Modeling Future Evolution in Autoregressive Video Generation
A afnon-android-tv-launcher Turns prayer times, qibla direction, adhan reminders, and Quran radio into an Android TV home screen, with offline data for 74 cities as a solid detail, aimed at Muslim families in the living room. Scope is restrained and the description doesn't oversell; the downsides are sideload-only, limited Google TV/TCL compatibility, and while phone apps abound, TV versions are rare.
Afnon (أفنون): Android TV launcher and home screen for Muslims. Offline prayer times (74 cities), Qibla, adhan/azan, prayer reminders, Ramadan and Friday modes, 16 Quran radio stations and reciters. A…
A UltraSlim An Android tool for offline batch transcoding of JPEG/HEIC/AVIF/JXL while preserving Ultra HDR gain maps, for people processing photos on their phone without uploading to the cloud. Preserving gain maps is more careful than most transcoding apps; the limitation is transcoding only, with no editing.
Android offline photo transcoding tool: convert between JPEG / HEIC / AVIF / JXL, preserve Ultra HDR / ISO 21496-1 gain maps, supports batch and background conversion
A php-eol-CVE-2026-109618-109619 Docker images for end-of-life PHP 7.4/8.0/8.1 with backported fixes for CVE-2026-109618/109619, for ops teams locked to old versions. It's practical and the description is restrained; the limitation is that it patches only one CVE and can't replace a real upgrade.
PHP 7.4 / 8.0 / 8.1 Docker images with the CVE-2026-109618 / CVE-2026-109619 fix backported
A naskah-id An AI skill for Indonesian academic writing that rewrites and proofreads per EYD V and KBBI norms, serving Indonesian researchers. Both language and scenario are specific, more targeted than general writing assistants; the limitation is Indonesian only, and effectiveness still needs real-text validation.
Skill agen AI untuk menulis dan menyunting karya ilmiah berbahasa Indonesia yang natural, baku (EYD V, KBBI), dan jujur.
A synodeb A hands-on record of installing Debian 13 on a Synology DS923+ with working NIC, drive bays, and fan, including the official installer over SSH. It genuinely saves time for the few wanting to escape black Synology. The limit is it only covers one model and reads more like notes than a general script.
Debian 13 on the Synology DS923+: every piece of hardware working, installed with the official Debian installer over SSH
A confess Before AI-written code is committed, it makes you explain the change in your own words and blocks you if you can't. Aimed at people who code with AI but can't read the code, it's a clever angle. The limitation: the repo offers only a slogan, with no detail on how it blocks or which hook it uses.
Your AI wrote the code. You still have to explain it before it ships.
A rl-env-use Uses a Browser Use agent to mine real web pages for cases and auto-generate RL environments, solving the lack of real task distributions for agent training. Its 102 stars in 0.8 days ride the browser-use brand and RL training hype; the path is a paid API for environment generation as training data or evaluation; the risk is questionable environment quality and reproducibility, competing with in-house benchmarks.
Build RL environments for any domain from real cases that Browser Use agents mine from the web
A kindi An inference engine written for old cards like the Quadro K420, running ternary-quantized 2B models at 20+ tokens/s with an OpenAI-compatible service. Squeezing usable speed from old hardware is a rare approach and the numbers are concrete; the limits are targeting just one card and one model.
A custom LLM inference engine for the NVIDIA Quadro K420: Ternary-Bonsai-2-2B at 20+ tokens/s with an OpenAI-compatible server
A Show HN: Computer use on the phone itself: a 2B model that operates Android apps Runs a 2B small model on-device to drive Android app operations, solving GUI agents' latency, privacy, and API costs from cloud LLMs. Its 12 points are low but on-device computer-use is a real gap and easy to spread; paths are OEM preinstallation, accessibility automation, or mobile automation testing SDK licensing; the biggest risk is Google's own on-device agent and system permission lockdowns.
A ps5-bazzite-deck Brings Bazzite's Steam Game Mode image to the PS5, with Game Mode, VCN hardware decoding, and Wi-Fi all working. It is real progress for people tinkering with PS5 Linux; the limits are coverage of only specific firmware and models, with a high barrier for ordinary players.
Bazzite Deck (Steam Gaming Mode) images for the PS5 with working Game Mode, VCN and Wi-Fi
A video-to-blender Converts reference videos or images into editable Blender procedural node trees, with a tested library of Geometry, Shader and compositing nodes, packaged as a coding-agent skill. People recreating 3D references save a lot of node-building time; the limits are dependence on a specific agent client and 16 stars with no verified results.
Turn a reference video, image or idea into an editable procedural Blender node system: a coding-agent skill with a tested Geometry/Shader/compositor node-group library
A HistorianRCA A root-cause-analysis benchmark built from 90 days of simulated chemical-plant historian data, with faults buried for users to locate. Industrial time-series RCA has long lacked a public reproducible dataset, so process-control and time-series diagnostic teams should take a look; limits are a single scenario and small scale.
HistorianRCA: a root cause analysis benchmark. 90 days of simulated plant historian data from a steam-heated hold tank, with hidden faults to find and diagnose.
A matinee A quiet, ad-free launcher for Google TV and Android TV, small in scope with an honest description, and TV launchers are indeed a long-standing pain point. Mature alternatives like Projectivy exist, so it will stand or fall on polish.
A calm, ad-free home screen for Google TV and Android TV
A freecad_solidworks_plugin Lets FreeCAD open SolidWorks SLDPRT/SLDASM and Parasolid files directly, saving a paid conversion step. Mechanical design has long lacked this path and the scope is clearly stated; the open question is fidelity on complex assemblies, which nobody has verified.
FreeCAD addon: open and import SolidWorks parts (.SLDPRT), assemblies (.SLDASM) and Parasolid files
A php2shell-lab Packages offline reproduction of a PHP RCE vulnerability and the detection rules written from it into a rerunnable Shell lab, explicitly excluding exploits and requiring your own samples, which is directly useful for blue-team detection engineering. The scope is restrained and the description honest; the limit is that it covers a single CVE and you still have to find samples yourself.
Offline, reproducible lab to safely analyse an untrusted PoC for CVE-2026-109618/109619 and build detection rules. No exploit included; bring your own sample.
A OpenSWOS Rewrites Amiga's Sensible World of Soccer in C as a cross-platform, network-playable open-source version; classic game remakes are always a community need. The risks are copyright on original assets and doubts about completeness for a repository one day old.
An open-source C port of Sensible World of Soccer (Amiga), for Windows, macOS, Linux, Amiga and the web, with online multiplayer.
A toolkit A pure Go middleware that adds tool-calling capability to any LLM API, so Go backends can use function calling without binding to the Python ecosystem. The positioning is clear in one sentence and the scope restrained; the challenge is the ecosystem advantage of mature gateways like LiteLLM.
A pure Go middleware library that adds tool-calling capability to any LLM API.
A quiet-tizen A native, clean launcher and reversible debloat tool for the Samsung U8000F/Tizen 9, with notes written for agents to read. Targeting one specific model feels honest, but it only works on that TV and touching system partitions is risky.
Native Quiet Home launcher, reversible cleanup tools, and an agent-readable guide for the Samsung U8000F running Tizen 9.
A torrentplayer A desktop torrent player written in Rust that streams while downloading and seeks in seconds, explicitly offline with no account or telemetry. It's genuinely convenient for people who watch torrented video often, but Stremio and WebTorrent already cover similar needs, and stable decoding is the real test.
Stream any torrent, video or audio, and skip to any moment in about a second. Free, open-source desktop app for Windows, macOS and Linux. No servers, no accounts, no telemetry.
A ComfyUI-Ming-3-tts Wraps Ming-3.0-tts as a ComfyUI node so video dubbing users can generate speech directly in their workflow. The scope is restrained and the description honest, but it is essentially a port whose ceiling depends on the upstream model, with similar TTS nodes as alternatives.
Native ComfyUI port of Ming-3.0-tts, supporting 0.5B and Tiny 9B speech generation.
A EAGLE-4 The official inference code for the EAGLE family of speculative decoding, directly determining LLM inference throughput and a key component likely to be integrated into inference frameworks. Its 554 stars and fast growth show inference acceleration is the hardest need. The path is becoming the default draft model for vLLM and SGLang; risks are reproduction quality and competition from Medusa and others.
Official Inference Code of EAGLE-4
A fx0 A lightweight design agent that produces a single HTML page from one description, essentially a minimal v0. Its 337 stars come half from vercel-labs backing and half from single-file delivery; it could fold into the Vercel ecosystem for traffic and paid credits, but the risk is head-on competition with mature products like v0 and Lovable.
A lightweight design agent built with libfx. Describe a page and an agent writes it as one HTML file.
A aerial-tv Tackles boot ads and bloated launchers on Chinese smart TVs by taking over the boot process with one click, and it genuinely works. 1,619 stars in a day show how many users are tormented by TV ads; the path is preinstallation with ROM vendors or box makers and paid custom versions, while the risk is vendor blocking and fragmented adaptation across ROM differences.
A clean launcher for China-market TVs: an open-source Android TV launcher designed after tvOS, no ads, auto-takes over on boot. A tvOS-style launcher for Xiaomi, TCL and other China-market TVs.
A ra-jit A Rust compile daemon built on rust-analyzer, offering incremental cargo builds and hot reload while running, targeting Rust's long-standing slow compile pain. The 171 stars reflect the community's obsession with iteration speed; it could become a team-level cache service or IDE integration. The risk is being swallowed once rustc's official incremental compilation and hot reload improve.
Rust compiler daemon on rust-analyzer: fast incremental cargo builds and live edits to running programs. Try: nix run github:indexable-inc/ra-jit
A Show HN: VoidAccess – open-source dark web OSINT, from query to a graph Automates the whole pipeline from dark web collection and denoising to NSFW filtering and relationship graphs, solving analysts' extremely inefficient manual dark web browsing. Its high 127 points reflect real security-team demand, and it could target Maltego and Recorded Future as an open-source intelligence platform with subscriptions; the biggest risks are compliance of dark web scraping, data quality, and upstream site blocking.
Hi HN! Im Katriel, and im a Security Researcher/Enthusiast and in the wave of AI, i wanted to build that next gen of security tooling that surpasses the current tools, and even the paid industries wit…
A ElyOffice An offline-first desktop office suite in Rust and GPUI, targeting writing and spreadsheets where data never leaves the device. Its 674 stars show real developer demand for lightweight native office tools. Monetization could be free for individuals plus paid team collaboration, or becoming a GPUI ecosystem sample; the risk is the huge engineering effort of an office suite, competing with LibreOffice and Feishu.
Ely Office: offline-first desktop apps in Rust and GPUI
A ai-live-avatar Assembles real-time digital humans, danmaku filtering, outfit changes, and OBS overlays into a runnable AI livestream room, directly serving e-commerce and unmanned livestreaming. Its 261 stars come from the real paid scenario of digital-human livestreaming. The path is selling templates, managed operations, or per-stream billing; risks are reliance on closed APIs like Vidu and platform regulation of digital-human livestreams.
AI real-time livestream room: Vidu S2 real-time digital human + danmaku filtering and scheduling + wardrobe changes + OBS overlay (character assets not included).
A Ming-3.0-tts Ant Group's inclusionAI open-sourced a TTS model family with a 0.5B dense and a 7.9B MoE model, supporting multiple languages, voice cloning and natural-language control. The 126 stars reflect the lack of commercially usable large open speech models; the path is cloud API or on-device SDK. The risk is head-on competition with established options like CosyVoice and F5-TTS.
Ming-3.0-tts is an open-source TTS model family with 0.5B dense and 7.9B-A1.3B MoE variants, supporting multilingual speech, voice cloning, voice design, and natural-language control.
A OneLineage Uses a user-space pivot_root container to let rooted Samsung phones hot-switch between One UI and LineageOS GSI without repeated flashing. The idea is novel but the audience is narrow, only useful to Samsung users who tinker with ROMs; partition and rollback risks are not yet explained.
Hot-switch a rooted Samsung phone between One UI and a LineageOS 23.2 GSI via a userspace pivot_root container
A whistle-esp32 Runs fully offline speech recognition on an ESP32-S3: hold a button, say a color, choose from 7 languages, and the onboard LED responds. The scope is tiny and the goal clear, a good example of embedded STT; the limitation is a color-only vocabulary and model and memory tuned per board.
Whistle speech-to-text running fully on an ESP32-S3: hold a button, say a colour in one of 7 languages, the LED does it. No network.
A one-click-eu-lord A 166KB single-file tool that unlocks the features the EU DMA brought to Windows 11 while keeping what the EU version removed, persisting after updates with one-click restore. It's practical for people who want to try features without breaking their system; similar registry scripts abound, and its selling point is solid reversibility.
One-Click EU Lord — give Windows 11 every gain from Ireland's EU/EEA (DMA) policy without the features the EU build removes, and keep it across updates. One 166 KB exe, no runtime, fully reversible. T…
A tcl-tv-toolkit A debugging toolbox for China-market TCL and Android TV: ADB authorization, sideloading, proxy, ad removal and a status HUD in one place. It addresses concrete annoyances of specific models with solid documentation; the downside is scattered features and no unified entry point, with value mainly in China-market adaptation experience.
China-market Android TV debugging and tuning toolbox: ADB auth, sideload, proxy, debloat, status HUD | China-market Android TV toolkit: ADB auth, sideload, proxy, debloat, status HUD
A SimpleCAD Adds SketchUp-style direct modeling to FreeCAD: push-pull faces, edit faces, and convert STL meshes into editable solids. It addresses a clear pain point for CAD users put off by parametric workflows, with restrained description and a single goal; but the repo is under a day old, so installation and completeness remain unclear, so wait and see.
Direct modeling for FreeCAD, in the spirit of SketchUp: draw, push/pull and edit faces directly, and turn STL meshes into editable CAD.
A luma-tv A minimalist tvOS-style launcher for Android TV and Google TV, explicitly ad-free and tracking-free, replacing the official recommendation feed. Anyone tired of the Google TV home screen can just install it, but it competes with FLauncher and Projectivy for the same ground, and the information so far is only screenshot-level.
A calm, tvOS-inspired home screen for Android TV and Google TV. No ads, no tracking.
A theboringtv-lgwebos Gives LG webOS TVs a Google TV-style home page, with a single clear goal and clean description rather than platform hype. It has real value for people who find the native home page hard to use but don't want a new TV; limits are that installation likely requires developer mode and compatibility is unverified, though a few similar launchers already exist in the webOS community.
Google TV-style home screen for LG webOS TVs
A frameo-homeassistant-dashboard Turns a cheap Frameo digital photo frame or idle tablet into a Home Assistant wall panel with a tiny kiosk app that solves one specific thing. The scope is restrained and the description plain, and it works once installed; the limits are that it only serves HA users and several similar community solutions exist, though this one wins on lightness.
Fast Home Assistant wall dashboard for a Frameo photo frame or Android tablet, in a small kiosk app
A ComfyUI-CharacterCardStudio A ComfyUI node that generates character turnaround sheets or setting cards in one click and auto-splits and reassembles them, targeting Qwen-Image 2.1. People doing AI character design can install it and skip manual cutting and alignment; the limitation is being locked to a single model version, requiring changes when the model updates.
ComfyUI custom node: one-click character reference sheet (character card) generation, auto slice & recomposition for Qwen-Image 2.1
A Show HN: KiviDB, a faster Redis-compatible memory database and vector cache Addresses Redis memory bloat after snapshots, OOM and weak multithreaded throughput with a Rust and io_uring compatible replacement that includes vector search. 258 stars, 62 points and benchmarks against Dragonfly give it momentum; a hosted cloud service could charge, but the AGPL core faces Redis, Valkey and Dragonfly, and the last 1% of command compatibility is the deadliest gap.
Hi all, So I have been working on this project KiviDB from February 2026. It started as a project for me to understand rust, having a product with no GC pauses, memory blow up after snapshots and gett…
A pdf-figure-vector Converts figures from paper PDFs into vector SVG and editable PPTX with a pixel-level comparison preview, solving the real hassle of redrawing charts for papers and slides; install and go. The hard part is fidelity on complex charts and embedded fonts; the alternative is tracing by hand in Illustrator.
Convert academic PDF figures into self-contained vector SVG and editable PPTX with pixel-level comparison and live preview.
A agents-contract Turns AGENTS.md conventions into a lintable contract plus a SessionStart hook, usable with Claude Code, Codex and Cursor; it addresses the new problem of rule drift in multi-agent collaboration, with zero dependencies and a restrained scope, though the ecosystem is still very early.
AGENTS.md contract for AI coding agents + zero-dependency linter + SessionStart hook. Works with Claude Code, Codex, Cursor, Copilot.
A Lampline A self-hosted voice agent platform that imports Dialogflow CX agents and handles calls on your own machine via jambonz SIP. It appeals to small teams avoiding per-minute fees, but claiming 20 model providers in 0.7 days suggests an overly broad scope with unproven maturity.
Self-hosted voice agent platform: import your Dialogflow CX and CX Agent Studio agents, build and test them, and take calls on your own hardware. 20 model providers, phone and SIP through jambonz.
A grammai An offline English-German grammar checker with local AI rewriting and translation, giving model choice back to users, a clearer positioning than yet another cloud wrapper, appealing to privacy-minded people who don't want subscriptions. Limitation: only English and German, while the LanguageTool ecosystem is more mature.
AI writing assistance. Free & open source. Offline English and German grammar checks, with rewrites and translation through your own AI.
A SignalScope An Android network signal checker written in Flutter and Kotlin, covering SIM, LTE/5G, Wi-Fi, GNSS and latency tests. It is aimed at field and on-site debugging engineers and is cleaner than installing a pile of carrier apps. It overlaps with Network Cell Info but wins by being open source and modifiable; it lacks iOS and detailed documentation.
SignalScope — An Android network-signal inspector built with Flutter and Kotlin. SIM, LTE/5G, Wi-Fi, GNSS, and latency tests.
A huashu-motion Turns a coding agent into a commercial-grade product promo video generator, with composing and pointing-at-frame shot edits, folding video production into agent workflows. 281 stars in half a day shows one-sentence video output spreads fast. Monetization: templates plus render cloud subscription. Risk: native video models like Sora and Kling moving downmarket, and unstable output quality.
Let your agent produce commercial-grade product launch films in one shot: it composes music and can change things by pointing at the frame. Commercial-grade launch films by your coding agent — point a…
A Show HN: Sphinx – A hook that quizzes you on your AI's code before shipping Pops up a questionnaire before pushing to production to verify you truly understand AI-generated code, targeting the comprehension debt of vibe coding. Its 238 points reflect widespread team anxiety about AI code quality; it could become a CI or PR gate sold per seat, but risks being dismissed as process noise and being built into platforms like Copilot.
FOSS, Basically it plugs into your existing coding agent workflow and every time you push to prod it'll throw up a questionnaire that checks if you understand how the code you're pushing works.
A tv-vibehack Solves Android TV boot ads and bloated systems slowing the experience, using one-click ADB ad removal and a tvOS-style desktop. Nearly two thousand stars in a day spread via Claude Code buzz and the hard need to remove TV ads. Monetization could be a launcher subscription or preinstalled device deals; risks are Google tightening ADB permissions and launchers like Projectivy already occupying the space.
Your TV, minus the ads. Debloat any Android TV over ADB and give it a clean, tvOS-inspired launcher. Built in an afternoon with Claude Code.
A cn-math-taxonomy Breaks the middle-school math curriculum into a knowledge graph with 1,407 problems and adaptive diagnostics, fixing the lack of structured paths for tutoring firms and self-learners. 321 stars in half a day shows K12 structured data is a real need. Monetize via institutional SaaS or a problem-bank API. Risk: content copyright and double-reduction compliance, with incumbents like Yuanfudao holding similar graphs.
Primary and secondary school math knowledge graph + prerequisite diagnostics: curriculum-standard aligned, People's Education Press grades 1–9, 531 knowledge points, 2,271 questions, adaptive diagnost…
A map-tunisia Turns Tunisia's five-level administrative boundaries, down to 2,083 imadas, into SVG/GeoJSON/TopoJSON that frontends for data visualization or local map services can use directly. The levels and counts are stated concretely, a small dataset done thoroughly; limitation: only one country, and no license or rendering examples are visible.
The most accurate open-source SVG maps of Tunisia: country, regions, districts, 24 governorates, 279 delegations & 2,083 imadas · 5 levels of detail · 8 modes · SVG, GeoJSON, TopoJSON
A a7c2-fl23-research Ports FL2/FL3 to Sony A7C II firmware 2.01 with custom LUT/DCP import, including backup, write-back verification and restore flows. For photographers who want to add a look to their camera, it unusually documents test results and known limitations; the limitation is that it covers only this one model.
FL2/FL3 porting and custom LUT/DCP import tools for Sony A7C II (ILCE-7CM2) firmware 2.01, with backup, installation, readback verification, and restoration workflows. Includes single-camera test resu…
A decks Makes slides an agent-first format: leave a comment under a page and Claude Code or Codex edits that page and replies, an interaction closer to real revision workflows than AI-generated PPT. It is currently an HTML approach, with collaboration and export still to be proven.
Agent-first slide decks: comment under a slide, and your coding agent (Claude Code or Codex) makes the change and replies on the slide.
A cve-intel A local Rust program that bundles 31 CVE data sources and 32 analysis tools, outputs risk scores and exposes them to AI agents via MCP. Security teams doing initial vulnerability triage can skip the scripting work, and local-first is genuine; the risk is that data source quality and scoring criteria remain unvalidated.
Local security intelligence for CVE triage - 31 data sources and 32 tools, risk scores and an MCP server for AI agents, all in native Rust. Private by default.
A Table-Tennis-Rockstar-PC Static recompilation brings Xbox 360 Table Tennis to Windows x64 with FSR/FXAA, frame cap and offline menu fixes; you supply the ISO. A locally buildable release for replaying old games on high-refresh screens, with a clearer build process than peers. Risk: recomp ports are crowding, and each game is hand-ground.
Windows x64 recompilation with up to 4K, 60/120/144/240 FPS targets and uncapped mode, FSR/FXAA, gamepad intro skip and offline menu fixes. Builds locally from your own Xbox 360 ISO.
A ha-frame-starter A paper-and-ink style Home Assistant wall dashboard with clock, weather, calendar, blinds and AC, plus a config page that helps write YAML, saving manual layout work for anyone mounting an old tablet on the wall. Scope is restrained and the description is honest; the downside is its dependence on the button-card ecosystem.
A calm paper-and-ink wall dashboard for Home Assistant: clock, weather, calendar, people, lights, blinds and climate on button-card, with optional calendar, energy, Flightradar24, rubbish, appliance, …
A im-mcp Read-only exposure of local WeChat 4.x/3.x and DingTalk chat records as MCP with FTS5 full-text search, letting agents query your own chats; it solves a real problem for people wanting AI to dig through old records. Limits are that database formats change with versions and only local desktop clients are covered.
Fast, local, read-only MCP Server for WeChat (4.x/3.x) & DingTalk (7.x V2/V3) chat history with FTS5 search
A Show HN: Things to Do API Fills the gap where local event data (pop-ups, shows, one-off events) has no unified supplier, exposed directly to agents as MCP, using influencer accounts for fan-out aggregation faster than real-time crawling. 32 points show the pain is real but reach is limited; the author is validating paid demand. Path: become the event data API for the agent era, benchmarked against Eventbrite data services. Risk: narrow coverage, SF-only, easily copied by big tech or ticketing platforms.
I built an api/mcp for local events: pop-ups, one-off events, concerts, etc. currently covering the SF bay area. I've been building a bunch of consumer apps lately around personalization and recommend…
A blender-ai-mcp-addon Adds MCP and a visual chat to Blender so non-technical 3D users can edit scenes in natural language and run multi-agent workflows, removing the bpy scripting barrier. 502 stars in two days ride both the MCP ecosystem and Blender's huge user base. Path: paid plugin marketplace plus studio team subscriptions; once MCP standardizes it could become the default 3D access layer. Risk: Blender's built-in AI and rival copilot plugins.
AI assistant and MCP server for Blender: visual chat, scene tools, multi-agent workflows. Works with OpenAI, ChatGPT, Claude, Gemini, Ollama, OpenCode. Blender AI addon and plugin, copilot for 3D, bpy…
A Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware Solves the information gap where buyers cannot get real tok/s before buying local inference hardware, turning measured results across hardware combos into interactive comparisons. 175 points come from the neutral-data vacuum amid a consumer GPU upgrade wave; exploding in two days validates demand. Path: hardware vendor partnerships, procurement consulting or enterprise POC data services. Risk: official llama.cpp benchmarks and OpenRouter-style leaderboards dilute it, and data credibility may be questioned.
A MOSS-Image A 5B DiT base model from OpenMOSS focused on text rendering and layout generation, long the hardest part of controllable generation. 384 stars in two days come from lab backing and the scarcity of open base models. Path: fine-tune into poster/e-commerce design SaaS, or become the default base for Chinese layout generation. Risk: Flux, Qwen-Image and others already occupy the space, and a pure base model is hard to monetize alone.
Official repository for MOSS-Image, a 5B-parameter base DiT model with high-quality text rendering, layout generation, and efficient fine-tuning.
A reason-network Lets multiple coding agents within a team communicate, solving the context silos of one person running many agents doing their own thing; 170 stars/day is a real signal of demand for an agent orchestration layer. Path is clear: open protocol to set a de facto standard, then enterprise hosting and audit services. Biggest risk: A2A, MCP and other protocols converging fast to grab the entry point, or Claude Code and Cursor building similar orchestration in-house.
Reason Network: your team's coding agents, talking to each other. The engine behind `reason network`.
A Show HN: Gemini 4 Argon plays StarCraft Has LLMs play StarCraft BW by writing code, essentially a public benchmark for agent long-horizon planning and API operation; 233 points come from the natural spread of model-arena topics. Path: become the standard leaderboard cited by eval media and vendors, taking sponsorships, API distribution and consulting. Risk: game benchmarks age fast, and aggregators like OpenRouter building their own leaderboards dilute citation value.
I've refreshed StarSkirmish Bench, a benchmark where LLMs play StarCraft BW by writing code, with results from the flagship models of OpenAI, Anthropic, Google, and Muse
A Show HN: rStation – A shared research lab for your AI agents Turns the agent research process into a shared lab, solving the silo problem where agent exploration results in a team cannot be reused or audited. 54 points show multi-agent collaboration is moving from toy to real need. Path: a subscription collaboration platform for engineering teams, like a shared notebook for agents. Risk: observability platforms like LangSmith and Weights & Biases casually absorbing this.
A leanrea Packages reverse engineering into an 8KB single skill, replacing MCP with a pure CLI, claiming 38% faster and 35% fewer tokens, directly challenging the consensus that MCP is the answer. 231 stars in two days show the lightweight agent-skill route is recognized. Path: enter skill distribution ecosystems or enterprise reverse-engineering audit tools. Risk: existing REA-style skills iterate fast, and a pure CLI lacks structured output, making a commercial loop hard.
One 8 KB skill that replaces REA (Reverse Engineer Anything): plain CLI tools, no MCP. Benchmarked: 38% faster, 35% cheaper.
A 1c-cf-vendor-merge Turns the Russian-region headache of upgrading 1C custom-development configurations into an Agent skill that uses XML export for three-way git merges. The scope is very narrow but the problem is real, and the description is accurate without hype; the limitation is that it only matters within the 1C ecosystem.
Skill for AI agents: updating a customized 1C configuration to a new release via three-way git merge of XML dumps
A claude-eyes Lets coding agents "watch" video: build an index, then skim, search, zoom in and quantify editing style, all running locally; it fills a missing capability for people building video-related agents and is infrastructure, though accuracy still needs verification.
Let coding agents watch videos: index once, then skim, search, zoom, and measure editing style. Runs locally.
A sentryward It automates formal verification for warp-specialized GPU kernels, hitting exactly the gap where more kernels are hand-written but nobody proves correctness. The direction is rare and the description restrained; but the repo is still thin, so it remains to be seen whether it can actually run a CUDA example.
Automatic formal verification for warp-specialized GPU kernels.
A whiteboard-teacher-video A skill that turns explainer scripts into hand-drawn whiteboard videos, drawing shapes stroke by stroke and rendering locally in Chrome without image-generation APIs, so knowledge-video creators can use it directly; it's close in spirit to Manim/motion-canvas but wins on being ready out of the box.
Illustrated explainer video skill** — turns a script into a "hand-drawn animated" explainer video. Graphics are drawn stroke by stroke in code, text is written character by character, and it draws whe…
A opendevicedb A catalog of over twenty thousand devices known to support open firmware (OpenWrt, LineageOS, coreboot and others), with a JSON API and MCP service so people and agents can check before acting; data accuracy is the key to success, and the database is only half a day old.
Open directory of 21,000+ devices whose protocols or firmware are already open (OpenWrt, LineageOS, coreboot, Zigbee, Z-Wave, appliances…), so nobody — human or AI agent — redoes the work. Website, JS…
A OpenGCUA A general computer-use agent stack open-sourced by Tongyi Lab, combining code execution, GUI operation, and cross-device control, usable as a baseline by those doing CUA and automation. For now it's more of a platform prototype; whether it's usable depends on docs and a reproducible demo.
An open-source stack for generalist computer-use agents, combining code execution, GUI interaction, and cross-device control.
A safe-disk-cleaner A GTK4 disk cleanup tool for Ubuntu whose selling point isn't scanning but telling you what is safe to delete. The positioning is restrained and the description honest, making it friendlier than baobab for ordinary users; the limitation is that it only covers systemd-based distributions.
See where your disk space went — and understand what is safe to delete. GTK 4 app for Ubuntu and systemd distros.
A sydrahynth Emulates the Hydrasynth hardware synthesizer on Linux by running its firmware, packaged as a VST3/LV2 plugin; the approach is hardcore and the description restrained, filling a gap for Linux musicians, with risks around firmware licensing and stability.
Linux plugin (VST3/LV2/standalone) that emulates the Hydrasynth Desktop by running its firmware
A quittio An e-invoicing library for Laravel covering ZUGFeRD/Factur-X, XRechnung, UBL and Peppol BIS, implemented per EN 16931. PHP teams doing European B2B billing benefit directly, and the description is precise without hype; competitors are mature packages like horstoeko, so format coverage depth is what to watch.
E-invoicing for Laravel: ZUGFeRD/Factur-X, XRechnung, UBL and Peppol BIS based on EN 16931.
A walk-kcal Converts "how much more can I eat" into "how far I still have to walk," using McDonald's official nutrition values plus nearby restaurants to compute the allowance. The concept is small and concrete, understandable in one sentence; the limitation is narrow data coverage, more fun than long-term practical.
How many calories you can still eat today depends on how far you're willing to walk — a meal decision tool that trades walking distance for calorie budget. Official McDonald's nutrition values × nearb…
A home-care-agent An elderly-care solution for Home Assistant users: five copy-paste automations plus an optional local AI summary alert. It explicitly says "care, not surveillance," showing good judgment; the limitation is tight coupling to HA and sensor placement.
Local-first Home Assistant elder-care safety net: 5 paste-ready automations + an optional local AI care agent (digest + urgent alerts). Care, not surveillance.
A sm64-plaque-decode Decodes the Super Mario 64 courtyard stone tablet texture line by line, including font identification, a forward model of the 1996 texture and full reproduction code; a specific, charming question is taken seriously, suited to game archaeology and reverse-engineering fans, with limited practical value.
Decoding the Super Mario 64 courtyard plaque ("L is real 2401"): font identification, forward model of the 1996 texture, line-by-line decoding, report and full reproduction code.
A yapfast A native Discord client written in Rust, pitched as lightweight and low on memory for desktop users tired of the bloated official app. Several Rust/Go third-party clients already exist; its edge is timing and blunt messaging, with feature coverage the biggest unknown.
The chat app that doesn't eat your PC. A lightweight, native Discord client in Rust.
A Hy-Memory Tencent Hunyuan's team drew 500+ stars in two days, so the attention is real, and the name points to long-term agent memory, a currently needed component. But the listing has no explanation, so it's unclear whether it's a memory framework, training component, or paper code; read the README before investing.
A prototype-switch It uses ReXGlue to statically recompile the Xbox 360 version of Prototype to the Switch with native Vulkan rendering, and explicitly states it contains no game files and isn't sold. Such ports only became feasible in the past two years, so the engineering value is clear; playability remains to be verified.
Prototype and Prototype 2 (Xbox 360) for Nintendo Switch: static recompilation with ReXGlue, native Vulkan renderer. Free, not for sale. No game files.
A ko-paper A Claude Code skill that translates English paper PDFs into Korean while preserving the original layout, built on BabelDOC, for Korean researchers who need to read literature quickly; dependencies are honestly stated, but it's essentially a wrapper around a mature tool.
A Claude Code skill that translates English academic PDFs into Korean while preserving the layout (BabelDOC + Claude)
A exodus-radar Connects a Raymarine Quantum radar to a Raspberry Pi, with AIS, anchor alarms and Signal K integration viewable in a browser or on the Pi screen; open-source options for this marine scenario are very few, the description honestly notes a non-commercial license, and the downside is a very narrow audience.
Raspberry Pi radar display for the Raymarine Quantum: AIS, anchor watch, Signal K, OpenPlotter, mayara. Runs on the Pi screen or in any browser. Free for non-commercial use.
A BRUTAL Solves the C# ecosystem's lack of a low-level game framework that directly connects to Vulkan and native APIs, and it's the real foundation of Kitten Space Agency and the RimWorld Waylith engine, not a toy. Its 1300 stars in 1.5 days show strong demand for an independent engine stack beyond Unity; monetization is engine licensing plus support, or becoming a standard .NET game dev dependency. Risk: building an engine ecosystem is extremely hard, against Godot and MonoGame.
A free, open source C# framework for making games, or your own game engine. Direct access to native APIs and Vulkan, with the low-level libraries around them. The foundation under Kitten Space Agency …
A flux-3-image-klein Black Forest Labs officially released inference reference code for FLUX 3 Image Klein; its 54 stars in 0.7 days are driven directly by the BFL brand, a move from a leading image model vendor. It helps developers reproduce the latest model's inference, with a clear monetization path: weight licensing, API and cloud services. Risk: it's only a reference implementation with an incomplete toolchain, and it must compete head-on with closed-source image models.
FLUX 3 Image Klein: reference inference code
A mwoyeo Real-time on-screen translation on macOS: a lens overlays text in place with your language, running fully on-device without screenshot uploads. Useful for foreign UIs and games; the hard parts are layout reflow and latency, and it's closest to existing screen OCR translation overlays.
Live, in-place screen translation for macOS — a lens that rewrites the text under it in your language, on-device.
A AutoIllustrate A SakanaAI LLM agent tool that generates editable draw.io diagrams from a paper's method section; worth trying for researchers writing papers and system diagrams. The limitation is that output quality depends on the method description, and while draw.io editability is the selling point, layout still needs manual finishing.
Generate publication-ready, editable draw.io figures for research papers from a methods section, using LLM agents.
A mcd-mcvoice Builds the full voice-ordering flow for visually impaired users, with restrained scope and an honest description; accessibility is rare on this list. But it is a McDonald's programmer-day demo tightly coupled to a specific ordering pipeline, hard for outside developers to run, so its value is more a reference voice-interaction prototype than an install-and-use tool.
McVoice · Visually impaired users complete the entire McDonald's ordering flow by voice (choose store → order → price calculation → checkout) · Entry for the McDonald's Programmer's Day creative devel…
A mdastic A TypeScript Markdown/MDX parser claiming output identical to remark's mdast but 17-28x faster—same interface, new implementation, the easiest thing to verify and most practical. Static-site and editor builders can swap it in and test; risks are edge-case syntax and plugin ecosystem compatibility.
Markdown and MDX parser in TypeScript. Exact remark mdast output, 17 to 28x faster.
A cobblestone.ts A TypeScript implementation of the new chunked AEAD algorithm Cobblestone-128, usable as an npm package or single .js file for streaming encryption in browsers and Node. The author also wrote age, and the description is restrained with clear boundaries; the limitation is that the algorithm itself is young, so do not rush it into production.
A TypeScript implementation of Cobblestone-128 chunked encryption, available as an npm package or as a bundled .js file.
A open-scion Following Anthropic's "rethinking source control" idea, it implements centralized version control for agent-heavy monorepos in Rust; the problem is new, the scope is clear, and it explicitly states it is unaffiliated with Anthropic. Currently only a design skeleton, far from replacing git—first see whether the first real workflow can run.
Server-centric version control for agent-heavy monorepos, in Rust. An independent open-source implementation of the design from Anthropic's "Redesigning source control" post. Not affiliated with Anthr…
A FlashMask PaddlePaddle's mask-aware distributed attention implementation targets long-context training by handling attention masks and parallel sharding together; worth a look for infrastructure developers doing long-text pretraining. The limitation is tight coupling to the Paddle ecosystem, and the description gives no performance numbers or integration cost.
Efficient Mask-Aware Distributed Attention for Long-Context Training
A AniCraft Claims to be a 2D animation editor written from scratch in Rust, targeting the open-source gap beyond Flash and After Effects. The direction is rare and the description restrained, but the repo is only a day old with no release or screenshots, so maturity is unknown; first see if it runs.
AniCraft: An open-source 2D animation editor written from scratch in pure Rust.
A touri Puts coding agents like Codex, Claude and OpenCode into a desktop app, each running in an isolated VM with browser and computer control. The description is restrained and the boundaries clear, but similar sandbox products are clustering, it's only half a day old, and installation is unverified; first check whether the isolation is solid.
touri: open-source desktop app for AI agent bots (Codex, Claude, OpenCode) with isolated VMs, browser and computer use.
A knoedel-watch German vocabulary flashcards animated on an M5Stack ESP32-S3 watch, for hardware fans who want to study on their wrist. Small scope, clear goal, no fluff. Limits: you need that watch, and whether the firmware works out of the box needs verification; the alternative is Anki on a phone.
Animated vocabulary cards for learning languages, on an ESP32-S3 (M5Stack StopWatch). First course: German.
A net-switch Lets you assign each Windows app to a specific network cable or adapter, something multi-NIC and dual-broadband users have long hacked around with ForceBindIP; this turns it into a clickable little tool. Scope is restrained; worth a look at whether binding is stable.
Choose which internet connection each Windows app uses.
A bebox Natively emulates PowerPC and Hobbit BeBoxes on Haiku, implementing even obscure CPUs itself, a rarely touched retro computing subject for old BeOS users and emulator fans; the scope is restrained and description honest, limited by a very narrow audience.
Haiku native PowerPC and Hobbit BeBox Emulator
A scout An open-source browser agent that operates web pages like a human for scraping and workflow automation, directly hitting paid enterprise scenarios of repetitive web work. 646 stars in 1.6 days show the market caught it on release, and Outcrawl has a commercial entity for a hosted version—a typical open-source-to-SaaS funnel. Risk: earlier players like browser-use already have ecosystem and funding, anti-scraping is an ongoing cost, and turning scraping reliability into a paid moat is key.
Scout: the open-source browser agent, by Outcrawl. Mirrored automatically from Outcrawl's own repository.
A Show HN: Petal, open-source CoScreen/ScreenHero finally free Solves the problem that shared windows in pair programming can only become grid tiles: shared windows exist as real windows you can click into, type and scroll. After Screenhero was shut down by Slack and CoScreen by Datadog, a vacuum appeared; the original team rewrote it in eight weeks and promises never to shut it down, and 102 Show HN points plus founder backing is strong momentum. Hosted instances and self-hosting are the monetization path, but the no-charge pledge is the biggest uncertainty, with Tuple and VS Code Live Share as rivals.
Hi HN! I’m Jason Thomas, one of the founders of CoScreen. As some of you may know, Datadog sunset the CoScreen service on July 31st. Four of us from the original CoScreen team have spent the last coup…
A mosh-rs A Rust rewrite of mosh, wire-compatible with mosh-server 1.4, adding a native Windows client and scrollback—gaps mosh has left unfilled for years. People who SSH over weak networks benefit directly; risk is the rewrite's unknown maturity, so try it in parallel first.
Rust reimplementation of mosh (the mobile shell): wire-compatible with mosh-server 1.4, native Windows client, scrollback
A indra A fast CPU-only OCR pipeline—line detection, CTC recognition, reading order and Markdown output—hitting the pain of high GPU costs and fragmented pipelines in document digitization. 207 stars in 1.5 days show real demand, and the company behind it already has a commercialization base with Marker, so it could be widely reused as a standard dependency in document pipelines; risk is that strong open-source rivals like PaddleOCR and docling are more mature in accuracy and language coverage.
Fast CPU OCR: line detection, CTC recognition, reading order, markdown
A isocast An OBS-style recorder for macOS with scenes, per-source ISO recording, and streaming, all driven by CLI and MCP. macOS streamers who want recording wired into scripts or AI workflows are exactly the gap it fills; but beyond the description there are no sample recordings or config examples, so real stability is unknown.
OBS-style recorder for macOS: scenes, per-source ISO recording, streaming, and full control from a CLI and MCP.
A Show HN: T0: We trained a 256M SOTA open-weights time-series forecasting model A 256M open-weight time-series foundation model that tops benchmarks like GIFT-Eval with independent ERCOT validation, runs on CPU and Apple chips, and installs with a few pip lines. Forecasting is clear paid enterprise demand (retail, power prices, capacity planning) and the company sells uncertainty reduction, so the path is clear; risk is competition from TimesFM, Chronos and peers, and whether benchmark leadership converts to real customer retention.
Hi everyone, we are excited to share t0-gamma, the latest model in our t0 family. It is a 256M parameter open-weights foundation model for time series (TSFM). It runs on CPU, GPU or Apple chips. https…
A html-disasm A single-HTML-file offline disassembler covering ARM64, x86, WASM, CIL, Java bytecode, and ELF/PE/Mach-O/APK, with no file uploads. Ready to use for reverse engineers and CTF players, though its analysis depth falls far short of desktop tools like IDA.
Offline disassembler for ARM64, ARM32, x86-64, x86 (32-bit), WebAssembly, .NET (CIL), and Java bytecode. Supports Mach-O, ELF, PE, APK, and IPA files from macOS, iOS, Linux, Android, and Windows. Runs…
A scry A Windows whole-disk file search tool in Rust that returns filename results in milliseconds, can also search file contents, and is invoked with Alt+Shift+Space plus a tray icon. Useful for those annoyed that Everything only searches filenames; but the space is already filled by Everything and AnyTxt, and 14 stars in a day leaves index overhead and real speed unclear.
Whole-disk file search for Windows: any file by name in a millisecond, search inside your files, a search bar on Alt+Shift+Space and a tray icon.
A anan-iphone-use ESP32 firmware for the M5Stack watch that sends input to an iPhone over USB via MCP, letting an agent operate the phone directly. This hardware-plus-MCP combination is rare and worth a look for on-device automation; limits are being locked to a specific board and unverified stability and compliance boundaries.
ESP32 MCP firmware for iPhone USB input, built for M5Stack StopWatch with guided board adaptation
A css-composite-font Pure CSS composite font typography using different fonts and sizes for Japanese kana versus Latin. Front-end developers building Chinese or Japanese sites will use it immediately, and the approach is light with no dependencies; limits are reliance on unicode-range and font loading, with complex fallback cases still needing real testing.
Composite fonts in pure CSS: separate fonts and sizes for CJK, kana and Latin. 复合字体 / 合成フォント
A LeanCIL Uses Lean 4 to produce machine-checkable correctness proofs for compiled .NET CIL, covering both C# and F#, from Nethermind, with a rare direction and an honest description. A new entry point for formal verification and high-assurance software, with the barrier being that you must know Lean first.
LeanCIL is a Lean 4 formal verification framework for .NET programs. It enables machine-checked proofs of correctness and safety against compiled Common Intermediate Language (CIL), supporting applica…
A gongwen-kit A local tool that generates and machine-checks Word official documents per GB/T 9704, automatically checking format, numerals and style on output. It is a real need for government and state-owned enterprise staff writing official documents, with an accurate pain point; the limitation is heavy dependence on templates and rule coverage.
Official document generation and machine-check tool: generates and repairs Word official documents per GB/T 9704, and automatically checks format, numbers and writing style on output; runs locally, ai…
A sparkLimiter A small tool that persistently locks GPU clocks on NVIDIA DGX Spark and supports batch deployment across multiple machines. It addresses the specific power and temperature pain points of these machines with a restrained scope and clear command-line usage; the limitation is that it only matters to Spark users.
Cap the GPU clock on NVIDIA DGX Sparks across all your machines, persistently
A ScaleScout Local zoom for YOLO small-object detection that is explainable and budget-equal: comparing crop strategies under fixed compute, filtering edge artifacts, giving per-image evidence and an offline viewer. Researchers in remote sensing or detection can reuse it directly; the limit is its focus on YOLO and fixed budgets, with generalization unverified.
Explainable budgeted zoom for YOLO small-object detection. Equal-budget experiments, crop-edge filtering, per-image evidence, and an offline viewer.
A claude-bulkhead Targets the port conflicts, branch switching, and machine freezes that occur when running multiple Claude Code sessions at once, isolating a real pain point. Immediately installable for people running Agents in parallel, though it only covers the Claude Code toolchain.
Keeps parallel Claude Code sessions from killing each other's servers, fighting over ports, freezing the machine, or switching a branch under each other.
A OpenABL Adds a graphical boot menu and a standard Linux boot flow into the ABL bootloader on Qualcomm devices, aimed at handheld and custom firmware enthusiasts. It solves the concrete problem of awkward flashing entry on such devices, but the risk is modifying the boot layer directly, where a bad flash is costly.
A graphical boot menu and standard Linux boot for Qualcomm handhelds, built into Qualcomm's Android bootloader (ABL).
A hyundai-pleos-europe-ha Brings Hyundai, Kia and Genesis European vehicles into Home Assistant via the Pleos Vehicle Data API, explicitly read-only. It fills a real gap for HA owners and the description is restrained; the limit is reliance on official API authorization, with different coverage from the existing hyundai_kia_connect_api.
Read-only Home Assistant integration for Hyundai, Kia and Genesis cars in Europe via the Pleos Vehicle Data API
A quickRS-player A practice player for your own Rocksmith 2014 tracks: scrolling tab, looping, slowdown, scoring, real-tone cable support, and psarc to feedpak conversion. It is a solid supplementary tool for Rocksmith players practicing; limits are Windows-only and dependence on the game's native track format.
A free practice player for your own Rocksmith 2014 songs: scrolling tab, loops, slow-down, scoring, Real Tone Cable support, and a psarc-to-feedpak converter (Windows)
A immich-video-optimizer A terminal tool that shrinks videos in a self-hosted Immich library, transcoding to HEVC with hardware encoding while preserving metadata. The pain point is concrete and scope restrained; Immich users short on storage can use it today, with limits of quality loss and dependence on ffmpeg and GPU encoding.
Interactive terminal client to reclaim gigabytes on your Immich server by re-encoding videos to hardware-accelerated HEVC while preserving 100% of metadata
A SimdYolo A pure C#/.NET YOLO inference library with SIMD CPU and Vulkan/Metal GPU backends, reading .pt and ONNX directly and covering detection, segmentation and pose. It meets a real need for .NET developers who don't want to drag in Python; next is whether it has credible benchmark data.
Pure C#/.NET YOLO inference: SIMD CPU, Vulkan/Metal GPU, direct .pt and ONNX; detection, segmentation and pose.
A Show HN: SolidCraft – Open-Source Parametric CAD in Rust (With MCP Support) SolidCraft is an open-source parametric CAD tool in Rust with MCP support, and 133 Show HN points show high community expectations for AI-drivable CAD. CAD has long been dominated by SolidWorks and Fusion 360, so an open-source alternative is itself a chance to become a standard dependency, and MCP lets it plug into agent workflows. The risk is that parametric CAD kernels are extremely hard to build deeply, and mature competitors are far ahead in features.
A Show HN: Agor, multiplayer AI – stop running agents alone Agor focuses on multiplayer AI, putting team members and agents in the same collaboration space, and 1,446 GitHub stars show multi-agent human-AI collaboration is a real current need. Commercialization could be team collaboration SaaS or an enterprise edition. The risk is that orchestration frameworks like LangGraph and CrewAI are all moving toward collaboration, and only 9 HN points suggests broad awareness hasn't opened up yet.
A inkline-town An isometric line-art town simulation running on a self-built bevy-ts ECS plus Pixi. People wanting to learn TypeScript ECS or Pixi rendering can read the source directly, and the author is a long-time technical writer with good judgment. The limitation is that it is more a tech demo than a usable tool.
A town simulation in pale isometric line art, on the bevy-ts ECS and Pixi
A exploded A Claude Code skill plus runtime that splits a live UI component into a drawing layer, state and tokens, claiming no re-rendering and no hand-filled values. A rare angle for design-system people; whether "no re-rendering" holds needs your own verification.
Take a real UI component apart. A Claude Code skill + runtime that explodes a live component into paint layers, states and tokens. Nothing is redrawn, no number is typed.
A csp-forge Scans build artifacts or live URLs to compute sha256 for inline code and generate CSP, outputting Nginx, Vercel, Helmet and other formats, with --check in CI. The need is real and the scope clear, though most existing tools are evaluators rather than generators.
Scan a built static site (dist/, build/, out/) or live URL and generate a tight Content-Security-Policy with sha256 hashes for inline code. Outputs header, meta, Nginx, Apache, Netlify, Vercel, Helmet…
A kalman-percentile-bands A TradingView Pine v6 indicator combining a Kalman mean with three empirical quantile bands on each side for mean reversion, ready to paste and use, with restrained and honest documentation. The limitation is that volatility-band indicators are already plentiful, and the quantile methodology and backtests aren't provided, so its value depends on whether you trust its parameters.
A mean-reversion indicator for TradingView (Pine Script v6). It plots a volatility-adaptive Kalman mean with three empirical percentile bands on each side.
A vibereceipt Prints a receipt for each AI coding session: token counts, cost at API prices, and how many times the agent said you're right. Fully local, supports Claude Code and Codex, and cost awareness is the point behind the joke; ccusage already does finer stats.
Print a store receipt for your AI coding: tokens, what they'd cost at API prices, and how many times the agent said "You're absolutely right". Claude Code, Codex, OpenCode. 100% local.
A mcu-timer-calc An 8051 timer initial-value calculator as a CLI with zero third-party dependencies and full unit tests, also packaged as an Agent Skill; small but clean, ready to use for embedded developers, with the only downside being its narrow function.
8051 timer initial value calculator: a command-line tool with zero third-party dependencies and fully unit-testable, also an Agent Skill.
A onionKidsOS Wraps OnionOS on the Miyoo Mini Plus in a kid-sized launcher so children only see games they can play. The idea is clean and the scope restrained, but the audience is only parents with this handheld, and 11 stars means it is unproven.
A game console for small children: OnionOS for the Miyoo Mini Plus with a child-sized launcher
A net-merge Aggregates Wi-Fi, Ethernet, and phone-tethered networks into one link on Windows with per-connection load balancing, built on TUN/Wintun — one of the few multi-NIC solutions that actually works. 49 stars in a day shows real demand; the limits are Windows-only coverage and no effect on a single TCP connection.
Use all your network adapters at once on Windows: per-connection load balancing over Wi-Fi, Ethernet and phone/USB tethers (TUN + Wintun)
A rsi_eval A lightweight SDK/CLI for unified evaluation of recursive self-improvement methods, running on existing agent benchmarks; the scope is restrained and the interface is clear, worth a look for agent training researchers, though the ecosystem is still early.
A lightweight Python SDK and CLI for evaluating recursive self-improvement (RSI) methods across agent benchmarks.
A chinese-web-typography An Agent skill that packages Chinese web typography experience, with interactive before-and-after examples. Frontend developers regularly handling CJK-Latin mixed text, punctuation squeezing, and line height can use it directly, and it's more concrete than abstract typography articles. The limitation is that the content is still small and less deep than mature CJK typography standards.
A skill for thoughtful Chinese web typography, with interactive before-and-after demos.
A keel A code-based CAD language on NVIDIA's usd-brep kernel: 1D paths, 2D sketches, 3D parts, outputting exact USD. Worth a look for those tired of OpenSCAD mesh precision who still want scripted modeling. Limitation: 11 stars, very early, ecosystem far behind CadQuery.
A CAD language on NVIDIA's usd-brep kernel: 1D paths, 2D sketches, 3D parts, exact USD output
A kitchen-frame A family dashboard for cheap Android photo frames: calendar, weather, Todoist, recipes, and agent messages, running on a Cloudflare Worker, with the author noting inspiration from Wes Bos. Honest and restrained; usable if you have a frame.
A calm family dashboard for a cheap Android photo frame: calendar, weather, Todoist, recipes, music and agent messages on a Cloudflare Worker. Inspired by Wes Bos's photo-frame-dashboard.
A yoto-audiobookshelf-bridge Pushes audiobooks from a self-hosted Audiobookshelf to the Yoto kids' player with two-way progress sync — a specific, tightly scoped problem solvable with one Python service. A must-have for families using both, but the audience is very narrow and it relies on Yoto's undocumented API.
Play your Audiobookshelf audiobooks on Yoto players, with listening progress kept in sync
A transnap-oss A macOS menu bar word-lookup translator built on Apple's own Translation framework, local-first and requiring no account. For Mac users who care about privacy and don't want DeepL or Bob, it's a clean little tool; the limits are reliance on a new OS, macOS-only coverage and a narrow feature set.
A free, local-first macOS menu bar translator built with Apple Translation and SwiftUI.
A new-term-radar Catches new models and terms on X, HuggingFace, GitHub, and sitemaps before keyword tools do, serving SEO and brand monitoring's timeliness needs; 532 stars in two days shows the pain is real. Could become a keyword-intelligence subscription like Ahrefs' new-term monitoring, but risks are source bans, high scraping costs, and users replacing it with their own scripts.
Daily radar for new AI terms: names of new models, products and repos, caught on X, Hugging Face, GitHub and sitemaps before keyword tools list them.
A Show HN: PS2 games statically recompiled to WebAssembly Statically recompiles PS2 games to WASM to run natively in the browser, technically harder than a normal emulator; 260 Show HN points is the strongest momentum in this batch, from nostalgic players and the reverse-engineering community. It could become a PS2-to-WASM porting pipeline service for porting studios, but Sony copyright and ROM compliance are hard risks, and a demo alone is unlikely to become a standard dependency.
A d2r-mobile Ports Diablo II: Resurrected for Switch to iPhone and iPad, solving the pain of needing your own image files to play AAA retro titles on mobile; 687 stars in two days came from nostalgic players and the porting community spreading it organically. It could become a general Switch-to-iOS porting layer for small teams to reuse, but Nintendo DMCA lawsuits and single-game dependence are the biggest risks, making it hard to become a standard library.
Open-source iOS / macOS compatibility layer for Diablo II: Resurrected with the Reign of the Warlock expansion. Runs your own Switch copy natively on iPhone, iPad and Mac — full touch controls, 60 fps…
A Show HN: Atticus – Hold-to-talk voice assistant for macOS that runs on Claude -p Press a hotkey to talk; simple tasks finish locally via rules in 0.08 seconds, complex tasks go through Claude -p without an API key, and successful actions are saved as replayable scripts to cut costs; 55 Show HN points and a young author's story boosted spread. Commercialization could be subscription or enterprise, but it faces Raycast AI, Siri, and the macOS 26 barrier, and a personal project has a thin moat.
Hi HN! I'm Saarth. I'm in 9th grade, and I built Atticus. Atticus is a fast and cheap Mac assistant that can do tasks for you. You hold down Control-Option-Space, speak or type your task, be it comple…
A pulse-proxy-checker A local proxy batch checker written in Tauri that tests HTTP/SOCKS proxies and gives country, quality and risk scores, with offline GeoIP avoiding external queries. It's handier than online checkers for people maintaining proxy pools. Similar scripts are plentiful, and long-term value depends on whether the risk scoring is actually accurate.
⚡ Fast local proxy checker for HTTP/HTTPS/SOCKS4/SOCKS5 — live results with country, quality & risk, offline GeoIP, saved pools. Free & open-source (Tauri + Rust). macOS & Windows.
A vtype-linux Runs the official WeChat input method engine inside Fcitx5, with cloud candidates, hot words, and a lexicon shared with phone/Mac, addressing Linux Chinese input's long lack of good options. The description is concrete and includes a one-click agent install script; the risk is compatibility tied to the official engine.
Use WeChat Input Method on Linux: run the official engine in Fcitx5, with cloud candidates, hotwords, personal dictionaries and phrases synced with phone/Mac. Lets agents like Codex install it in one …
A solid-glass Offers a complete design language emphasizing material and physical feel for AI apps, filling the gap left by homogeneous AI product interfaces; 229 stars in two days show strong demand for ready-made AI UI systems. If adopted by a shadcn-style ecosystem it could become a standard component library with paid themes, but the risk is competition from Tailwind, shadcn/ui and various AI templates.
A design system for AI applications. Material, physical, tactile, technical.
A win4wine Brings Windows XP's Explorer to Wine as a Linux desktop shell; almost nobody would use it daily, but the idea is rare, the implementation concrete and the description restrained. Suited to people curious how far Wine can be pushed, with practical value near zero.
Run Windows XP's Explorer as the Linux desktop shell on Wine
A solidcore Ports a subset of NVIDIA SMLib's B-rep kernel to safe, deterministic Rust with wasm support, with a restrained goal and precise description. CAD kernels have long been held by C++, so this path is worth following, though it is only a subset with no examples yet.
A Rust port of NVIDIA's SMLib B-rep kernel (CAD subset): safe, deterministic, wasm-compatible
A uma A native macOS menu bar client for self-hosted Umami analytics by egoist, with a tiny scope and honest naming; install it and you can see your data, but expect nothing beyond a dashboard.
A native menubar app for Umami.is
A livelight.koplugin A KOReader plugin that adjusts brightness and color temperature in real time by swiping at the screen edge, restoring a feel previously only on stock Kobo. Tiny scope and accurate description; e-reader users can install it in ten minutes.
KOReader plugin: adjust brightness and warmth in realtime by dragging your finger on the screen edges, just like in Kobo’s own software.
A mxoptions Adds official button, gesture, Actions Ring and haptics settings to the Logitech MX Master on Linux, plus a D-Bus interface. It fills a real gap for Linux office users, and the author openly says Claude wrote it. The limit is that it only covers the MX Master series.
Settings app for Logitech MX Master mice on Linux: buttons, gestures, Actions Ring, wheels, haptics, D-Bus API. Written by Claude.
A ComfyUI_NVIDIA_RTX_VSR_Long_Video Turns NVIDIA RTX video super-resolution into a ComfyUI node, with disk paging and low-memory handling for long videos, preserving audio tracks and encoding via NVENC. A ready-made piece for batch-restoring old footage, if you have an RTX GPU and ComfyUI set up.
NVIDIA RTX Video Super Resolution for ComfyUI, optimized for long videos with disk-backed processing and low RAM usage. Supports AI upscaling, denoising, deblurring, H.264/HEVC NVENC encoding, audio p…
A EnglishPractice An offline English practice app for Turkish speakers that runs Ollama, Whisper, and Kokoro locally, with A1–C1 courses and spoken correction. Full scope and fully local operation are the draw; the limits are serving only one native language and a nontrivial install.
Yerel, çevrimdışı ve taşınabilir İngilizce pratik uygulaması (Türkçe konuşanlar için): yapay zekâyla sesli/yazılı sohbet ve anında düzeltme, A1–C1 dersler ve sınavlar, Türkçe sesli öğretmen, kelime ve…
A minecraft-metal-shaders Brings shadows, volumetric clouds, water reflections and HDR to Minecraft Java on Apple Silicon via native Metal, forked from mcopt with honest attribution. Shaders have long been missing on Mac, so it is worth a look; the limit is keeping up with upstream and game versions.
Native Metal shaders for Minecraft Java on Apple Silicon Macs: shadows, volumetric clouds, god rays, reflective water, HDR. A fork of mcopt.
A qz-browser-agent Alibaba's local-first Chrome agent uses its own models for browser automation and data extraction, and the engineering foundation is worth a look. But browser-use and nanobrowser already crowd this space, leaving only local plus bundled models as differentiation.
Local-first AI agent for Chrome: browser automation, web research, and data extraction with your own model.
A folivox Splits ebooks into chapters and uses CosyVoice 3 to clone your voice locally into an audiobook, with online engines as an alternative. Fully local and free is the selling point; limits are that voice cloning needs samples and compute, and quality depends on text cleanup.
Turn any e-book into a chaptered audiobook in your own voice: free, local, CosyVoice 3, with optional online engines
A miscale-dashboard Turns Xiaomi Body Composition Scale 2 BLE data into a self-hosted real-time dashboard with multi-user, body composition and history, letting owners of this scale escape the cloud app entirely. openScale and ESPHome already offer similar solutions; it wins on out-of-the-box use but has a narrow audience.
Self-hosted live dashboard for the Xiaomi Mi Body Composition Scale 2 over Bluetooth Low Energy: profiles, body composition, history
A msfs2024-glass-cockpit-dlss-fix A ReShade plugin that removes DLSS and FSR ghosting on glass cockpit displays like the PFD and ND in MSFS 2024, covering both monitors and VR. The problem and fix are both concrete, so sim pilots can try it directly; the risk is dependence on the game version.
ReShade add-on that removes DLSS and FSR ghosting from glass cockpit displays (PFD, ND, ECAM) in Microsoft Flight Simulator 2024, on a monitor and in VR
A schematic-manager A cross-platform desktop tool built with Tauri for Minecraft builders to manage and preview .litematic projection files, addressing a concrete need previously served only by scattered scripts. With 23 stars in half a day, feature depth remains to be seen.
Cross-platform Minecraft projection (.litematic) file management and preview tool | Tauri + Vue + TS
A awesome-iran-mcp Wraps Iranian local services like Digikala and Snappfood into 19 MCP servers and 308 tools so agents can check local prices and orders. Regional MCP is a real gap, but the "awesome"-style naming and huge tool count raise doubts about maintenance depth.
19 open-source MCP servers (308 tools) that give AI agents the Iranian web: Digikala, Snappfood, Jabama, MrBilit, Paziresh24, Taaghche and more. Prices in Toman.
A Show HN: BetterWispr – Free, open-source dictation for Mac A free, open-source Mac dictation tool aimed directly at paid products like Wispr Flow, with 140 points in half a day proving voice input is a real need with many underserved users. Open source speeds adoption, then team editions or local models could charge; the risk is Apple's system-level dictation improvements and the incumbent's brand.
A SimRun Runs iOS simulator apps directly on the Mac, removing the friction of repeated Xcode installs; 116 stars in 1.6 days shows iOS developers can use it immediately. It could become a widely cited dev utility or even be absorbed officially, but the risk is reliance on private APIs that break with Xcode updates.
Run iOS Simulator apps directly on your Mac.
A glovebox Builds on Mac a compatibility layer to run Half-Life: Alyx on Apple Vision Pro, requiring you to supply the game. Far harder than wrapper streaming, and 34 stars in a day shows people are waiting; limits are single-game support, high setup门槛, and feel needs real testing.
Play your own copy of Half-Life: Alyx on Apple Vision Pro, built on your Mac
A sieve Reads existing package inventories from hosts and containers, checks CVEs via OSV.dev, then ranks by CISA KEV and EPSS, with no scanner install or vulnerability DB sync. Useful for ops wanting a "fix this first" list; coverage depends on OSV's catalog, and it competes with Trivy and Grype.
Find the CVEs that matter on a Linux box and its running containers. Reads the package lists already there, asks OSV.dev, ranks by CISA KEV and EPSS. No scanner binary, no database, no server.
A nixos-vex Combines OpenVEX statements with NixOS modules to judge which CVEs actually hit your real system config instead of accepting them wholesale. A sharply targeted piece for NixOS machine maintainers; narrow and NixOS-only, but the description is restrained and the boundaries are clear.
OpenVEX feed and NixOS module to manage NixOS CVEs, evaluated against your system's configuration.
A kadi Places a local decider before calling a frontier model, answering routine decisions locally and escalating to the large model only beyond a confidence interval, with conformal prediction bounding disagreement. The idea is more concrete than typical routing wrappers and the zero-dependency Python is easy to try, but integration and measured benefits are not yet evidenced.
⚖️ Kadı — a calibrated judge in front of your frontier LLM. Routine decisions answered locally with a conformal-prediction guarantee (≤ α disagreement with the oracle); everything else escalated and l…
A getseed-ble-protocol Fully reverse-engineers the BLE protocol of the Dedao AI recording card, with a command table, firmware parsing, and a pairing-free direct-connect tool; anyone holding this device can use it today, a narrow audience but the only option for them.
Dedao AI recording card (GetSeed/GSC10) BLE protocol reverse engineering: complete command table, firmware parsing, PC direct-connection tool (plaintext Ogg / no pairing / no key)
A pi-durable-assistant A self-hostable persistent personal agent that can resume tasks after a process restart, with sub-agents, scheduling, long-term memory, and a WeChat channel. Persistent execution is indeed an underbuilt part of today's agents, but covering scheduling, memory, and multiple channels at once with installation and runtime details still unclear risks stopping at the demo level.
A self-hostable persistent personal Agent: restores tasks after process restart, supports sub-Agents, scheduled plans, long-term memory, a web interface, and a WeChat channel
A shrike A debugger for people building web apps with AI: click an error and see the full chain down to the specific line, then let the Agent fix it after confirmation; the approach fits how development works now, but as an early version it still needs validation against your stack.
The open source debugger for people who build web apps with AI. Click the thing that breaks; see the whole trip to the exact line; your agent fixes it after you say OK.
A AquaType A macOS Zhuyin input method that still suggests the right word after a mistyped key, a real pain point for Taiwanese users that Xiaomai Zhuyin doesn't address. 19 stars in a day shows demand; limits are macOS-only, Zhuyin-only, and fuzzy matching feel needs hands-on testing.
A Zhuyin input method for Mac that suggests the right words even when you press the wrong keys.
A ha-thread-pulse Draws the mesh relationships of Thread/Matter devices as a live topology map in Home Assistant, solving the concrete pain of not being able to see network structure. Very useful for HA users with Matter devices deployed; narrow scope and tightly bound to the HA ecosystem.
Live map of your Thread network for Home Assistant – watch your Matter devices talk
A wg-ez Packages WireGuard management into a single static binary using about 1MB of memory, clearly aimed at self-hosted VPN users who want to save resources. The scope is restrained and the description honest, lighter than container approaches like wg-easy, but the README is still thin and the config path and permission model are not fully explained, so it is not yet worth a deep read.
A modern WireGuard manager packaged into a single static Rust binary, using 1MB of RAM.
A finder-menu Adds new file, open terminal here, copy path, and go to parent folder to the macOS Finder context menu—things the system has always lacked. The scope is restrained and it is native Swift, usable right after install; similar small tools and paid apps exist, but this wins on being free and open source.
Right-click menu for macOS Finder: New File, Open Terminal Here, Copy Path, Go to Parent Folder
A talorys Deploys a personal agent with memory, tasks, notes, and scheduled reminders to the user's own free Cloudflare quota with one command, emphasizing no telemetry and single-user privacy; 355 stars in 0.6 days rides the personal agent wave. It could serve as a hosted template or entry point to a private agent platform, but free-friendly single-user positioning compresses paid space; strong competitors are big-vendor hosted assistants and open-source self-hosted agent frameworks.
Your personal AI agent in your own Cloudflare account. Chat, memory, tasks, notes and scheduled reminders, deployed with one command: npx create-talorys@latest. Free-tier friendly, single-user, no tel…
A muguet Ports GTA V to iOS and macOS as a native Mach-O build with multiplayer, a rare engineering feat; 291 stars in 0.6 days comes almost entirely from the impossible-port narrative, proving its porting toolchain has reuse value. But commercialization is completely blocked by Rockstar/Take-Two copyright, leaving only open-source results to trade for influence and recruiting leverage; the biggest risks are legal takedowns and long-term maintenance difficulty.
A native Mach-O port for iOS/macOS of GTA V, with multiplayer
A countinghouse A local-first personal finance tool for UK users: import statements, output git-versioned JSON, and compute assets, spending and tax-year projections. The scope is clear and data stays local, more honest than cloud bookkeeping apps; the limits are that it only makes sense for UK tax rules and the UI is still very early.
Counting House: a private, local-first UK personal finance tracker. Statements in, git-versioned JSON out: estate value, spending, tax year and projections.
A StrimKeep A small tool for NAS/Emby users managing dual STRM libraries: quality selection, missing-episode comparison, and follow-up subscriptions, with concrete pain points and ready to run. MoviePilot and nas-tools already exist, but this focuses only on STRM scenarios; however, it takes on exploration and notifications too, so its boundaries are scattered and long-term maintenance is uncertain.
A NAS-oriented STRM dual-library movie and TV management and governance tool, supporting quality-based selection, missing-episode comparison, movie/TV discovery, follow-and-subscribe updates, and libr…
A FinderSearch Tackles the old macOS problem of slow, weak Spotlight/Finder fuzzy search; native Swift with an fsearch core can immediately speed things up for heavy file users, and 553 stars in 1.7 days reflects real demand in the Mac tools community. A free-plus-Pro desktop tool could monetize, but it runs head-on into Raycast, Alfred and system Spotlight, capping paid upside for an independent search tool.
A native macOS file browser with fast fuzzy filename search, powered by fsearch.
A ai-promo-video-maker A Remotion pipeline that breaks making a product promo video into research, 4K page capture, beat-synced editing, TTS narration, and line-by-line verification, runnable as an Agent Skill. The workflow is concrete and suits indie developers making launch videos, but the description is marketing-heavy and final quality depends on footage—try it yourself.
Have AI make a product promo video for you in one sentence! Tell a product name or URL to AI coding tools like Cursor, Claude Code, Codex, and it researches the product itself, draws sample images in …
A gggodlin-cc-harness A governance layer for a personal Claude Code workflow: gates, hooks, reviews, and spec-to-ticket conversion, which the author marks as reference-only and not ready to use out of the box. Valuable for people building agent pipelines, but you'll have to dig through the configuration yourself.
Governance layer of a personal Claude Code harness: gates, hooks, review and spec-to-ticket workflow. Reference, not a turnkey install.
A ConcordRTS An RTS engine written in C++20 that's compatible with RA2/YR rules and map formats and explicitly ships no official assets. Attractive to Red Alert modders and engine developers, but it's still early and OpenRA is the unavoidable comparison.
C++20 RTS engine source code, supporting RA2/YR rules and map formats, does not include official game assets.
A macos-wechat-export It exports WeChat 4.x local chat history on Apple Silicon: copy and re-sign to get the key, then open read-only with SQLCipher. A real need for people reclaiming their own data, and the approach is clearly explained; the limitation is being locked to specific versions and platforms.
macOS WeChat 4.x local chat history export: temporarily re-sign a copy to extract keys + SQLCipher read-only (Apple Silicon)
A workhold-durable-work-queue A Python persistent task queue with named queues, fenced leases, and atomic follow-up tasks. The semantics are clearly explained, suiting backends that want at-least-once delivery without duplicate execution; peers include procrastinate and pgqueuer, so it needs to prove its difference.
Durable per-application work queue with named queues, fenced leases, and atomic follow-up tasks.
A WindWakerHD-Android-Plus It statically recompiles The Wind Waker HD to Android with Vulkan rendering, a built-in Portuguese translation, and HUD editing, requiring your own game. A real piece of work for the recompilation and handheld porting scene, but it's a single-game branch with a high barrier to entry.
The Legend of Zelda: The Wind Waker HD natively on Android (static recompilation, Vulkan). Built-in Brazilian Portuguese translation, HUD editor, .wua support, controller gyro, and more. Bring your ow…
A Show HN: Let your AI agents paint big arrows, boxes and text on your screen It gives CLI agents like Claude Code and Codex a visual feedback channel to draw arrows, boxes, and text on the Mac screen, solving the real pain that agents can't point at things for people; 364 points plus 444 stars is the strongest dual-source signal in this batch. It could distribute through the agent-skill ecosystem or offer a paid team collaboration layer, with the risk of being swallowed by system-level screenshot and annotation features.
A big-arrow-on-the-screen Gives CLI agents visual feedback on the Mac screen by drawing arrows, boxes, and text, with a single command, click-through, and auto-dismiss, solving the communication gap where agents can only output text; 444 stars in two days plus 364 HN points is a strong dual-source signal. It could join the agent-skill ecosystem or offer a paid team collaboration version, with the risk of being subsumed by system-level annotation features.
Let your AI agents paint big arrows, boxes and text on your Mac screen. One CLI, click-through, gone by itself. Skill for Claude Code and Codex. MIT.
A AssemblyForNeovim Shows inline x64 and aarch64 assembly for C/C++/Rust inside Neovim, aimed at low-level developers who read compiler output daily. The scope is clear in one sentence with no hype, and 38 stars in half a day show people were waiting for it. Limits: only two instruction sets, and it must be wired into the build flow; closest to Compiler Explorer but wins by staying in the editor.
Small lua script that shows inline assembly in neovim (x64, aarch64) for C, C++, and Rust.
A open-appleremote It unlocks and parses touch and buttons on the original Siri Remote on Linux, with tvOS gestures and a virtual keyboard, in pure standard-library Python. HTPC remote users benefit directly; it only covers two hardware generations and the project is still very early.
Use the first-gen Apple Siri Remote (A1513/A1962) on Linux: unlock it, decode touch and buttons, tvOS-style gestures, a virtual keyboard. Pure-stdlib Python.
A once A single-file Go CLI that runs a command once and caches its output for a while to reuse, installable and ready to go. It solves a small but real problem (repeated slow commands) with clear boundaries; a shell alias could do the same, but this wins on being ready out of the box.
Run a command once, reuse its output for a while.
A work-humanizer Removes the AI tone from work emails without changing the meaning, packaged as an Agent Skill with blind evaluation and examples of meaning changes checked by humans, which is more honest than similar de-AI-ifying tools. Suited to people who write English emails daily; the space is crowded and results still depend on the model used.
A humanizer for work writing: takes the AI tone out without changing what your text says or commits to. Agent Skill + evidence: blind judges, hand-checked meaning changes, Pangram.
A z7-qemu Turns the Nikon Z7 into a full-machine model runnable in QEMU, with working RAW/ISP/LCD, requiring your own firmware. It's rare infrastructure for camera firmware reverse engineering and ISP research; the audience is very narrow and it's still experimental, but the description is restrained and the boundaries are clear.
Experimental Nikon Z7 machine for QEMU, functional RAW/ISP/LCD and research tools. Bring your own firmware.
A filter-ast It parses $op(value) filter objects into a database-agnostic typed AST, zero-dependency and about 50ns per parse. Suited to backend developers building their own filter DSL; scope is restrained and metrics concrete, but it only covers parsing and long-term maintenance is uncertain.
Parses a declarative filter object with $op(value) strings into a typed, database-agnostic filter AST. Zero dependencies, ~50 ns per call.
A lipwise Local voice dictation across three platforms: hold a hotkey and speak, recognition runs on-device, then AI handles spoken corrections like 'delete that last sentence'. A clear choice for people who care about privacy and don't want cloud dictation subscriptions; the repo is new, with no data yet on model distribution or latency.
Private voice dictation for Windows, macOS and Linux. Hold a shortcut, talk, and Lipwise types what you meant into any app: speech recognition runs on your computer, and an AI editor understands "scra…
A MacRAR Wraps RARLAB's rar/unrar and 7zz in a native AppKit interface, supporting 30+ formats, encryption, split archives, and Finder shortcuts. macOS really lacks a decent RAR GUI, so it is worth trying; depending on external binaries, long-term maintenance is the key risk.
WinRAR-style archive manager for macOS: native AppKit app around RARLAB rar/unrar and 7-Zip 7zz. RAR, 7z, ZIP, TAR and 30+ formats, encrypted and multi-volume archives, Finder Quick Actions.
A VibeOffice A local-first office suite running in the browser with an Office 2003-style interface, where files are not uploaded and no account is needed — an appealing trade-off. But docs, spreadsheets, and demos are all lumped together, and 91 stars in two days looks more like a demo than a usable product, so depth is doubtful.
A free office suite that runs in your browser, in the style of Office 2003. Nothing to install, no account, no server: documents are opened, edited and saved on your own computer.
A intelligent-ui-demo Dissects how ChatGPT's generative UI works, giving a runnable recreation with a streaming DSL compiler plus iframe/Worker sandbox, compared against AG-UI and A2UI, for frontend devs building LLM-driven interfaces. Analysis plus runnable code beats talk; the limit is demo level with no sign of long-term maintenance.
Deconstructs the implementation of ChatGPT Intelligent UI and compares it with AG-UI and A2UI; includes a runnable replica (streaming DSL compiler + iframe/Worker sandbox)
A Ricoh-GRMod An unofficial modding tool for the Ricoh GR IV that imports xmp/cube presets into Image Control and adds custom aspect ratios and shutdown screens, aimed at photographers who want a personal color profile on their GR. The niche is tiny but clearly explained, and 26 stars fits a genuine small demand; limits are that it covers one model only and relies on undocumented interfaces.
Unofficial modding tool for the Ricoh GR IV Series: import .xmp / .cube presets into Image Control, add custom cut aspect ratios, custom power-off images.
A relight-anime An After Effects plugin that uses depth estimation to light AI-generated animation assets, porting a TypeGPU lighting model to C++/CUDA, a concrete and technically substantive path. Aimed at animation creators, it requires After Effects and has only been tuned on anime assets.
Relight Anime: After Effects plugin that adds light and shadow to AI anime footage using estimated depth (TypeGPU lighting model ported to C++/CUDA)
A ARTEX An autonomous AI penetration testing system and Baidu agent+ attack-defense competition champion, with 696 stars in 0.8 days, the strongest momentum in this batch. It chains reconnaissance, exploitation, and reporting into a closed loop, directly targeting pentest-as-a-service and enterprise red-team subscriptions with the clearest commercial path. Risks are liability for accidental damage and pressure from big-vendor security platform bundling.
AI autonomous penetration testing system | Champion project of Baidu's "agent+" attack-defense challenge
A Show HN: K10s – A Clickable Kubernetes TUI (Go, Bubble Tea) Upgrades read-only TUIs like k9s into a clickable single binary with logs, exec, and port-forwarding, saving repeated kubectl context switches. 238 stars in 0.2 days and 61 HN points is the strongest early momentum in this batch. It could commercialize as a team console, with the risk being Lens/Rancher and k9s ecosystem inertia.
A answerui A generative UI layer that renders model replies as charts, forms and calculators, compatible with any OpenAI-compatible model, suited to indie developers building interactive answers. Lighter than frameworks like CopilotKit, but built in one day, so component reliability and fallbacks for unstable model output need real testing.
Open-source alternative to ChatGPT's Intelligent UI. Answers arrive as live charts, forms and calculators. Works with any OpenAI-compatible model, in the cloud or local.
A WinShun A Windows quick launcher summoned by double-tapping Ctrl, supporting filename, pinyin and full-text search in Qt6 + C++20, with clearly stated scope. Everything and Listary are rivals; its edge is Chinese pinyin and content search, and large-index responsiveness needs real testing.
A fast search launcher for Windows: double-tap Ctrl to summon, search by filename, pinyin and text content. Qt 6 + C++20.
A agentic-kb A knowledge-base agent for regulatory documents that chains dense and sparse retrieval, RRF, cross-encoder reranking, GraphRAG and audited ReAct, plus retrieval evaluation tooling. Teams building enterprise compliance Q&A can borrow from it, though generality depends on the corpus.
Regulation/procedure knowledge base agent: hybrid retrieval (bge-m3 dense + sparse → server-side RRF → cross-encoder reranking) + GraphRAG + a ReAct agent with three gates and auditing. Includes retri…
A cli Excalidraw's official CLI that script-generates and processes .excalidraw diagrams, also serving agent workflows. Official and restrained in scope; the only issue is it's one day old and interfaces will still change.
Official Excalidraw CLI for humans and agents
A PiDurableKit It brings durable execution to Apple platforms, running agent logic on a JS engine with Swift-side APIs—same lineage as Cloudflare Workflows but on-device. The scope is clear and momentum fast; worth seeing how it handles checkpointing and recovery.
Durable AI agents for Apple platforms: pi-durable on JavaScriptCore with a Swift API
A betterwispr Local dictation for macOS: hold a hotkey, speak, and it types, supporting Whisper, Parakeet and Apple speech, explicitly targeting Wispr Flow and fully offline; the need is clear and onboarding straightforward, so it comes down to whether recognition quality and latency can match the commercial version.
Free, open-source, local-first dictation app for macOS. Hold a shortcut, speak, and text is typed into any app. On-device speech-to-text with Whisper, Parakeet and Apple speech. Private, offline, open…
A map-motion A Claude Code skill that turns real Amap routes into map-animation videos, with 12 styles including dive, altitude and flight path, and a clear positioning. Travel or cycling vloggers can produce clips fast; the limit is being locked to Amap data and the Claude Code ecosystem.
Map motion-graphics video Claude Code skill: based on real Amap routes/boundaries/satellite basemaps, creates map animations such as Earth dive, vlog "I am here", hiking elevation, cycling journey, fl…
A PaperBenchX A benchmark evaluating whether research agents can reproduce papers end to end, covering 12 research fields, hitting the gap in verification for AI-driven science. Worth a look for those doing agent evaluation or research automation; the limits are that question sourcing and scoring criteria need close reading.
A benchmark for scientific research agents, evaluating end-to-end paper reproduction across 12 research areas.
A freecenter It adds a web admin console for free ESXi, which lost vCenter after Broadcom, with migration, backup and a Terraform provider. The pain point is real, but claiming so much in just over a day—check the code before trusting it.
A vCenter-style management console for the free VMware ESXi hypervisor: web UI, REST API, migration between hosts, backups, Terraform provider
A streamgres Juspay's Postgres subscription-based sync engine, built on logical replication for cross-database sync. The company backing and 66 stars suggest real demand, but how it beats PeerDB-style options requires reading the code.
A subscription-native sync engine for PostgreSQL.
A tailscale-nx It turns Tailscale into an Atmosphère system module so homebrew like Moonlight on the Switch can join the tailnet directly, a concrete idea with clear boundaries. People streaming on hacked consoles will want it immediately; the limit is serving only already-jailbroken Switches.
Tailscale for the Nintendo Switch: an Atmosphère sysmodule that puts homebrew (like Moonlight) on your tailnet
A Self_Gradient_Forcing_Plus It decouples the parameters of context writing and denoising in autoregressive video generation, claiming a 5-second training window can generate up to 24 hours. The idea is concrete with complete code, an extension of the Self Forcing line. Whether it holds depends on VRAM and quality costs; suitable for long-video generation people to reproduce.
Autoregressive video generation with decoupled context-writing and denoising parameters, enabling up to 24-hour generation from 5s training windows.
A tf2mt Runs TF2 on Apple Silicon via Wine+DXVK+MoltenVK, listing tested fixes (mouse feel, menu stutter) plus a native launcher — directly useful to Mac players. The limit is that it serves one old game and isn't reusable across titles.
Team Fortress 2 tuned for Apple Silicon: Wine + DXVK + MoltenVK with measured fixes (smooth 120 Hz mouse, no menu stalls) and a native macOS launcher
A ApiHashing A Windows x64 API hashing library that bypasses EAF and resolves forwarded exports, ordinals and api-ms-* API Sets, from Maldev Academy — practical for red teaming and malware development. The description is precise and unhyped, but the audience is very narrow and it's a dual-use tool.
Windows x64 API hashing library that bypasses EAF and resolves forwarded exports, ordinals, and api-ms-*/ext-ms-* API Set DLLs.
A stocking-texture-tool Adds moiré-free knit texture to stockings in illustrations, a narrow but real problem, described accurately in both Chinese and English without hype. People doing AI image post-processing will want to try it; the limitation is it covers only this one material, so generality is limited.
Adds a realistic, moire-free knit texture to the stockings in an illustration.
A MirrorAct Mirrors iPhone/iPad screens to a Mac with device frames plus screenshots and recording, a real need for developer demos and App promo material. QuickTime only works over USB, so the AirPlay part is the differentiator; 11 stars is too early and stability remains to be verified.
Mirror, frame and record your iPhone or iPad on your Mac – via USB or AirPlay. Device frames, backgrounds, screenshots, recordings, presentation mode. Open source (GPL-3.0).
A OpenComputerUse An MCP server that gives agents background computer control, driving the desktop without stealing the foreground, written in Rust with a narrow scope. The pain point is real, but it's one day old, so the permission model and long-task stability are unverified.
Background computer use for agents, as an MCP server
A loadout A harness manager for Claude Code configs that switches by profile, shows where each setting comes from and toggles them individually, solving the real pain of config sprawl. Scope is restrained, but it depends heavily on Claude Code's config format and must chase upstream changes.
Claude Code harness manager: switchable profiles, scope provenance and on/off for every harness element
A MathModel-Copilot A math-modeling assistant and agent skill for CUMCM and MCM/ICM, covering problem reading, model comparison, running Python, result verification and paper support, matching real contest rhythm. The audience is competing students and the scope is clear; similar tools abound, so the key question is whether result verification is truly reliable.
A mathematical modeling AI assistant and Agent Skill for CUMCM and MCM/ICM: problem analysis, model comparison, Python execution, result verification, paper assistance and team collaboration. Mathemat…
A ts-rust Rewrites the TypeScript 7 compiler for order-of-magnitude faster type checking, hitting the JS ecosystem's most painful tsc performance bottleneck; ping.gg as author plus 434 stars in 0.8 days shows real development investment and community attention. If the performance holds it could be absorbed into build toolchains as a standard dependency or sold as a hosted type-checking service; the risk is competition from the official tsgo/native port route.
An experimental Rust port of the TypeScript 7 compiler (tsc)
A deckcraft It addresses a long-standing open-source gap: a clean reimplementation that can genuinely replace PowerPoint, including format parsing and rendering, rather than another HTML slide generator. 144 stars in under a day show real existing demand for presentation tools, and the ArtCraft series' steady release cadence adds momentum. Monetization lies in turning the Rust rendering kernel into an embeddable SDK/library for document and collaboration platforms. The biggest risk is the huge engineering scope (animations, charts, compatibility) and facing LibreOffice directly.
Presentations and slide shows: an open-source, clean-room reimplementation of Microsoft PowerPoint in pure Rust. Part of the ArtCraft Crafting Apps.
A Hyperliquid-Leaderboard-Analytics Views Hyperliquid leaderboards in the terminal: tracking addresses by PnL and ROI, separating "one lucky streak" from steady winners, using public read-only APIs and explicitly self-labeled unofficial. Practical for copy-trading and on-chain research; the difference is only the terminal form and tracking of ranking changes.
Hyperliquid analytics right in your terminal. Track top traders by PnL, ROI, and account value, monitor ranking changes over time, and identify consistently profitable wallets while separating them fr…
A fayda-mcp Connects Ethiopia's Fayda national digital identity (MOSIP eSignet) to MCP, letting agents perform zero-trust e-KYC verification. A concrete scenario with a clear target user, a rare integration with national-level identity infrastructure; the limitation is that the audience is basically local Ethiopian developers.
A Model Context Protocol (MCP) bridge connecting AI agents to Ethiopia's Fayda National ID (MOSIP eSignet) for zero-trust e-KYC and identity verification.
A ai-data-extractor It unifies conversation-history export across Claude Code, Cursor, Windsurf, Aider, Cline, and other assistants, making data retention and review easier. The need is real and the description plain; similar scripts exist scattered, so its value is coverage, while export-format depth still needs a look at the code.
Free open-source extractor for AI coding assistant chat histories. Supports Claude Code, Cursor, Windsurf, Aider, Cline/Roo Code, and more.
A BIGWORDS.PAGE Encodes the large text to display directly into the URL, so opening it shows a full-screen display with no backend and no account. It is for people who need on-site screen projection or slogan displays, and its restraint, with almost no extra features, is exactly the strength; the drawback is that it can only show plain text.
Full-screen text for any screen. The message lives in the URL, so there's no backend, accounts or storage.
A llm-openai-decisions A plugin by Simon Willison for the llm command line that connects to OpenAI's new Decisions API, letting one command try the new interface and log it into llm's system. It is small, and its value is the usual restraint and composability; people who do not install llm cannot use it.
LLM plugin for the OpenAI Decisions API
A foner Makes and receives calls on a KDE desktop through a paired Bluetooth phone, filling the HFP gap that Linux desktops have long lacked. Very practical for those using Linux as their main machine who must take calls; limitations are dependence on BlueZ versions and phone compatibility, and it is not yet packaged for distribution.
Make and receive phone calls through a paired Bluetooth phone, from your KDE desktop
A storage-rescue A Mac disk cleanup Agent Skill: verify first, then move to an external drive, with confirmation before every deletion. Plain text with no runtime and restrained boundaries—exactly what a skill should be; the limits are macOS-only coverage and cleanup policies you still must review yourself.
Agent Skill that frees up storage on a Mac without losing data: verified offload to an external drive, or safe cleanup with no drive. Asks before every deletion. Plain text, no runtime.
A Wobbly Brings the Compiz-era jelly window effect to macOS, implemented in Swift, scoped to just one thing. Mac users wanting some desktop fun should give it a try; it's purely decorative and doesn't solve productivity, and it tends to break after system updates.
Compiz-style wobbly windows for macOS
A Aegisub-NEXT Aegisub has gone unmaintained for years, and fansub groups have long lacked a modern successor; this project aims to be compatible with Aegisub projects and is rewritten in C#. The direction is right, but it gives no differences from the 3.3 branch or completion status, so for now it is worth watching, not migrating to.
A modern aegisub compatible subtitle tool
A matrix-multiplication-all-fields It generalizes the proof of the 9/4 matrix multiplication exponent to arbitrary fields in Lean, which is serious formal math work with a precise, unhyped description. It is useful to people working on Lean and algebraic complexity; the audience is very narrow, and ordinary developers will only get a spectacle.
Lean formalization work extending the matrix multiplication 9/4 argument to arbitrary fields
A Show HN: Parseable, an open observability datalake, handles 100M time-series/min An open-source observability data lake claiming 100 million time-series points per minute on a single machine, directly targeting Datadog/Grafana log and metric storage costs with HN 78 points of ops recognition; a hosted commercial version exists, giving a clear open-source plus cloud model, but the space is crowded with giants and migration costs are high.
A openchart From the same source as a related HN post, 80 stars in 1.3 days confirms OpenChart's user acquisition speed; an open-source chart plus agent workbench could become a standard dependency for trading tools, monetized through low-latency market-data subscriptions. The risk is that it is currently macOS Apple Silicon only, with narrow platform coverage and strong competition from TradingView.
OSS Tradingview with your own AI, free and unlimited.
A Show HN: OpenChart – OSS TradingView alternative with your own AI agent Solves the pain that agents can only interact with markets via CLI and cannot see graphical charts, running locally with private data; Show HN 35 points plus 80 stars show real early demand. Monetization could be low-latency paid quotes and team subscriptions, with the biggest risks being TradingView's ecosystem dominance and high market-data licensing costs.
Hi HN, I'm Weilun, cofounder of OpenChart. We built OpenChart because we wanted Claude and Codex to interface with the markets, like most of humans do, through charts rather than through CLIs. With yo…
A kandinsky-6 A foundation model that generates video and audio together, addressing audio-video sync, a major current weakness in video generation, with 129 stars in 1.6 days reflecting demand. It can monetize through both API and open weights, but the risk is being crushed by Runway, Sora and Kling on data and compute.
Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation
A cc-mod-image-view A small patch for Claude Code that shows a thumbnail after pasting an image instead of the [Image #n] placeholder. The cut is tiny but a real daily pain, and reaching thirty-plus stars in a day shows genuine need; the limit is relying on Claude Code's patch mechanism, which may break when versions change.
Claude Code mod: thumbnails of pasted images above the prompt instead of bare [Image #n] tags
A home-dns-adblock A home DNS blocking ruleset for Chinese apps like Hongguo and Tomato, also effective on the go with Tailscale, with 198 rules aimed at zero false positives. DNS ad blocking is not new, but the rules are concrete and the scenario clear, one of the few things today you can copy and use right away.
🏠 Home DNS ad-blocker for Chinese apps (Hongguo Short Drama/Fanqie Novel) — works at home AND away via Tailscale. 198 precise rules, zero false positives. Home DNS ad blocking, usable at home and on t…
A huashu-art-motion Packages 35 artistic-style animations into directly callable skills, filling the gap in artistic direction when agents generate motion; 628 stars in 0.8 days is the strongest growth in this batch, validating demand. It can monetize via skill distribution or design-tool plugins, but the risk is that the style library is easily copied and lacks a moat.
Art animation skill: 35 art styles, 9 narration grammars, bring drawings to life with code.
A Show HN: MailAccess – the true Email OSINT framework Codifies email OSINT methodology into a tool that outputs sources and confidence scores instead of bare lists, hitting pentest teaching and practice with 66 points of genuine security-community resonance; a hosted paid version already runs, enabling security SaaS subscriptions, with risks of data-source bans and compliance disputes.
Im a security engineer and i built mailaccess. When i started learning pentesting, i came across multiple lectures and notes of people listing out tools and websites, which gives the emails for a part…
A unstream-proxy A zero-dependency proxy that specifically fixes corrupted SSE streaming responses from Anthropic-family upstreams—a pit many have hit but nobody made into a standalone small tool. Scope is clear in one sentence and it works once installed; the limit is covering only this one upstream type, so the audience is narrow.
Zero-dependency proxy for Anthropic-type upstreams with broken SSE streaming
A punctum An adjustable dot-matrix font with two axes, dot diameter and dot shape, on a 5×7 grid, previewable directly in the browser. Designers and front-end developers building pixel-style interfaces or posters can try it today; the scope is clear, the description honest, with no extra packaging. Limitations: it covers only the dot-matrix style, glyph coverage and licensing are unstated, and comparable options are existing pixel font packs.
Punctum: a free variable dot-matrix typeface. 5×7 dots, two axes: dot size and dot shape. A variable dot-matrix font.
A picofold A readable PyTorch protein structure predictor of about two hundred lines, explaining the core steps of folding in a minimal way. Suitable for people new to structural bioinformatics, its value is teaching rather than accuracy, occupying the same niche as mini AlphaFold reproductions.
A *minimal* protein structure predictor written in ~200 lines of *readable* PyTorch.
A TWLazUI A Tailwind-style component library for Lazarus/Free Pascal built on BGRABitmap, an area almost nobody covers, useful to desktop Pascal developers. Scope is restrained and the description honest; limits are the tiny ecosystem and unproven theming and documentation depth.
"A modern, Tailwind CSS-inspired UI component library for Lazarus and Free Pascal, powered by BGRABitmap." (Pustaka komponen UI modern bergaya Tailwind CSS untuk Lazarus dan Free Pascal, ditenagai ole…
A uwpatch Patches Ultima Underworld's UW.EXE so floors and ceilings render with correct perspective. Extremely narrow but very clearly explained, and the author has clearly dug into that old engine; well worth a look for DOS-era game-fix enthusiasts, though you must supply the original files and it serves only one 1992 game.
uwpatch patches the original DOS game’s UW.EXE so floors and ceilings are drawn perspective-correct.
A agent-duel A local desktop app that has two coding agents blind-judge the same prompt head-to-head, turning which model is better into your own reproducible vote. Simple idea but solves real confusion, more trustworthy than leaderboards; the limitation is that winning still relies on subjectivity.
Blinded battles between coding agents: two agents, one prompt, you pick the winner. Desktop app, runs locally.
A claude-my-privacy Replaces real names of people and companies with pseudonyms inside Claude Code, restoring them in tool calls and replies, filling the privacy gap of coding assistants sending context out—more practical than just not using them. Small scope, clear docs, but Claude Code only, and you maintain the replacement rules.
MyPrivacy: local name and company pseudonymization for Claude Code, with restoration for tools and assistant reply display.
A kube-bmc Turns bare-metal server BMCs into K8s resources, with health status, Prometheus metrics, and out-of-band power control, aimed at operators managing their own data centers. Lighter than Metal3 with zero-config discovery as a highlight, but the feature list is rather full and real-world readiness remains to be seen.
Your servers' BMCs as Kubernetes resources: zero-config in-band discovery, hardware health, Prometheus metrics, a slick dashboard and out-of-band power control. IPMI & Redfish.
A Pawboard Pins a whiteboard beneath the Windows desktop icon layer so doodles stay on the wallpaper. The idea is concrete and the niche small, more appealing than yet another notes app; the limits are Windows-only support and a very narrow feature set, making it a small utility rather than a productivity mainstay.
A whiteboard that lives on your Windows desktop, behind your icons 🐾
A MinigamesBenchmark A seventy-mini-game benchmark from CMU testing computer-use agents' various fine-grained skills; it exposes weaknesses better than generic agent evaluations, and GUI agent builders can test directly. Limited in that game coverage bounds how far conclusions generalize, and community scores aren't in yet.
A benchmark of seventy minigames for computer use agents, testing a variety of skills. Think your agent can beat them all?
A sd-report-collector The server side of a systemd-report uploader: mTLS intake, zstd archiving by hostname, then plugin analysis, by a longtime systemd/pam maintainer. People managing Linux fleets who need to preserve failure scenes will like this narrowly scoped tool. The limit is that the companion collector is not yet formed, so it is more a slice of the author's own infrastructure.
Server counterpart to "systemd-report upload". Accepts report pushes over mTLS, stores each one zstd-compressed under a per-hostname directory, and runs configured plugins against the saved file.
A PSXS5 A native port of a PS1 emulator based on PCSX-ReARMed to jailbroken PS5, adding a cover shelf and xBR upscaling, with origin and features clearly stated. PS5 homebrew users can use it directly; the originality is mainly in the frontend and adaptation layer.
PSXS5 (PlayStation X Super 5): a PlayStation 1 emulator running natively on jailbroken PS5, based on PCSX-ReARMed. Cover-flow shelf, xBR upscaling, RetroAchievements, cheats.
A Show HN: Minigraf – An embedded, bi-temporal graph database in Rust A self-contained bitemporal graph database crate giving agents embeddable graph memory without an external database; 31 HN points and 108 stars show developers are paying attention, and it is the author's second iteration. It could become the standard agent memory layer or a hosted service, but the graph DB space is crowded and bitemporal demand is niche.
A successor project to my previous attempt at building a temporal graph database ( https://news.ycombinator.com/item?id=23455516). This one builds on its predecessor's concept, but drops the dependenc…
A simcity3000-gta-mode A native plugin for 1999's SimCity 3000 that lets you possess citizens, steal cars and get arrested—reverse-engineering the old engine to put GTA into a city builder. Very geeky and modestly described, but only meaningful to players still running SC3000.
GTA mode for SimCity 3000 Unlimited: possess citizens, steal cars, fight, get busted. A native plugin for the original 1999 game.
A iron-laws A CLI check for the security, structure and maintainability of AI-generated code, hitting the real pain that AI code needs automated review, with a decent start at 108 stars. It could monetize via paid CI scanning or an IDE plugin, but risks competing with CodeQL, Sonar and various AI linters, needing unique rules to break out.
Iron Laws - a source code inspection CLI that points out security, structure, and maintainability problems in AI-generated code.
A Deeds Reads the code diff of each commit and classifies it as new capability, fix, or maintenance to quantify actual AI coding output, which is more informative than counting PRs. It is a CLI plus Claude Code plugin you can try today; the limitation is that classification still relies on model judgment and can easily become a KPI.
A CLI and Claude Code plugin that reads every commit's diff and counts what actually got better: new capabilities, fixes, and upkeep. A way to measure work done with AI that doesn't depend on counting…
A manifold-csg-unity Brings the Manifold geometry kernel into Unity, offering runtime boolean operations (union, intersection, difference) that can be repeated. Game developers building level cutting, destruction or modeling tools can install it today; the scope is restrained and accurately described, making it one of the most practical small libraries here.
Runtime CSG for Unity – subtract, union and intersect meshes, again and again. Powered by Manifold.
A brand-identity-skill A Claude Code skill that generates logo, typography, and color palette from the same brief and outputs a brand guidelines kit; for indie developers needing quick proposals, it is more systematic than one-off image generation. The limitation is that review still relies on model self-assessment, and deliverables are mostly concept drafts.
A Claude Code skill that designs a brand identity as one system: logo, typography and colour palette built together from one brief, measured, compared as identity cards, and delivered as a brand guide…
A SCM The repo behind a 134-point Show HN, with 246 stars in 1.3 days, doing deep retrieval over local photos and video frames and hitting both privacy and efficiency needs, with native macOS experience as a differentiator. Open source plus a subscription indexing service could monetize, but Apple and Google's free semantic search keeps closing in and local models carry compute costs.
Deep AI search for every photo and every frame of video in any folder on macOS
A Show HN: AI search for every photo and every frame of video on macOS Semantic search over every local photo and video frame, hitting the macOS pain of finding images without uploading to the cloud, with privacy as a hard selling point. 134 points on Show HN is a hot post and the linked repo already has hundreds of stars, doubly validating demand; monetization is an open-source kernel plus subscription indexing; the risk is Apple's free visual search closing in.
A live-panel-skill Turns architecture diagrams from manual drawing into a JSON-driven animated video or web page, hitting the repetitive work of documentation and presentations. 429 stars in 1.3 days, released as a Claude Code skill, rides the agent-skill ecosystem boom. It could become a content-production SaaS or be built into documentation tools; the risk is that Mermaid, Graphviz, and AI image generation already cover most scenarios, leaving a thin moat.
Config-driven animated architecture diagrams: turn one JSON file into a terminal-style, always-running diagram or a light-theme infographic that moves. Outputs H.264 mp4 or a live web page; also a Cla…
A hermes-gadget-sdk Provides ESP32 firmware, a Linux client and a desktop simulator for Hermes Agent, bringing voice-text agents onto physical devices and targeting the AI hardware long tail; 177 stars in 1.6 days is good, and the SDK form could become a standard dependency. Risks are Hermes' small ecosystem and competition from OpenAI and the ESPHome voice stack.
Build voice and text devices for your Hermes Agent, with ESP32 firmware, a Linux client, and a desktop simulator.
A hashyy Runs a full unpruned 177B MoE on a 12GB GPU at 11.5 tok/s by adding expert streaming to llama.cpp, with measurement scripts and documented dead ends. A rare empirical result for local large-MoE runners, though still too slow for daily use.
A full, unpruned 177B MoE model at 11.5 tok/s on one 12 GB GPU. Expert streaming for llama.cpp, with the measurement harness and every dead end.
A omnarr Combines Calibre, Audiobookshelf, Jellyfin, Sonarr/Radarr, Komga, and RomM into one search, browse, request, and playback entry, hitting the real pain of scattered self-hosted media stack entry points. The integration list is specific and honest; the risk is wildly differing APIs, hard to fully connect in one day, and it overlaps with portals like Organizr.
One front end to search, browse, request and play everything on your self-hosted media server (Calibre, Audiobookshelf, Jellyfin, Sonarr, Radarr, Komga, RomM…)
A tmux-scrollable Brings niri/PaperWM-style horizontal scrolling tiling to tmux: new columns slide in from the right instead of squeezing existing panes. Small and clear, it will immediately appeal to anyone running a dozen panes in tmux, and the script is easy to try; the risk is tmux layout API limits, with stability and plugin compatibility unverified.
Scrollable tiling for tmux, like niri and PaperWM: new columns scroll in from the right instead of squeezing your panes
A gears-judgment-recomp Statically recompiles Xbox 360's Gears of War: Judgment into a native Windows program with high frame rates and widescreen, requiring your own game files. It is the furthest along among recent 360 recompilation projects, but stability and compatibility remain completely unknown.
Gears of War: Judgment for PC, a native Windows build made by statically recompiling the Xbox 360 game. Bring your own game files.
A lowpaper A bash wrapper around mpv for dynamic video wallpapers on Linux, compatible with X11 and Wayland and designed for old, low-spec machines. Video wallpapers have long been missing on Wayland, and this approach is small and direct; the feature set is thin, so complex playback needs mean editing the script yourself.
Ultra-lightweight live / animated video wallpaper for Linux (X11 & Wayland), designed for low-end and old hardware. Minimal Bash script around mpv.
A auto-handoff A Claude Code plugin that automatically writes a handoff document when a session goes idle and the prompt cache is about to expire, so the next conversation can continue; it targets the real pain of long-task cache invalidation with a very narrow cut. The alternative is taking notes yourself, so whether it's worth installing depends on how much you rely on Claude Code.
Claude Code plugin that writes a handoff document before an idle session's prompt cache goes cold
A Jelly5 A native Jellyfin client for jailbroken PS5s with its own interface and hardware-decoded 4K HDR playback. For people who want to use a console as a living-room player with a self-hosted library, there is almost no alternative; the barrier is the jailbreak requirement, the audience is narrow, and long-term maintenance is uncertain.
A native Jellyfin client for jailbroken PS5: its own liquid-glass interface, hardware-decoded 4K HDR
A bw-picker Adds global autofill to Bitwarden on Windows: a hotkey types passwords into games, launchers, Discord and other desktop apps, calling the official CLI and openly labeled unofficial. It fills a long-standing gap, but only runs on Windows and depends on CLI sessions.
Bitwarden auto-type for Windows: fill logins into any desktop app (games, launchers, Discord) with a hotkey. Unofficial; uses the official Bitwarden CLI.
A edgextract Extracts markdown into a knowledge graph based on a self-written ontology, using closed yes/no judgments instead of free-form JSON—a rare ontology-first-then-extract approach for RAG and document structuring. 37 stars show some recognition, but ontology design has a high bar, and results depend on how detailed you write it.
Turn markdown into a knowledge graph with an ontology you write and a cutoff you own. Closed yes/no decisions, not free-form JSON.
A ps5-ai-cli Puts the Codex CLI into PS5 homebrew, using a phone browser as terminal and the controller to type—a typical "just make it run" experiment. Interesting for console tinkerers wanting AI on the TV; but usability depends entirely on the console environment, the description omits dependencies and limits, and it's closer to a demo than a finished tool.
Run OpenAI Codex on PS5 with controller text entry and a browser terminal for your phone or computer. Experimental homebrew.
A SharpForge A C# subset compiler in JavaScript emitting ECMA-335 IL/PE, with its own managed runtime, debugger, and browser IDE. Ambitious but clearly scoped, it's a rare readable sample for compiler and .NET fans; subset coverage and performance are unstated, making it more a teaching experiment than a usable toolchain.
C# subset compiler written in JavaScript with ECMA-335 IL/PE output, a managed runtime, debugger and browser IDE (SharpForge Studio).
A answer-me-with-html Lets agents answer complex questions with a single readable HTML page, hitting the pain of hard-to-read long LLM text answers; 305 stars in 1.6 days is the strongest growth in this batch, showing real demand for output formats. The path is becoming a standard skill in the agent-skill ecosystem and being absorbed by IDE-like products; the risk is a very low barrier, easily swallowed by native rich model output or similar skills.
Answer me with HTML — an agent skill that answers hard questions with a one-page HTML you can actually read. Let AI Agents answer complex questions with a single page of HTML.
A Navi48-MacOS Gets an AMD RX 9070 XT to render images on macOS via a custom kext, a RADV port to Darwin, and a Metal-on-Vulkan layer. This is driver-level dirty work few would attempt, and success would matter a lot for Hackintosh workstations; it's only a PoC now, with app coverage and stability unverified.
Proof-of-concept native macOS support for the AMD RX 9070 XT (Navi 48): bring-up kext, RADV Darwin port, Metal-on-Vulkan bundle
A ESP32-DATV Builds a DVB-S amateur TV transmitter on a bare ESP32-C3, from QPSK to 16APSK in the 13cm band with no extra RF hardware — an embedded project that pushes one thing to its limit. It's appealing to hams and SDR hobbyists; the barrier is needing a license and real-world verification, so ordinary developers can only watch.
DVB-S digital amateur TV transmitter using a bare ESP32-C3 — up to 1 MSymbol/s QPSK, 8PSK, 16APSK in the 13 cm band, with no additional RF hardware.
A eslint-cssicorn Turns CSS checks into ESLint rules, from a trusted name in the ESLint ecosystem, with restrained naming and scope. Suited to frontend teams already on ESLint who don't want another stylelint setup; limits are mature CSS lint options exist, so check which extra rules it adds.
Powerful ESLint rules for CSS
A esp-idf-injection-ng A one-line objcopy weak-symbol patch on ESP-IDF v5+/v6+ that enables native 802.11 frame injection, useful for Wi-Fi security research or ESP32 low-level work. It clearly explains which symbol is changed; the limit is that it only targets new IDF, so older projects and existing injection libraries need manual migration.
Raw 802.11 frame injection on ESP-IDF v5+/v6+ via a one-command objcopy --weaken-symbol patch. Enables Authentication (SAE Commit), Deauth, Reassoc, Disassoc, and Control frames (RTS/CTS/ACK) through …
A lowpaper-live-wallpaper A few dozen lines of Bash wrapping mpv to give X11 and Wayland a dynamic wallpaper that runs on low-end machines, explicitly optimized for old hardware. The scope is clean and the description honest, and lightweight options are genuinely scarce on Linux desktops; limits are that it only supports formats mpv can play and has very few config options.
Ultra-lightweight live / animated video wallpaper for Linux (X11 & Wayland), designed for low-end and old hardware. Minimal Bash script around mpv.
A ctx-handoff-mod Automatically hands off to a new session when context fills up while keeping the prompt cache warm, hitting Claude Code's long-task interruption pain point. The approach is clear and usable now; the risk is relying on undocumented cache behavior that an official change could break.
Claude Code mod: auto handoff to a fresh conversation when context hits a threshold, keeps the prompt cache warm while you're away
A book-to-course-skill Turns EPUB/PDF books into beginner courses with quizzes, flashcards, and unit-test-checked exercises, a rare complete loop among Claude skills, suited to self-learners or internal training. Features are detailed but there's no usage feedback yet, and you must handle copyright for source books yourself.
Claude skill that turns a book (EPUB, PDF, HTML, DOCX, Markdown) into an interactive beginner-friendly course: step-by-step lessons, animations, quizzes, chapter tests, flashcards and coding exercises…
A kyojin A ROCm inference engine for AMD Strix Halo (gfx1151) built on ExLlamaV3, capable of running 300B-class MoE models on a single 128GB mini-PC. A real gain for local-inference users who can't afford multi-GPU setups; the catch is it covers only one architecture and is tightly bound to upstream.
Kyojin: the Yamz inference engine for AMD Strix Halo (ROCm, gfx1151), built on ExLlamaV3. Runs 300B-class MoE models on one 128 GB mini PC.
A ccprogress Adds terminal and desktop progress bars to long-running Claude Code sessions, easing the black-box anxiety of waiting, and works out of the box. A tiny but real patch-style tool; similar status plugins abound, and keeping up with CLI version changes is its biggest uncertainty.
Live step-by-step progress bar for long Claude Code sessions, in the terminal and the Desktop app
A dgx-spark-memory-saver A UVM page-table packing patch for DGX Spark that reclaims usable memory from 64KiB kernels, with build tools and measured data. The audience is small but the problem concrete, and laying out the verification process makes it far more credible than performance-bragging peers.
Recover usable memory on DGX Spark with a 64 KiB kernel: an opt-in NVIDIA UVM page-table packing patch, build tools, and measured validation.
A remocn-studio An agent-native video editor that lets AI operate the timeline directly rather than just spitting out assets, hitting the real pain of time-consuming editing; 96 stars in 1.4 days lands right on the agent-operating-creative-tools wave. The path is a subscription editing tool or retreating to an agent interface layer for editing software. The risk is CapCut and Descript doing the same agentification, so it must hold its lead through workflow depth.
Agent-native video editor
A ai-agent-gateway An open-source gateway to control and audit agents' MCP tool calls and LLM traffic, meeting enterprises' compliance need to avoid granting agents broad permissions; 48 stars in 1.5 days is early but the direction is right. The path is clear: security team procurement, charged by traffic or seat, with the LLM gateway ecosystem already proving budgets exist. The risk is cloud vendors and model platforms offering built-in governance, so it must hold differentiation through MCP depth.
AI Agent Gateway - an open-source gateway that secures, governs and observes AI agents' MCP tool calls and LLM traffic
A bootloops Provides certifiable high-precision physics and quantitative computation engines for LLM agents, hitting the pain that agent-generated numerical code is untrustworthy with uncontrolled error; 146 stars in 1.4 days show quant and research users actively seeking such a foundation. The path is a trusted computation API agents can call, entering research and quant procurement. Risks are high verification cost, a narrow user base, and competing with SciPy/JAX for the standard slot.
BootLoops 1.0: certified computational tools and house engines for exact and high-precision physics and quantitative science, built to be driven by LLM agents. MIT; docs CC BY 4.0.
A Defender-Bender-Box Wraps the Windows Defender x64 engine in Docker so you can drop suspicious files in on Linux and get local scan results, saving sample-triage researchers from spinning up a Windows machine. Scope is restrained and the description does not overclaim; limits are single-engine results weaker than multi-engine sandboxes, and you must confirm licensing before redistributing Microsoft components.
A Docker container running the x64 Windows Defender engine to scan files for malware. Put suspicious files in the box; let Docker bend Defender; get a local scan result
A spark-scope A read-only dashboard for new machines like DGX Spark, GX10 and EdgeXpert, showing vLLM/SGLang inference status and a rack view. It is ready to use out of the box and stays narrowly scoped, but it only supports these few devices, and general GPU monitoring already exists via nvtop and DCGM.
Read-only dashboard and rack panel for DGX Spark / ASUS GX10 / MSI EdgeXpert nodes running vLLM or SGLang
A dell-dw5821e-dw5829e-esim-linux A complete practical write-up of using Dell DW5821e/5829e cellular modems on Linux with lpac for eSIM, QMI/PPP dialing, udev and netns routing. Few people have these modules and fewer can explain them clearly in one place; the limit is that it covers only these two cards and will age with kernels and firmware.
Dell DW5821e eSIM & DW5829e eSIM (Snapdragon X20) on Linux: eSIM profiles with lpac, QMI/PPP data sessions, udev rules, per-uplink netns routing and recovery notes
A gpuslider A shader-drawn slider using WebGPU with a WebGL2 fallback, 4.7KB core with everything else as plugins. The scope is cleanly cut and the promises match the code size; worth a look if you want non-standard controls, but don't expect it to work out of the box.
A slider drawn by shaders: WebGPU, or WebGL 2 where there is none. A 4.7 KB core, everything else a plugin.
A skkmonkey Turns the SKK Japanese input method into a pure front-end userscript that runs directly in Firefox and Chrome without a server, solving the awkwardness of installing Japanese input on Chromebooks and Linux browsers. The scope is small and the privacy notes restrained. The limitation is that it only serves longtime SKK users, far narrower than mozc.
A local SKK Japanese IME for Firefox and Chrome, written in TypeScript and built with Node.js. Runs as a userscript; no server is required and typed text is never sent over the network.
A Show HN: Ledge.sh – Runnable Markdown Notes Embeds executable commands and SQL directly into Markdown notes, eliminating copy-paste between notes and the terminal; 74 points is the top HN heat in this batch, proving the pain is widely validated. Monetization can come from ledge-server team collaboration paid plans. The risk is many competitors like Obsidian and its plugins and various runnable-note tools, plus security concerns from executing arbitrary commands.
Hi HN, Ledge is a Markdown notebook that runs shell commands, code, SQL, etc from inside your own notes. I built Ledge because I spend much of my day copy/pasting commands from my notes into the termi…
A DeepEP-Ascend A high-performance collective communication library for Ascend training and inference, filling a key gap in the distributed stack; 182 stars appearing alongside DeepGEMM shows DeepSeek's systematic investment in domestic hardware stacks. It could become a standard dependency for Ascend distributed training and bring cloud vendor partnerships. The risk is fast ecosystem iteration and CUDA plus NCCL still dominating the market.
A high-performance communication library for machine learning training and inference on Huawei Ascend NPUs.
A universal-modder Chains reverse engineering, fal asset generation and in-game testing into Claude Code skills so non-programmers can mod PC games; 802 stars in 0.8 days comes from the novelty of AI meeting modding. Mod distribution or asset generation could monetize, but the risk is game-vendor DMCA and anti-cheat, plus mods lacking sustained payment scenarios, making it more a tool showcase.
Point Claude at any game. Skills, tools and the fal MCP that let Claude Code mod almost any PC game you own: recon, reverse engineering, fal-generated art/3D/audio, in-game testing, showcase videos.
A Show HN: I built a free Burp/Caido alternative but, zero setup - API Testing Solves the high barrier of initial Burp/Caido setup and APK traffic capture, with zero-config out-of-the-box use as the core selling point; 30 Show HN points show practitioner resonance but no breakout yet. Monetization can be pentest team subscriptions or SaaS. The risk is that Burp and Caido already own mindshare, security tool trust builds slowly, and single-person maintenance can't support enterprise features.
This is Katriel, the architect of the tool APIaxess. I have been API pentesting for a good time now, and starting with API pentesting was a bit of a rough patch. Specially the setting up and having to…
A TIRx-harness An open compiler harness for agentic GPU programming, giving AI-written kernels a unified compile and evaluation entry point; backed by mlc-ai, 75 stars is few but a strong quality signal. It has the best chance to become a standard dependency and benchmark for GPU code-generation agents, but the risk is a very early field where NVIDIA's closed tools and the Triton ecosystem hold ground.
An Open Compiler Harness for Agentic GPU Programming
A oido Packs open-vocabulary speech recognition into a $5 ESP32-S3 at 3.7% WER without cloud or NPU, bringing ASR to ultra-low-cost IoT; 45 stars in 0.5 days, with the metric itself as the传播 point. It is most likely to enter IoT and toy supply chains via firmware licensing, but the risk is the metric's credibility in real noisy scenes and competition from native solutions like Espressif.
Oído: open-vocabulary speech recognition that fits in a $5 ESP32-S3. 3.7% LibriSpeech WER, no cloud, no NPU.
A hypoarena Uses synthetic literature with controllable causal chains and Elo matchups to validate hypothesis-discovery algorithms, solving the lack of ground truth in research agent evaluation; 545 stars in a day reflects the research community's hunger for reproducible evaluation. Commercialization is narrow, and it's more likely to become a benchmark component for research agents. The risk is an audience limited to ML researchers with no clear paying party.
Scientific hypothesis-discovery workbench: grounded hypothesis-evidence graphs, synthetic literature with planted causal chains, generate-debate-evolve loops over pluggable offline adapters, Elo tourn…
A compress Compresses Codex tool output with trajectory-faithful compression to save about 30% of input tokens and cache costs, hitting the agent billing pain point; 67 stars and a Show HN appearance validate the demand. The author offers an SDK and API that could grow into cost-optimization middleware, but the risk is native cache optimization by model vendors and accuracy loss from truncation.
Free CLI from Everest that compresses Codex tool output.
A Show HN: Strata – an expressive semantic layer that can say no to your LLM A semantic layer with strict naming and partition-aware routing lets LLMs reject wrong definitions, solving inconsistent metrics in self-service analytics and agent data retrieval; 18 points is low because it's a closed-source beta. It offers 25 free users plus enterprise subscriptions, directly against Looker and Cube. The risk is that closed source weakens spread and dbt Semantic Layer already occupies the ecosystem.
Hello HN, I'm Ajo and I built Strata. I spent 4 years at Netflix solving self service for non-technical business users. I think I cracked it with my unique approach to semantic layer design. The key c…
A Ely-GPUI-Components A light and dark theme component library for Zed's GPUI, filling the biggest component gap in this emerging UI framework; 235 stars in 0.8 days shows scarce supply and hungry demand among GPUI developers. It could become the de facto GPUI component dependency and lead to commercial projects, but the risk is limited GPUI adoption while React and web component ecosystems remain dominant.
A component library for GPUI, in light and dark. Every component runs live in the browser.
A passportsim Emulates an ESP32-C3 in the browser via WASM or on desktop, running unmodified ESP-IDF firmware, built specifically for developing FoloToy AI Passport without hardware, plus an MCP service for agents to call. The scenario is small and specific and hardware-free development is a real pain point; the limitation is being tied to a single firmware.
Develop and debug FoloToy AI Passport firmware without the device: an ESP32-C3 emulator that runs unmodified ESP-IDF images in the browser (WebAssembly) or on macOS and Windows, with a CLI and an MCP …
A maclator A 'reverse Rosetta' for running arm64 macOS programs on Intel Macs: an interpreter with an x86-64 JIT, AOT cache, and a Metal bridge to the host GPU. The direction is rare and the description doesn't oversell; worth reading for binary-translation folks and those extending old machines' lives, though maturity and compatibility are clearly early.
Run arm64 macOS apps on Intel Macs: a reverse-Rosetta emulator (interpreter + x86-64 JIT + AOT cache) with a Metal bridge to the host GPU