Worth watching304
2026-09
A AIHOT Solves the repetitive work of aggregating RSS and social sources into an industry daily-news site, with a framework design where changing sources changes the site. 3,000+ stars in a day validates demand for automated content sites. The business path is an intelligence briefing SaaS for vertical media and enterprises, but LLM-written content is highly homogenized, and the biggest risk is general AI summarizers and RSS aggregators siphoning the value.
A website framework that finds trending topics on its own and writes its own daily digest. Swap in your own sources and curation criteria, and it becomes your industry trending site.
A jeeves PostHog's official account builds inference enhancement for Jev-style decision models, with 267 stars in a day and backing from a well-known vendor—the most endorsed of this batch's 'Jev' signals. If Jev becomes a lightweight decision paradigm, the inference layer could be absorbed by agent frameworks as a standard component. The risk is an overheated Jev ecosystem with unvalidated real demand and no independent user data.
Jeeves – Reasoning improves Jev-like decision models
A tusk A macOS-native, keyboard-first database client supporting 20 databases plus an AI assistant, directly targeting the paid market of TablePlus and DBeaver where demand is rigid and willingness to pay is clear. Rust-native performance and open-source freedom are the differentiators, but the database-client space is mature with high switching costs, so it must open a gap through AI-assisted querying.
A fast, native, keyboard-driven database client for macOS. 20 databases, an AI assistant that shows its work, free and open source.
A parquet-xray A Parquet file inspector that lays the bytes out for you, aimed at data engineers debugging encoding, compression and page layout. The scope is small and specific, filling the gap where parquet-tools gives only metadata and no byte view. The library is still small, but the direction is clear.
See how a Parquet file is laid out, byte by byte
A homedroid Turns an idle Android phone into a home server without root: SSH, web hosting, Jellyfin hardware transcoding, Immich, qBittorrent, Home Assistant and a dashboard, with a checklist concrete enough to follow. Very practical for anyone wanting a free low-power NAS. The downside is that no one guarantees long-term stability or cooling.
Turn an old Android phone into a home server: SSH, web hosting, Jellyfin with hardware transcoding, Immich, qBittorrent, Home Assistant and a web dashboard. No root.
A elm-compiler A Rust rewrite of the Elm 0.19.1 compiler, published as a drop-in replacement on npm, with a narrow scope and restrained claims. A real option for people still maintaining Elm projects who find the official compiler slow; the risk is a stagnant ecosystem with long-term maintenance by one person.
Fast, unofficial Elm 0.19.1 compiler written in Rust — drop-in replacement for elm (npm: @spottt/elm-compiler)
A claude-sap-kit Connects Claude Code to SAP: running ABAP reports, previewing NAST print output as PDF, and editing SmartForms as XML. A real time-saver for ABAP developers who tweak reports daily, and a far more specific entry point than general coding assistants. The limitation is dependence on its own ARC-1 extension, with thin docs and ecosystem.
Claude Code plugin + ARC-1 extension + ABAP kit: run SAP reports, preview NAST print output as PDF, edit SmartForms as XML
A synpath Unifies Kalshi and Polymarket market data and order placement into one Python SDK, with tick-by-tick order book history. It's convenient infrastructure for prediction-market strategists; the cost is heavy dependence on both exchanges' APIs, with stability yet to be verified.
One open-source API for prediction markets. Trade Kalshi, Polymarket and Polymarket US with one Python SDK: unified market data, smart order routing across venues, and advanced order types. Plus tick-…
A ThermalVortex-Mod An unofficial character mod for Slay the Spire 2 with 119 cards, a monster field, fusion summons and a meta reward pool, plus a directly installable package, making it an immediately usable sample for Chinese players and mod authors. The limitation is tight coupling to a single game version, worthless outside that circle.
Yugi Muto | Unofficial character mod for Slay the Spire 2. 119 character-related cards, monster field, Fusion Summon, and meta reward pool building, with a directly installable mod pack.
A vita-flash-emulator Brings Ruffle to the PS Vita; one VPK install lets you pick games and play Flash, with touchscreen, button mapping and a pause menu. The platform is niche but the need is real, and it honestly notes it is based on Ruffle. The limitation is that Flash content itself is dying, so the audience is limited to Vita homebrew players.
A Flash player for PS Vita, built on Ruffle. One VPK: launch it, pick a game from your library, play — with touch, button mapping and a pause menu.
A Show HN: OpenAPPA – open-source deterministic guardrails that don't break agents Provides deterministic guardrails for multi-tool agents: a data-oriented policy language plus DualLLM, claiming leakage drops from about 10% to single digits without sacrificing utility. Enterprise agent deployment needs permission and leak-prevention consoles, a clear budget line backed by papers and benchmarks, but Cedar/OPA and big-vendor native policy systems compete, and only 23 HN points suggests real customer adoption is still needed.
Hi Hacker News! Matvey, one of the authors, is here. While building enterprise agents, we ran into a problem: the more tools you connect to the AI, the higher the chance it will run out of control and…
A BaryBind Research code framing multimodal alignment as a Wasserstein barycenter problem, credited as a NeurIPS 26 Oral and aimed at researchers in multimodal fusion; the wording is precise and the idea concrete. The limit is paper-level code lacking reproduction scripts and baseline comparisons, so whether it runs remains to be seen.
(NeurIPS 26 Oral) Binding Multiple Modalities via Multimodal Wasserstein Barycenter
A action-dist-verify It rebuilds JS Actions pinned in workflows and compares the committed dist against source, filling exactly the supply-chain poisoning gap, usable with one command in CI. The limitation is that it only covers JS Actions and the community hasn't validated it yet.
Rebuild the JavaScript GitHub Actions your workflows pin and check that their committed dist matches the source.
A shellstrict It adds a check ShellCheck doesn't cover to shell scripts: missing set -euo pipefail, and no trap cleanup after mktemp. People writing CI pre-hooks will want to try it—small scope, honest description, easy to install as a Go single binary. The limitation is narrow capability, with shellharden already covering the strict-mode part; 13 stars don't yet show real usage.
Lint shell scripts for missing strict mode and for mktemp files with no trap cleanup, complementing ShellCheck.
A booking-truth Judges AI scheduling agents by final calendar state rather than conversation logs, with fault injection, pass^k, and false-success-rate metrics, plus a constrained scheduling implementation. It offers a rare acceptance-testing perspective for people shipping agents, limited to scheduling-type tasks.
Tests any AI appointment-setting agent on the calendar's end state, not its transcript: fault injection, pass^k, false-success rate. Ships a guarded Cal.com / Google Calendar / HubSpot booking agent.
A taskdistill Distills an expensive LLM call into a small local model: capture traffic, clean it, LoRA on MLX, evaluate against the teacher, then fall back via an OpenAI-compatible cascade. The pipeline is complete and clearly explained, suited to individual developers cutting inference costs; it is only convenient on Apple Silicon.
Distil an expensive LLM API call on a narrow task into a small local model: capture traffic, curate, LoRA fine-tune with MLX on Apple Silicon, evaluate against the teacher, and serve an OpenAI-compati…
A anotherworld-zx-spectrum-48k A genuine port of the Another World virtual machine to the 48K ZX Spectrum, complete with intro, a very narrow but very hard concrete achievement. Worth a look for retro computing and game porting fans; the limitation is a tiny audience and a high barrier to entry.
Actual VM implementation of Another World game, with intro
A price-check Aligns and validates OHLCV data across multiple market sources, offering adjusted returns, deviations and a consensus dataset to help quant and backtest people catch dirty data. The scope is narrow and the problem concrete rather than yet another framework; the limit is that it only covers the validation step, is one day old, and does not yet say which exchanges it supports.
Cross-provider market data validation for OHLCV prices, adjusted returns, discrepancies, and consensus datasets.
A shahrah A proxy that does geo-sharding at the Postgres protocol layer: sharding by key, read/write splitting, and online migration of a key to another region, all without changing drivers. Attractive for data-compliance teams, but whether it actually works needs verification.
A PostgreSQL proxy that keeps each user's data in their own region. Shards by a key you name, splits reads from writes, pools connections, and moves a key between regions while the application keeps r…
A geometry-as-address A paper repo from Zhejiang University 3DV that treats geometric positions as addresses and routes attention to visual memory for long-sequence camera-controllable video generation; the topic and direction are clear and worth following. The limit is that only paper-level explanation exists so far, with no code or weights, so it cannot be used soon.
[ARXIV 2026] Geometry as Address: Routing Attention to Visual Memory for Long-Horizon Camera-Controlled Video Generation
A onetake Packages continuous long-take filmmaking into a Claude Agent Skill, using an oracle to verify shot continuity against the most common AI-video problem of jarring cuts. 540 stars in 1.8 days shows real marketing-video demand and low skill distribution cost; it could be agency templates/subscriptions or be adopted by video tools as a standard skill, but many similar skills appeared within a week and output quality depends on upstream models, leaving a thin moat.
Motion films that never cut to the next slide: every beat grows out of the one before, one continuous camera, continuity measured by an oracle. A Claude Agent Skill for product launch films and featur…
A mcp-audit-tool Scans MCP server configurations for tool poisoning, rug pulls, hardcoded secrets, command injection and supply-chain risks, outputting SARIF directly into CI right as MCP security governance has a vacuum. 105 stars in 1.9 days shows good momentum, with a clear path as an enterprise CI insertion point, security SaaS, or adoption into official toolchains; the risk is Anthropic or open-source peers like mcp-scanner standardizing first.
🛡️ Security audit CLI for Model Context Protocol (MCP) servers — scan AI agent configs for tool poisoning, rug pulls, hardcoded secrets, command injection & supply-chain risks. Pure Python, SARIF + CI…
A trellis Provides infinitely nestable, scalable layered panel layouts for web apps that keep every view alive, addressing window-management pain in IDE and dashboard applications. 136 stars in 1.7 days is not explosive, but the author has community influence, and adoption by a mature frontend ecosystem could make it a standard layout dependency; the risk is high paradigm-migration cost and Figma/Linear-style panels already locked into in-house systems.
Fractal layouts for web apps: nest panels to any depth, zoom to the part you need, and keep every view alive.
A Show HN: Building a Markdown editor for Mac, iOS and web A local-first Markdown editor across Mac, iOS and Web with in-place rendering, draggable tables and WebRTC peer-to-peer sync where the server only sees a random room ID, earning 62 points for 10 months of product polish. Free plus optional sync subscription is a clear path, but the editor space is extremely crowded with Bear, iA and Obsidian, so it must target a niche through typography and private sync.
I've spent the last 10 months building Beauty: a local, powerful greco-roman Markdown editor where the page looks finished while you type. It runs on the Mac, iPhone and in the browser, and the browse…
A beamer A small service that lets one keyboard and mouse control both Mac and Windows on a LAN with bidirectional switching, clearly described and easy to set up. The problem is that Deskflow and Synergy already do this thoroughly, with no explanation of what gap it fills.
One keyboard and mouse for a Mac and a Windows PC over the local network, in either direction.
A ppcosxkvm Runs PowerPC Mac OS X Tiger on Apple Silicon and uses Metal to emulate a Radeon 9700 PRO for Quartz Extreme and OpenGL. This is real technical work worth a look for retro Mac and emulator fans; the limitation is scripted setup and results that are hard to reproduce or verify.
PowerPC Mac OS X Tiger on Apple Silicon with an emulated ATI Radeon 9700 PRO: Quartz Extreme, Core Image and OpenGL via Metal
A polyxor A 128-bit general-purpose hash for file and data-stream integrity checks, with a Lean formalization, from the maintainer of libraries like fastrand, so credibility is decent. It's useful for fast checksums; the "cryptographic strength" claim still needs independent review.
A 128-bit universal hash function designed for fast file/data stream integrity checks with cryptographic strength, written in Rust.
A agentglass Browses, follows and takes over sessions of coding agents like Claude Code and Codex in the terminal, giving people running several agents a single entry point. The scope is restrained and the description not oversold; several viewers already exist in this space, and success depends on truly taking over rather than just watching.
A looking glass for your coding agents: browse, watch, drill into and steer Claude Code, Codex & fx sessions from one tiny native TUI
A mhi2_altscreen_carplay Open-source research on projecting CarPlay onto Volkswagen Group MHI2 head-unit virtual clusters, starting with Škoda and later expanding to SEAT/VW. The target and stage are stated modestly, a rare new problem in the head-unit modding scene; the limitation is it's pure research with no install steps and a very narrow audience.
Open research for CarPlay AltScreen/Auxiliary Screen navigation in MQB Virtual Cockpit on VW Group MHI2 — Škoda first, SEAT/VW planned.
A cartoon-terrain-studio A Godot editor plugin for 2.5D cartoon terrain generation, aimed at indie developers who want to quickly build stylized scenes; such tools are genuinely scarce in the Godot ecosystem. The limitation is that the repo is brand new, with no example project or export instructions yet.
Godot Tool to create 2,5D stylized cartoon terrain.
A RUG An ABNF grammar to LALR(1) parse table generator and runtime in C++98, with a narrow goal: usable on old compiler environments. It's appealing for adding DSL parsing to legacy codebases; the limitation is an empty ecosystem, with docs and examples you must supply yourself.
RUG — Rug Understands Grammar — is a C++98 parser generator and runtime library that compiles an ABNF grammar into a packed LALR(1) parsing table.
A hermes A local CLI that packages unfinished changes on disk into context and hands them to the next coding agent, supporting Claude, Cursor and Codex with no plugins or cloud. Multi-agent switching has a real handoff pain point and the idea is clean, but it's just a packaging convention whose effect depends on how each agent reads it.
Local-first handoff CLI that packs unfinished work on your disk and hands it to the next coding agent: Claude, Cursor, Codex, Antigravity, and more. No plugin. No cloud.
A formula1 An explodable, disassemblable 3D F1 car built with three.js to explain 2026 season technical regulations, useful for fans and technical visualization folks alike. The demo is complete; the limitation is that content is locked to one year's rules.
Interactive, explodable 3D F1 car explaining the 2026 FIA technical regulations (three.js)
A gutcheck "Semantic grep": ask questions directly about lines in logs, diffs, or JSON and filter instantly, running locally on CPU as a single Rust binary. It's far more restrained than dumping whole logs into an LLM, good for debugging production issues; the cost is accepting a small model's comprehension limits.
grep for meaning: ask a question about lines, logs, diffs or JSON and explore the answers live. Runs locally on CPU, one Rust binary.
A X4_VR A mod adding true stereoscopic VR and 6DOF head tracking to X4: Foundations via OpenVR, working with any SteamVR headset; space sims rarely get proper VR, so the need is clear. The limits are its X4-only audience and tuning-dependent results.
A mod that adds true stereoscopic VR with 6DOF head tracking to X4: Foundations, on any SteamVR (OpenVR) headset
A portkill A macOS menu bar tool that lists local listening ports, shows which project each port belongs to, and closes them with one click, filling the gap of not remembering lsof flags. Small scope, honest description, install and use; several similar GUIs exist, and the difference is how accurately it identifies project ownership.
Native macOS menu bar app that lists local listening ports, shows which project owns them, and stops them in one click.
A ilio-mini Turns Codex into small infrastructure for solving reverse-engineering CTF challenges; the author placed third in Flare-On, showing this is a real competition setup rather than wishful thinking. Great for binary analysis people and those building agent harnesses. The limitation is single-digit stars, and whether others' environments can be replicated is unverified.
A tiny infrastructure for Codex to solve CTF challenges in reverse engineering. 3rd in Flare-On 13.
A midden A Rust JVM heap dump analyzer that produces retained sizes, holder chains, and leak suspects directly from .hprof files as a single static binary without a JDK, lighter than Eclipse MAT. Practical for backend engineers chasing Java memory leaks, though its analysis depth still needs comparison with MAT and JXray.
Fast JVM heap dump analyser: retained sizes, owners and leak suspects from .hprof files. One static binary, no JDK needed.
A loopvec A compiler-side tool for SIMD auto-vectorization of Go loops, with a single clear goal that fills a long-missing piece in Go. Worth a look for how it hooks into SSA and lowers to assembly, though at one day old and 10 stars, loop coverage is unknown.
SIMD autovectorizer for Go loops
A ssi263-speech It emulates the SSI-263 speech chip alongside original firmware for the Speak-Out, Braille Lite, and Accent lines, packaged as an NVDA add-on. Scoped and accurately described, it's a genuinely usable tool for retro speech hardware and accessibility users; 10 stars in a day shows how small that circle is, so few others benefit directly.
The SSI-263 speech chip, emulated, with the original firmware of the Speak-Out, Braille Lite, Accent-mini and Accent SA driving it. NVDA add-ons.
A interference-search Lets a model reason in parallel across multiple explicit state branches, merging duplicates, pruning dead ends and advancing together, solving agent search divergence; 93 stars in 0.4 days is a strong signal. It has a path to becoming the standard layer for agent reasoning, with the risk of overlapping Tree-of-Thoughts implementations and mainstream agent frameworks, so it needs to ship benchmark data soon.
Reason over explicit states, many branches at once: merge duplicates, cancel dead ends, advance together.
A phantomat Infinite-canvas window management for Hyprland: windows become zoomable, pannable layers rather than tiles in a grid, a new way to organize research, writing, and multi-screen comparison. 158 stars in 1.7 days shows real demand; the limits are that it only serves Hyprland users and stability is unverified.
Phantomat: a zoomable, infinite-canvas window manager for Hyprland
A godot-lunastream Lets Godot 4's VideoStreamPlayer play RTSP and http streams directly, with FFmpeg demuxing plus hardware decode, GPU zero-copy, and no external runtime, filling a long-missing engine gap. Godot developers doing live casting or surveillance feeds can use it immediately; the risk is unverified GDExtension prebuilt coverage and stream protocol boundaries.
Godot 4 GDExtension: play rtsp:// / http(s):// / local files through VideoStreamPlayer — FFmpeg demux, hardware decode, GPU zero-copy, no external runtime.
A servery.nvim A small Neovim plugin that uses an existing fuzzy finder to jump between multiple server sessions, doing only that with an honest description. Immediately installable convenience for people running several nvim instances, but several session-switching plugins already exist, the scope is narrow, and it's hardly a new approach.
Jump between Neovim sessions using your favourite fuzzy finder
A open-beni It brings robot racing to the browser: MuJoCo compiled to WASM, ONNX policies running online, plus training source, more complete than most robot demos. Suited to those trying reinforcement learning and physics simulation cheaply, though it's a toy and far from a real robot stack.
OpenBeni browser robot racing with MuJoCo WebAssembly, ONNX policies and training source
A cdj-nxs2-qemu It runs Pioneer CDJ-2000NXS2 deck firmware under QEMU, a rarely touched corner of hardware emulation, interesting to those reverse-engineering DJ gear or wanting to test without the physical unit. The limit is that it's purely technical experimentation, and it's unclear whether a full playback chain can complete.
CDJ-2000NXS2 QEMU Emulator
A cida A small always-on macOS tool that calls a chosen LLM to translate and polish text in any input field; scoped, accurately described, install-and-go, from an author with a track record in open source. Suited to people who write English often, though similar Raycast plugins already cover most of the ground.
Cida: translate and polish text anywhere on your Mac using the LLM of your choice. Translate and polish text anywhere on your Mac.
A Moon-FruityLoops26-FileParser Parses FL Studio's binary FLP project files in C++, enabling procedural generation, batch project edits, cross-DAW migration and AI composition training sets; 195 stars in 0.1 days shows the music toolchain long lacked such a low-level library, with potential as a format middleware via open core plus commercial SDK. Risks are format changes across versions and vendor legal pressure.
FLP binary parser for FL Studio project files
A Show HN: PlaceCall (YC W26) – agentic API to call businesses and get things done An API that gives agents phone calls: natural-language tasks, automatic IVR navigation and waiting on hold, returning structured results, recordings and transcripts to fix stale store info. YC W26, per-call pricing and 250 free tries are a standard API model; risks are voice infrastructure like Twilio moving down-stack and compliance around placing calls on users' behalf.
Hi HN, this is Yarik and Vlad from VOYGR - we are building the tools for agents and apps to engage with local businesses. It all started with our own pain point at VOYGR: calling businesses to verify …
A tpu-megakernels Hand-fuses multiple TPU operators into a megakernel, targeting utilization loss from XLA compilation startup overhead and scheduling gaps, a scarce TPU performance skill. Inferact as an inference provider could package kernels as hosted inference or license them to cloud vendors, but Google's official Pallas and compiler keep absorbing such optimizations.
A collection of megakernels for TPU
A Show HN: Canary (YC) – Independent verification for AI code Independent verification for AI-generated code: takes a changeset and intent, then uses a swarm of agents in a remote sandbox to verify runtime behavior and security invariants, filling the gap between static review and unit tests. YC backing plus a per-repo CLI is a clear enterprise path. The risk is huge: Greptile, CodeRabbit and Qodo already hold the track, and only 5 HN points shows the narrative did not land.
Hey HN, we are Aakash and Viswesh and we are building Canary ( https://www.runcanary.ai/) - independent verification for AI code. Claude/Codex calls Canary with the changesets, intended behaviour and …
A wetype-ime-linux Ports the WeChat input method's Android engine to Linux and integrates Fcitx5, filling the long-weak Chinese input on Linux; 156 stars in 1.2 days shows Chinese Linux users really lack this, with a chance to be preinstalled by distros and become a standard dependency. Risks are closed-source engine licensing and upstream version tracking, with almost no direct charging.
Linux port and Fcitx 5 integration for the WeType Android input engine
A binaural-voice Renders TTS voices with KU100 near-field HRIRs so they sound close to the ear and moving around the listener, with position labels, directly useful for spatial audio and voice demos; the scope is small and specific, more honest than big voice platforms, though listening validation is hard and it depends on specific HRIR data.
Make TTS voices speak right by your ear and walk around your head: KU100 near-field HRIR binaural rendering + position labels.
A ipadScreen It uses a jailbroken old iPad (tested on iPad 2 with iOS 6) as a Mac extended display, sending JPEG over USB/Wi-Fi with NEON decoding at about 35fps; the scope is small and the metrics concrete. It meets a real need for people hoarding old devices, but depends on jailbreak and old systems, and the experience gap versus Sidecar is unstated.
Use an old jailbroken iPad (iOS 6+, tested on iPad 2) as a real extended display for your Mac, over USB or Wi-Fi. Native iOS 6 app with libjpeg-turbo NEON, ~35 fps.
A morph-ui Copyable animation components built on React 19 ViewTransition without pulling in an animation library, aimed at React developers wanting quick transition effects. The scope is small and clear, with code as documentation; the limitation is a shallow ecosystem and unknown component count.
Animated React components you copy and paste, built on the React's ViewTransition. No animation library, just React 19.3+ and a little CSS.
A buried-injections Uses 629 AgentDojo injection samples to show regex and Prompt Guard 2 detect almost nothing in tool outputs, with reproducible scripts. Direct evidence for agent security practitioners, more useful than talk of defenses; but it covers only prompt injection and conclusions depend on the sample set.
🛡️ Regex catches 0%, Meta's Prompt Guard 2 catches 1% of 629 realistic AgentDojo injection attacks when they're buried in tool output. Reproducible benchmark.
A bitwright A Rust bit-vector expression simplification library for binary analysis and deobfuscation, filling an open-source gap beyond SiMBA. Its scope is restrained and its purpose stated in one line, so it can plug into reverse-engineering and vulnerability toolchains; the limitation is that it's brand new, with no docs or benchmarks.
Bit-vector expression simplifier for binary analysis and deobfuscation
A cleanupper Solves the daily pain of macOS developers' disks being eaten by Xcode DerivedData, node_modules and Homebrew/pip caches, terminal-native, trash-first and zero telemetry, more trustworthy than CleanMyMac, and naturally distributable via Homebrew and social word of mouth. Commercialization could be a Pro rule library or team device cleanup, with the risk that Apple's built-in storage management and open-source peers like Mole already hold the space, and the anomalous 700+ stars in three days needs watching.
🧹 Free, open-source macOS disk cleanup CLI. Scan & safely remove caches, logs, Xcode DerivedData, npm/Homebrew/pip junk and stale node_modules from your terminal. Trash-first, zero telemetry the termi…
A Show HN: I built a post-mortem debugger for native Windows x64/x86 crashes Post-mortem debugging for Windows x64/x86 crashes, moving WinDbg's analysis and cross-correlation into a modern GUI and using MCP so agents reason over structured data with fewer hallucinations—hitting a long-standing gap in native Windows troubleshooting. 20 points on HN and beta licenses already sold show willingness to pay. Competitors are WinDbg, Visual Studio and TTD; the risk is a closed-source small team struggling to shake the Microsoft ecosystem.
Hello HN! I've spent years debugging Windows crashes with tools that were either friendly but limited (e.g. Visual Studio) or powerful but archaic (e.g. WinDbg). I developed patterns and methods for u…
A vellum.nvim Renders a GitHub-style Markdown preview beside Neovim buffers and draws mermaid diagrams as real images in the terminal. nvim users writing docs are missing exactly this, and it adds mermaid over markdown-preview.nvim; stability remains to be verified.
Live, beautiful GitHub-markdown preview beside your buffer in Neovim — mermaid diagrams as real images, in the terminal.
A magpie This menu-bar tool unifies model backends across agents (Codex on DeepSeek, Claude Code on Kimi), solving multi-subscription, multi-key switching and vendor lock-in pain. Author yetone has developer influence and it started fast; it could add team config hosting and gateway value, but faces established routing layers like LiteLLM and claude-code-router and must differentiate on experience.
Every agent's model. One place. Codex on DeepSeek, Claude Code on Kimi, from the menu bar.
A cc-usage It aggregates quota, tokens and spending from coding assistants like Claude and Codex into a cross-platform desktop app, an immediately useful need for heavy users. The scope is clear; the hard part is keeping up with each platform's API, and several similar CLI tools already exist, so it's a race on maintenance speed.
A cross-platform desktop app for monitoring AI coding assistant usage, giving a unified view of quotas, tokens, costs, and request logs across Claude, Codex, and other platforms.
A grit A Godot GDScript compiler written in C++, aimed squarely at the performance limits of interpreted GDScript; the direction is rare and the target concrete. It is still an early implementation, so do not expect to drop it into production.
GDScript Compiler for Godot.
A shapeshift A single input box that morphs by semantics into the right control—date picker, slider, selector—an experiment in form and natural-language interaction paradigms that, if made solid, could become a general input component for agent UIs and be licensed as a design system or SDK. Risks: heavy dependence on the undisclosed Jev model, offline promises and recognition accuracy, plus a strong gimmick factor, and mainstream design systems may simply build it in.
An input that becomes what you mean: one text box that morphs into the right UI as you type. Powered by TypeSafe Jev, works offline.
A oink A Rust text-adventure (gamebook) engine for e-ink devices, small in scope and clearly defined, with rendering and page turns evidently designed around e-ink traits. It suits developers wanting to write interactive fiction on Kindle-like devices; for now it lacks sample books and docs, so watch whether it can run a full work.
A librogame (gamebook) engine for e-ink devices.
A Show HN: RxFilm Studio–Create and edit your product videos with AI agent This hands the entire product marketing video pipeline—script, editing, effects, voiceover, subtitles—to AI agents, targeting indie developers and small teams producing ad assets, a real need with budget. Monetization via subscription and credits is clear, but Runway, Descript, InVideo and Sora-like tools are extremely dense, making differentiation and asset copyright the biggest risks.
Hi HN, I built RxFilm Studio, a video workspace that lets you create and revise videos with AI agent. The reason I started this project is simple: I’ve noticed that many independent developers and sma…
A PianoStudio A browser piano workbench that turns sheet-music images into playable sequences, with real samples and an optional OMR pipeline. The barrier to trying it is low; OMR accuracy and sheet-music copyright are its ceiling.
Piano Studio is a responsive browser-based piano workspace that turns a sheet-music image into a playable sequence. It combines a polished virtual piano interface, real acoustic piano samples, local f…
A ha-layers Adds a priority layer to Home Assistant lighting, solving the old problem of scenes and manual adjustments overriding each other, with simple installation and clear scope. Immediately useful for heavy HA users, though it's just scheduling logic and differs little from existing blueprint approaches.
Priority layers for Home Assistant lights
A sentinel It turns anti-money-laundering transaction monitoring into a NestJS service plus a separately installable rules engine, with idempotent ingestion, case management and audit logs. Open-source compliance tools are rare and useful to fintech backends; but whether the rules match real regulatory practice can't be proven by the README, so check the tests first.
Open-source AML transaction monitoring: a NestJS service and a framework-agnostic rules engine on npm (@sentinel-aml). Idempotent ingestion, case management, audit log, runtime rule tuning.
A myDATA A Claude skill that issues Greek AADE e-invoices with MARK, saves a JSON per invoice, and syncs to ERP — an extremely narrow niche but a real compliance workflow. Teams doing business in Greece and connecting to SAP B1 can use it directly; the limitation is ERP support for only one vendor, with the rest left to the community.
Issue Greek invoices via AADE timologio, get the MARK, keep one JSON per invoice, sync to your ERP (SAP B1 today; SoftOne, Pylon, Megasoft wanted). Claude skill, not a ΥΠΑΗΕΣ.
A liteagents A provider-agnostic agent SDK from BerriAI that follows the Claude Agent SDK's query() interface and can switch models per turn; it's a low-cost abstraction for teams already on LiteLLM, with the limitation that the interface tracks upstream and the ecosystem is still shallow.
A provider-independent agent SDK with the same query() interface as the Claude Agent SDK, allowing you to use the right model for every turn.
A agent-message-board It lets multiple coding agents share context through a SQLite message board, with humans watching via a read-only web page; the problem is real, the implementation restrained, and how to run it is obvious at a glance. The limit is that it's only a collaboration patch competing with MCP and shared workspaces in an undecided ecosystem.
Shared SQLite message board for AI coding agents, with a read-only human web UI
A talking-head-reel A Claude Code skill that turns casual talking-head phone videos into vertical clips, including selection, subtitles, and overlays — a concrete niche with real demand. 40 stars in a day shows the pain point exists, but end-to-end editing like this is sensitive to ffmpeg and model calls, and results are unverified.
Claude Code skill: turn a messy phone recording into an edited vertical reel (best takes, captions, animated overlays, memes)
A NiloPathTracing Adds RTX path tracing and DLSS Ray Reconstruction to Unity URP with one-click setup and NiloToon compatibility; Unity creators doing toon rendering benefit directly, with the limitation of being tied to RTX GPUs and the URP pipeline.
RTX Path Tracing + DLSS Ray Reconstruction for Unity URP. One-click setup, native NiloToon support.
A nafath-php It wraps Saudi Nafath identity verification's web redirect and app push into a PHP library and documents pitfalls the official docs miss, saving teams doing Middle East logins a few days. The scope is narrow and the description restrained; the downside is serving a single flow in one country, with long-term upkeep on the author.
Nafath identity verification for PHP — web redirect and app-push, with the traps the official docs leave out already handled.
A OpenWire A native Linux soundboard and effects processor on PipeWire, filling exactly the gap left by Windows-only tools like Soundpad; streamers and voice-chat users can install and use it, with the limitation that it only covers PipeWire desktops.
Native Linux soundpad and effects processor for PipeWire.
A expo-motion-blur Native motion blur for React Native views, implemented on both iOS and Android, a gap the RN ecosystem has long had. 18 stars in half a day shows people were waiting; worth checking on a real device, though it does only this one thing and nothing else.
💨 Native motion blur for React Native views on iOS and Android
A SafariKeePassXC Uses KeePassXC's own browser integration protocol to fill the gap Safari officially lacks, directly benefiting people who stick with Safari but won't give up a self-hosted password vault. Swift-only and just getting started, so you'll have to verify stability yourself, but the direction is right and the scope restrained.
Safari Web Extension that connects to KeePassXC via its browser integration protocol
A ImageForge It analyzes, patches, repackages and verifies Android images entirely in the browser, with no upload and no flashing, and the boundaries are clearly stated. Those modifying boot images can use it as a lightweight alternative, though it only handles the image files themselves.
Universal Android image patcher that runs entirely in your browser: analyze, patch, repack, verify and export. No server, no upload, and it never flashes a device.
A Show HN: Drop – A rootless Linux sandbox with gVisor support A rootless Linux sandbox using namespaces plus gVisor to give locally run untrusted dependencies and third-party programs a disposable isolated home, solving the pain of one package install compromising the whole machine; 153 points on HN shows strong developer resonance, turning around after two unnoticed launches thanks to gVisor support. Could serve as the execution base for AI-generated code. Risks: firejail/bubblewrap and containers already occupy the space, and it's Linux-only.
I created Drop because I always felt uneasy installing and running third-party programs using my main user account. A single compromised dependency means a full compromise of the system. What is even …
A agent-jev A 0.6B System 1 decision model that maps states like diffs, traces and logs into calibrated probabilities in one forward pass, returning results in 50ms with zero decoding tokens and saving agent inference cost; 202 stars in 1.3 days, the asset in this family closest to a model proper, with potential as a standard agent dependency. Risks: confusing naming among peers and metrics lacking third-party validation.
AgentJev-0.6B - a fast 'System One' decision model for AI Agents: feed it any unstructured state (diffs, traces, logs) and structured questions, get calibrated probability distributions back in one ~5…
A Show HN: InstinctFlash – High-Performance Serving Runtime for Robotics Models An inference acceleration runtime for robot VLA and world models that cuts diffusion sampling from 25/50 steps to 2/4, directly addressing control-loop latency, a deployment bottleneck; backed by Jetson Thor measurements and Samsung/Siemens usage, though HN gave 23 points and GitHub only 42 stars, so momentum is just starting. Could become the vLLM of robotics. Risks: AGPL and pressure from NVIDIA's Isaac/TensorRT ecosystem.
Hey HN, Guanming here, cofounder of General Instinct. We just released InstinctFlash, a high-performance serving framework for robotics models. It’s licensed under AGPL-3.0. On Jetson Thor, we see spe…
A Show HN: AI·rete·RAG – a Rete rule engine decides, RAG explains why Uses a Rete rule engine to issue verdicts while the LLM only explains with cited policies, solving decisions that must be auditable like credit risk and clinical use, with Verdict reproducibility as a real selling point; 34 points on HN and an MCP client show early developer experimentation. The monetization path is clear (compliance audit SaaS). Risks: a closed-source engine squeezed by Drools/IBM ODM and built-in model guardrails.
Hi HN, I built ai·rete·rag because I kept seeing teams put an LLM in charge of decisions that need to be auditable (lending, fraud, clinical triage), then bolt on "guardrails" after the fact. It runs …
A Show HN: Training a model to identify AI web content from structure alone Judges whether a web page is AI-generated by structure rather than words, with 98% accuracy that rewriting struggles to evade, addressing marketing teams' need to detect AI slop; 37 points on HN and YC backing show commercial traction, already embedded in the Sitefire product. The path is a content-quality API. Risks: an arms race between detection and evasion, and human samples limited to 2020–2022.
Hey HN! We’re Vincent and Jochen from Sitefire ( https://sitefire.ai). We have been working together for years, with backgrounds in RL/optimization at Stanford and software engineering from Technical …
A qwenimage-ncnn-vulkan Runs Qwen-Image on ncnn + Vulkan, by nihui of waifu2x-ncnn-vulkan fame, addressing the real need of local image generation without a discrete GPU. The repo has no description yet, so you need to read the code to confirm supported models and VRAM use; compare with local inference options like stable-diffusion.cpp.
A FluidUse This does local computer use on Apple silicon, using the Accessibility API to fill forms and a local small model for typed decisions, addressing cloud CUA privacy and latency concerns; two-day traction shows local-privacy backers exist. It could sell enterprise intranet automation, but risks Apple tightening permissions and being crushed by official computer use and Apple's own capabilities.
Local computer use on Apple silicon: on-device typed decisions with laya and form filling with CUA-S1-FORMS, driven through the macOS Accessibility API
A MCPBastion Adds a zero-trust gateway between AI agents and MCP servers: capability allowlists, redaction, rate and cost caps, auditable session logs. As MCP spreads, runaway permissions are a real problem and such gateways are still rare; only a day old, whether it truly blocks privilege escalation depends on the implementation.
Zero-trust security gateway between AI agents and MCP servers: scoped capabilities, redaction, rate and cost limits, and verifiable session logs.
A Mac-Arch-Installer Shell automation that installs Arch Linux ARM under Apple Silicon plus Parallels, neatly sidestepping the official ISO being x86_64-only and saving an evening of fiddling. Scope is restrained; the limitation is it only covers Parallels virtualization.
A lazy pack for installing Arch Linux in a Mac VM | automatically installs Arch Linux ARM on Apple Silicon + Parallels Desktop (the official ISO is x86_64 only and cannot boot directly)
A YuE2-Turbo Concurrent inference optimization for the open-source music generation model YuE2, reporting concrete numbers of 1.68× per song and 3.31× tracks per GPU with the model and recipe unchanged. Directly usable by people running music generation, though it only serves YuE2.
Fast, concurrent inference for YuE2. Same model and recipe: 1.68× faster per song, 3.31× more songs per GPU.
A tenanting A Rails multi-tenancy library from the GoRails author that solves one thing, explained in one sentence, and works once installed. The concept is not new and mature options like acts_as_tenant exist; what is worth seeing is how it handles tenant isolation with a smaller interface.
Multitenancy for Rails
A glance A new yoheinakajima project: run open-source vision models locally to ask typed questions about images and get probabilities rather than prose. Clean idea with clear boundaries for people using VLMs as classifiers; limited to small local models, with accuracy depending on the chosen weights.
Ask an open vision-language model typed questions about an image and get probabilities back, on your own machine.
A dev-0.4b Turns routing, judging, and scoring into a single forward pass of a 0.4B bidirectional model that outputs calibrated confidence; useful for teams wanting to replace generative judging in agents. The description is restrained and the architecture concrete, but it doesn't say whether weights are released, and similar repos have clustered in the past two days, so worth a look before deciding.
Dev (dev-0.4b): a 0.4B bidirectional decision model on ModernBERT-large for routing, yes/no verification, and 1-5 rating. Non-generative single forward pass with calibrated confidence.
A wled-toy A real-time shader and node-graph editor for WLED strips, turning lighting effects from hand-written JSON presets into instant node previews. Anyone playing with LED hardware will want to try it, since before there were only scattered web effect presets; the limitation is it's locked to the WLED ecosystem and just released.
Live shader and node graph editor for WLED Strips
A bookcard.koplugin A small plugin that changes the sleep screen in KOReader, with the author openly crediting CrossPoint as inspiration and keeping the scope tight. A ten-minute install-and-use project for e-reader users; the limits are obvious too, serving only the KOReader ecosystem and just swapping a standby image.
Book Card is a new sleep screen for KOReader, heavily inspired by the idea and visual design of CrossPoint's sleep screen implementation.
A arc-cua Provides a faster action-execution layer for computer-use agents, bypassing the slow loop of screenshot-then-reason at each step and directly targeting the latency bottleneck of GUI agents. 125 stars show CUA teams are looking for infrastructure, with potential to become the default action runtime; the risk is OpenAI and Anthropic's built-in toolchains absorbing this layer.
Superfast action layer for computer-use agents
A OpenDLSS-NR Rewrites NVIDIA's neural rendering super-resolution network in Vulkan and claims bit-exact parity with the original, addressing the unavailability of super-resolution on Linux and non-NVIDIA GPUs. 70 stars is few, but if it lands in Mesa or Proton it could become the open-source standard implementation; the biggest risk is patent litigation, and bit-exact parity needs reproduction.
A Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, bit-exact against the original.
A briefd A self-hosted context compiler that packs team docs from git into MCP bundles under a token budget for coding agents. It targets the real agent-context problem, but only serves teams already using MCP.
Your agents, briefed. Not flooded. Self-hosted context compiler: git-backed team knowledge served to coding agents as token-budgeted bundles over MCP.
A Show HN: Mini-AGI – Dynamic continual learning model trained on 8GB VRAM Uses an expandable-expert MoE with single-stream batch-1 training to squeeze continual learning of large models into 8GB of VRAM, directly attacking the inability to train large models on consumer hardware. 248 points is the strongest heat these two days, but weights aren't released and only scaling curves exist—a research-grade project with no sellable product short-term, more likely to be absorbed by frameworks.
Sorry for the pretentious name, I know, I know.. It just contains all the pieces I would like to see a AGI model to have, and I can't stand the temptation. Before throwing rocks at me, please take a g…
A nyra-forced-aligner A forced aligner for natural speech that outputs word- and phoneme-level timestamps, aimed at real speech with pauses and restarts. It fits voiceover and clinical speech better than general tools, but lacks public evals and its ecosystem is very new.
Word and phoneme level speech timing for spontanoues speech. Fast, accurate and robust.
A renfecli An unofficial command-line client for Renfe timetables and fares that also serves agent calls. Spanish commuters would use it immediately, and the author is upfront about it being unofficial; the niche is small and maintenance rests on one person if the API changes.
CLI no oficial de Renfe, pensada también para agentes — horarios y tarifas reales de tren desde la terminal
A dewarr Self-hosted audiobook discovery and auto-download that pulls Goodreads/Hardcover lists and integrates with Audiobookshelf, filling a gap in the *arr ecosystem. It depends on third-party list sources, and copyright for downloads is on you.
Self-hosted audiobook discovery, Goodreads and Hardcover lists, Audiobookshelf integration, and automatic downloads.
A atlas-vtt Brings battle maps, tokens, fog of war, dice and initiative into Obsidian, with a separate player view. A ready-made need for people taking RPG notes in Obsidian, and the scope is restrained; limited to single-table offline use, with Foundry VTT as the alternative.
Offline virtual tabletop for Obsidian: battle maps, tokens, fog of war, dice, initiative and a separate player view.
A omarchy-icloud-photos Browse iCloud Photos in a native window on the Omarchy desktop, synced via icloudpd, with day-by-day browsing and undoable deletes. Scope is restrained and docs honest; it suits Arch desktop users tied to iCloud, but functionality is narrow.
Your iCloud Photos library as a native window on Omarchy. Sync with icloudpd, browse by day, delete with undo.
A video-editor A rough-cut Skill for voiceover that automatically finds pauses, slips, and restarts, then exports with micro-fades to avoid pops. It can save voiceover creators hours, but relies on an LLM judging each segment, so quality and cost remain unverified.
Smart general-purpose video rough-cut and slip-of-the-tongue/blooper removal Skill: automatically detects speech pauses and silence, uses a large model with concurrent sliding windows to identify slip…
A job-filter-assistant A job-filtering extension for BOSS Zhipin and 51job that passively scrapes, matches each posting against an AI profile, and exports results, keeping data local. It saves effort for mass applicants, but breaks whenever job sites change their pages.
Chrome extension (MV3) for job filtering on BOSS直聘 / 前程无忧: passively captures job listings, AI profiling and per-job match analysis, multi-protocol AI integration, batch analysis and export. Data is s…
A Cove A live macOS desktop scene built with Three.js: an aquarium you can fish in and a rabbit that chases the mouse, for people who want their screen to feel alive. The description is honest and it is a clean little toy; the limit is that it is purely decorative, closest to Shimeji-style desktop pets.
A living scene on the Mac desktop: an aquarium where you can fish, and a lop-eared rabbit that chases the mouse broom and bites it. Open source, macOS, Three.js.
A stellaris-galaxy-forge Directly edits the galaxy map in Stellaris saves: move systems, draw hyperlanes, change nebulae and spawn points, with byte-level edits plus backup and undo. Immediately usable for map editors, with clear boundaries; it only breaks when the save format changes.
Galaxy map editor for Stellaris saves and static galaxy scenarios: move systems, draw and cut hyperlanes, shape nebulae, place initializers and spawn points. Byte-level edits, backups, undo.
A ZCode Z.ai's officially open-sourced coding agent harness wraps its own coding model into a ready-to-use terminal agent, competing for the entry position of Claude Code/opencode. Over a thousand stars in 0.6 days comes from company channels and coding-agent heat. The commercial path is clear: funnel to its own API and subscriptions. Risks: severe homogenization, and harnesses easily swallowed by native model capabilities.
Z.ai's coding agent harness. Powerful, intelligent, extensible.
A WinStadia Makes an idle Google Stadia controller appear as an Xbox controller on Windows, working over both USB and Bluetooth. A real need for anyone with one lying around, narrowly scoped and honestly described; limits are covering only this hardware, with DS4Windows still the general option.
Use Google Stadia controllers as Xbox controllers on Windows, over USB and Bluetooth.
A opencloak Replaces names, emails and other personal information before prompts are sent, with local detection, no server and no account. Useful for people feeding real tickets and emails to models; whether the redaction rules are complete cannot be verified from the description, and it competes directly with existing PII libraries.
Swap the personal details out of your AI prompts before they are sent. On-device detection, no server, no account.
A tmon A one-screen terminal dashboard answering whether a home server is healthy right now, with CPU, memory, RAID, SMART, UPS, Docker, backups, and reverse-proxy requests all in one place. NAS users no longer need to stitch together btop and glances; the risk is breadth over depth, since each item may not match dedicated tools.
A full-screen terminal status monitor for home servers and NAS boxes. One screen answers "is everything OK?" and shows what's going on: CPU, memory, storage and RAID, disk SMART, UPS, Docker services,…
A expo-face-vision Wraps Android ML Kit and iOS Vision face detection, head pose, and liveness checks into an Expo module, with a Skia overlay. React Native has long lacked this native glue for face login, and the platform split is clearly documented; but you must assess liveness accuracy and security boundaries yourself, not treat it as turnkey auth.
Native face detection, tracking, head pose, guidance, auto capture and liveness for React Native and Expo. Google ML Kit on Android, Apple Vision on iOS, Skia overlays.
A expo-infinite-media Paged media streaming for React Native/Expo: mixed video and images, player pools, preloading, and bounded caching, using AVFoundation and Media3 natively. RN teams recreating short-video feeds can use it directly; the cost is being tied to an Expo version, and complex lists still need your own tuning.
Native paged media feed for Expo and React Native CLI. Vertical or horizontal paging of mixed video and images, with pooled players, predictive preloading and bounded caching. AVFoundation on iOS, Med…
A job_search_tool Scrapes jobs from company ATS APIs and aggregators, deduplicates them, optionally scores with Claude, and reviews them in a local dashboard. Genuinely convenient for active job hunters with a clear pipeline; limits are that scraping quality depends on the site and scoring needs your own key.
Automated job search pipeline: fetches roles from company ATS APIs and aggregators, filters and dedupes them, optionally scores fit with Claude, and reviews results in a local web board.
A nimble Bespoke Labs extends its synthetic data pipeline into a combination of local typed decisions, contrastive data filtering and model evaluation, hitting the real pain of fine-tuning data quality; 421 stars in a day and a half plus an existing Curator user base support a hosted data and evaluation platform, with the risk that big players like Databricks push similar capabilities down-market.
Local typed decisions, contrastive data curation, and model evaluation.
A simple-jev Wraps any open-source model into a classifier or jev endpoint, capturing migration demand for the new typed decision paradigm, with 282 stars in a day and a half showing a clear gap in the ecosystem; it could charge for a gateway service or SDK, but the risk is that once Jev officially supports this natively, such thin wrappers get absorbed upstream.
Turn any open model into a classifier/jev endpoint
A crayfish A Compose Multiplatform image cropping component covering Android/iOS/desktop/Web from one codebase, by skydoves, so maintenance looks stable. You can add the dependency today; cropping isn't new, but full four-platform support is genuinely rare.
🦞 Compose Multiplatform image cropper for Android, iOS, desktop, and the web.
A glance-linux Face unlock for Linux with liveness detection: a PAM module plus daemon, also wired into the Omarchy lock screen indicator. Since Howdy there's been no decent replacement, and it states its boundaries clearly; the limits are hardware compatibility and security auditing, which need time.
Face unlock for Linux with real liveness detection: a PAM module, a daemon, and an Omarchy lock screen indicator
A OrcaBonsai-27B-Uncensored Runtime behavior ablation strips refusals from a ternary-quantized 27B model without changing weights or requantizing, addressing how compressed models' safety alignment gets baked into weights and becomes hard to adjust. 360 stars in 0.7 days came from the uncensored topic. It could monetize through OrcaRouter inference routing, but risks compliance and model license disputes, and many similar de-alignment tools already exist.
Runtime behavioral ablation for compressed LLMs. First target: Ternary Bonsai 2 27B — no weight modification or re-quantization. by OrcaRouter team
A Show HN: Scry, programmable internet search w/ congestion pricing Makes a 500TB web index queryable as read-only SQL/Datalog, allocating compute via congestion bidding so agents escape uncontrollable black-box search APIs. 42 votes show the search-controllability pain is real. It could charge per query as agent-facing search infrastructure. The biggest risk is search-engine-scale cost, Exa and Tavily already occupying the space, and data coverage being hard to match.
Meet Scry, a 500 TB NVMe internet index in ClickHouse that you can run ~arbitrary readonly SQL and some of Datalog over, and I handle the problem of resource-contention with congestion-based micro-auc…
A panasonic-doorphone-connect An unofficial bridge connecting Panasonic door intercoms to Home Assistant, virtualized as a Wi-Fi sub-unit for ringing, live video and two-way talk. It's hardware reverse engineering few would attempt, and it clearly states what it does and for whom. Limits are obvious: valuable only to owners of this intercom system, and dependent on undocumented protocols that firmware updates may break.
Unofficial Home Assistant bridge for Panasonic Doorphone Connect. Virtual Wi-Fi handset: ring, live video, two-way audio.
A ShizuStore An app store for the Shizuku ecosystem that pulls APKs straight from upstream sources and installs them; the author maintains the awesome-shizuku list, and the scope and boundaries are clearly explained. Android users already on Shizuku can use it right away; it's essentially an indexer, so its value depends on that list being maintained.
App store for Shizuku apps. Based on my awesome-shizuku list and installs APKs straight from their upstream sources
A snifftest A zero-dependency English prose linter that targets clichéd AI writing patterns, with countable rules and only one judgment handed to a model. It's a lightweight checker for writers and editors on deadline; the limits are subjective judgments and English-only coverage.
A prose linter that sniffs out AI writing tells. Zero dependencies, countable rules plus one judgment model.
A BC-250 Turns a retired ASRock crypto mining card into a living-room console running Bazzite and Steam, with docs, 3D-printed parts, source and troubleshooting notes all in one repo. It's specific enough for DIYers hunting mining cards; the limit is that it's tightly bound to this card, so it can be copied but not generalized.
Building a budget gaming console out of an ASRock BC-250 crypto mining card. It runs Bazzite and Steam. This is my documentation, 3D printing files, source code, reference links, and other notes for t…
A ShellDelphi Embeds a real PowerShell terminal into RAD Studio via ConPTY, letting Delphi developers run Claude Code, vim, and other TUI programs inside the IDE. Small in scope and honestly described, a plugin almost nobody builds for Delphi; the limit is it serves only Pascal users and maturity is unknown.
Terminal PowerShell dentro do RAD Studio, via ConPTY. Roda Claude Code, vim e qualquer TUI.
A study-reader Defines an open .study course format plus a KOReader plugin that turns any material into interactive lessons on an e-reader. Moving learning interaction to e-ink is a rare angle, and the open format is commendable. The risk is dependence on a generation pipeline and a nascent ecosystem; actual course quality depends on the examples.
Open .study course format + KOReader plugin: turn any material into interactive courses on your e-reader
A Show HN: MCPJam - the first testing & evaluations platform for MCP servers A testing and evaluation platform for MCP servers, addressing the inability to observe MCP services in agent clients or run regression tests; the founder comes from Asana and brings an open-source user base as momentum. The commercialization path is clear: per-seat CI integration pricing, with the risk of being squeezed by Anthropic's official Inspector and various free evaluation tools.
Prathmesh, CEO of MCPJam here. Users now start in ChatGPT, Claude, Cursor, and other AI clients. They reach your product through your MCP server. That means your users often aren’t in your product any…
A node-webgl Brings ANGLE's Metal/D3D11/Vulkan backends into Node, enabling offscreen WebGL and three.js rendering without a browser, usable today for server-side screenshots and visualization output. It adds WebGL2 and multiple backends over headless-gl, at the cost of needing your own native build environment.
Headless WebGL 1 & 2 for Node.js, powered by ANGLE (Metal, D3D11, Vulkan). Render any WebGL or three.js scene without a browser.
A Show HN: I built a new version of my fun spatial 3D online meeting app Turns remote meetings into a multiplayer game where you walk around a 3D space, addressing the tedium and lack of serendipity in video calls; rebuilt solo over years, its 92 points indicate real users. Commercialization means team subscriptions, with risk from competitors like Gather and the ongoing operating costs of 3D scenes.
Hi HN! flat.social is a fun spatial online meeting app for remote teams and communities. It's largely built like a web multiplayer game, and the 3D virtual spaces are highly customisable with elements…
A techo5-dot Flashes a second-gen Echo Dot with fully local firmware, removing Alexa, enabling all seven microphones, and connecting to Home Assistant as a voice satellite, with signed A/B updates and SSH. Hardware tinkerers will want to try it immediately, but it requires disassembly and flashing, and only covers this model.
Linux firmware for the Echo Dot (2nd gen): no Fire OS, no Alexa. A local Home Assistant voice satellite with all seven mics, Bluetooth speaker mode, signed A/B updates and SSH.
A idempotency-shield An Express middleware that caches responses by Idempotency-Key and replays duplicates, with memory or Redis storage, hitting the real pain of payment and order retries. Small in scope and accurately described, installable today; similar implementations exist, but this one works out of the box.
Express middleware that makes handlers safe to retry — replays the cached response for a duplicate Idempotency-Key instead of re-running it, with pluggable in-memory or Redis storage.
A dexterous-astra A simulation environment, zero-shot scripts and trained checkpoints for dexterous-hand pen spinning and Rubik's cube solving, aimed at embodied AI researchers, and such reproducible in-hand manipulation resources are scarce. The limits are binding to a specific simulation stack and compute, making near-term deployment hard.
Dexterous pen spinning and Rubik's cube: sim environments, zero-shot scripts, trained checkpoints
A zeptun A Zig tun2socks engine delivering TCP/UDP/ICMP, SOCKS5, io_uring and a C API in one package, a convincing performance-oriented alternative to Go/Rust implementations. Proxy and traffic-forwarding developers can compile and try it, though it needs the Zig toolchain and lacks a ready client.
High-performance tun2socks engine in Zig with TCP, UDP, ICMP, SOCKS5, io_uring, and C API.
A techo5-spot Local firmware from the same author for the first-gen Echo Spot, keeping the round touchscreen, camera, Bluetooth and calling, also connected to Home Assistant. The engineering details are clear, but the approach matches the previous device and covers only one discontinued piece of hardware.
Linux firmware for the Echo Spot (1st gen): no Fire OS, no Alexa. A local Home Assistant voice satellite with a round touch face, camera, Bluetooth, phone calls and signed A/B updates.
A wesp-research Reverse-engineering documentation and tools for Microsoft's Windows Endpoint Security Platform, by a well-known figure in EDR. Directly useful to people building antivirus, EDR, or attack-defense research; the limits are that the platform is very new, the audience is narrow, and you must verify it yourself.
Technical documentation, reverse engineering research, and tools for the Windows Endpoint Security Platform (WESP).
A jevmlx Constrained decoding with MLX on Apple Silicon, producing schema-conforming JSON in a single forward pass and skipping retries and post-processing. Practical for Mac developers running small local models needing structured output; nested schemas and streaming edge cases aren't yet addressed.
Jev-style parallel constrained decisions for any MLX model on Apple Silicon. Typed, schema-valid JSON in one forward pass.
A sf4enhanced Adds rollback netcode to Steam's Super Street Fighter 4, a real need in the fighting game community, and similar changes have been proven on other old fighters. Currently the repo only hosts downloads with no source or implementation notes, so how far it goes depends on the author disclosing details.
Rollback netcode for Ultra Street Fighter IV (Steam). Downloads only.
A Show HN: I built a router for agent tools OpenRouter for agent tools routes and calls tools from different vendors through one interface, fixing the fragmentation of rewriting code for every tool added. 67 stars plus Show HN hit a real standardization pain point; the path is take rates and marketplace distribution, with potential to become a standard gateway, but the risk is MCP official and ecosystems like LangChain building in the same capability.
A Show HN: Give your AI agents access to WhatsApp Wraps WhatsApp as an MCP so agents can read, write, and search messages, with a Web UI for non-technical users, solving repeated WhatsApp integration per project; it drew HN discussion. Monetization is a hosted API/SaaS, but it directly violates WhatsApp's ToS, with high ban and data-compliance risk, and the official Business API is a strong competitor.
Hi, I built Chat-Man because I wanted a cheap way to give my agents access to WhatsApp without integrating a WhatsApp library separately in every project. You get WhatsApp MCP server, so you can conne…
A claude-vault Targets the specific pain that switching accounts in Claude Desktop loses sessions, offering local backup and restore that works on install and does only this one thing. 37 stars in one day shows real demand; the limit is covering only the desktop app, so it must follow if the official storage format changes.
Keep Claude Desktop chats safe when switching accounts — local-first backup and restore for conversations.
A pi-rain-radar A rainfall radar display on Raspberry Pi and Docker with animated playback, offline continuity and multi-screen LAN support. People wanting a dedicated weather screen get a clear deployment path; functionality is single-purpose but complete, and it depends on external radar data sources.
A self-contained Raspberry Pi and Docker rain radar display with animated observed rainfall, local history, outage-resilient playback and LAN multi-screen support.
A rimagent Lets an LLM play RimWorld and rewrite its own strategy manual between runs, with a RimBridge mod, a rare self-improving agent experiment. Interesting to agent researchers and RimWorld players alike, though it requires the game and mod, and results depend heavily on the model.
An LLM that plays RimWorld and rewrites its own playbook between colonies. RimBridge mod + self-improving agent.
A Show HN: An e-ink frame that hears birds and draws them as 1800s illustrations A Raspberry Pi e-ink photo frame identifies birds offline, then uses local AI to generate 19th-century hand-drawn style illustrations, grounding audio recognition and image generation in concrete hardware. 1237 upvotes and 1415 stars make it the strongest momentum this period. It could go consumer birdwatching hardware plus subscription against Bird Buddy; the risk is that recognition accuracy and cost determine repeat purchases.
See also: Avian Visitors - https://news.ycombinator.com/item?id=48343424 - May 2026 (20 comments)
A Show HN: Pizza Bot – An inbox for AI agents that work in the background Turns long-running AI agents into an email-like self-hosted inbox, using MCP and SKILL.md to manage approvals, memory, and resumable interruptions, solving observability for unattended background tasks, validated by 2,000 people at Amazon. Monetization could be an enterprise edition and skill marketplace, but the risk is head-on collision with big-vendor desktop agents.
Hi HN - long-time lurker (since 2012!), first time poster. Pizza Bot is a self-hosted desktop app for Mac, Windows, and Linux that runs AI agents in the background and exposes them through an email-li…
A egodup Detects repeated segments in first-person video and locates their time ranges, real dirty work in robot data cleaning, with a narrow scope and accurate description. The limit is it only helps egocentric dataset builders, and you'd need to confirm an installable package was actually released.
Local-first detection and temporal localization of reused footage in egocentric video.
A Show HN: Check if your IP has appeared in a residential proxy network Lets users check whether their IP is being abused by residential proxy networks, targeting the pain of IPs getting blacklisted and accounts flagged by crawlers and fraud. The 49 upvotes reflect real anti-fraud demand; a free lookup plus paid IP reputation API could rival IPQualityScore and Spur, but detection rules are easily evaded by proxy vendors.
A Show HN: TabPFN-3.5, a Tabular Foundation Model for messy real-world tables A tabular foundation model aimed at real messy tables, saving substantial feature engineering and tuning, since enterprise tabular modeling is a real need. Prior Labs has a team and productization intent, and its release cadence shows someone is on it; monetization via hosted API and private deployment, but AutoML platforms and open tabular models keep pushing prices down.
A Thunderstorm Collects environments, effective permissions, and trust relationships into an attack graph written as .rage.ndjson, by a well-known red-team developer. The idea is clear with real use, but only the output format exists; the consuming side isn't built yet.
A collector and derivation engine. It maps your environment, evaluates effective permissions and trust, and writes a complete attack graph as a single .rage.ndjson file. The input to everything else.
A gpu-lexer-textarea The author uses the new OpaqueRange Web API to put GPU lexing inside an editable textarea, a concrete and honestly described approach. It's a demo, not a library, and needs Chrome 152+; editor frontend developers should give it ten minutes.
Demo of using gpu-lexer with the new OpaqueRange web API (Chrome 152+) to create a universal syntax highlighter within an editable textarea
A awdl-toggle A small tool adding a persistent AWDL toggle to macOS Control Center, useful for those bothered by AirDrop/AirPlay interfering with Wi-Fi latency. Very narrow and honestly named; limited to one problem, with prerequisites unstated.
A native macOS Control Center toggle that persistently enables or suppresses AWDL.
A TressFX-Godot A GDExtension port of AMD's TressFX hair and fur solution to Godot 4; real-time hair is a long-missing piece in Godot. It's clearly valuable to Godot developers doing character rendering; the risk is that neither porting completeness nor performance has been verified.
Real-time hair and fur for Godot 4: a native GDExtension port of AMD TressFX
A dmgly A visual editor for DMG installers for Electron, Tauri, and native macOS apps, adding dynamic backgrounds and copy to replace create-dmg's command line plus manual image editing. Narrow but specific; limited to macOS distribution.
Visual DMG editor with animated GIF backgrounds, custom text, and export for Electron, Tauri, and native macOS apps
A Treemapolis Renders disk directories as a flyable 3D city, driving DirectX 12 and NativeAOT directly from C#, both a tech demo and a genuinely usable spatial viewer. Positioning is restrained; limited to Windows only.
Your disks, as a city you can fly through. Treemapolis turns the Windows shell namespace into a 3D treemap: every folder is a slab, every file a building as big as it weighs on disk. It is a technolog…
A fgate A GPU-side quality gate for first-person video, released the same day as another repo from the same org, looking like one piece of a robot data pipeline. The idea is solid but the description is too short, never saying which quality dimensions it judges; you'd have to read the code.
GPU-native quality gate for first-person video
A wisp Gives agents computer control on macOS: reading the accessibility tree, synthesizing input, hooking DevTools, plus a CLI and MCP server. Desktop agent builders can plug it in directly; limited to one platform, and official tools already have similar rudiments.
Wisp: computer use for agents on macOS (accessibility tree + synthesized input + DevTools), with a CLI and MCP server
A moqlivecast A from-scratch QUIC/HTTP3/WebTransport MoQ live streaming server, pushing and pulling streams in the browser via WebCodecs without existing QUIC libraries. A rare runnable sample for Media over QUIC explorers; experimental and far from production operations.
Next-gen live streaming server built on Media over QUIC (MoQ). From-scratch C implementation of QUIC/HTTP3/WebTransport — no third-party QUIC library. Browser publish & play via WebCodecs.
A Show HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and cost It rebuilds Qwen3-TTS/ASR into a speech-specific inference engine, proving open-source speech can be cheaper than closed source with sub-50ms latency and top WER on the Coval leaderboard, selling endpoints directly. 59 HN upvotes hit the cost pain point. Commercialization is the clearest here, a usage-based inference service; the risk is Alibaba's official optimizations erasing the price gap and vLLM adding multimodal inference.
Hey HN, Toby from Nari Labs here. We've been working on making OSS speech models super-fast. Last year, we built Dia, the first OSS text-to-speech model capable of doing natural dialogue. Since then, …
A ToolReplay It hash-chains AI agent tool-call records, adds deterministic replay and out-of-scope detection, filling the forensics and compliance audit gap after agents go live. 171 stars in 0.6 days show agent observability is missing exactly this piece. The zero-dependency CLI is easy for CI and regulatory workflows to adopt, with enterprise audit subscriptions as the path; the risk is trace platforms like LangSmith covering it incidentally.
Audit AI agent tool-call transcripts: hash-chain sealing, deterministic replay, and scope overreach checks. Dependency-free Python CLI.
A knowledgec-privacy-audit A zero-dependency Python script that reads and can clear macOS's hidden knowledgeC.db behavior log, so you can see what the system records about you. Clear goal and restrained implementation suit privacy-minded Mac users; it only covers this one database.
See what macOS's hidden knowledgeC.db activity log has recorded about you, and clear it. Zero dependencies.
A fly-self-driving Wires the fruit fly connectome (165,000 neurons) into a driving simulation from pixels to steering, a rare whole-brain emulation experiment worth ten minutes for neurocomputing people. The limits are high reproduction cost and a one-off research demo far from a daily-use tool.
A fruit fly connectome (165k neurons, 25.6M synapses) trained to drive a simulated street from pixels. Code, patches, renderer, results.
A redis-near-cache Client-side near-cache on top of StackExchange.Redis, evicting stale values immediately via Redis invalidation notifications. It addresses a real pain point of reducing hot-key latency in .NET services with a small, clear scope; consistency behavior needs your own evaluation.
Client-side caching on top of StackExchange.Redis. Keeps the values you read in memory and drops them the moment Redis says they changed.
A HeadOrbit A macOS menu bar tool using AirPods head tracking: blur the screen when you look away, get reminded when your posture slumps—a concrete idea almost nobody has done. It depends on AirPods and newer systems, with accuracy and battery life as ceilings, but worth ten minutes to try.
Your AirPods know where you're looking. macOS menu bar app: look-away blur and posture reminder from AirPods head tracking.
A msteams Adds a Microsoft Teams-based communication channel to Mythic C2, hiding commands in normal enterprise IM traffic. It is a clearly useful channel addition for red-team exercises with a restrained scope; the limits are dependence on a Teams tenant, non-trivial setup cost, and serving only Mythic users.
Mythic C2 MSTeams Communication Channel
A HyMT2Sharp A C# library from sdcb that, judging by the name, brings the Tencent Hunyuan-MT translation model into .NET; the author has credibility in inference wrappers. Useful for local translation in .NET services, though the empty description means you must read the code to confirm.
A Pokemon-Bank-Offline-Mod An offline patch for Pokemon Bank on 3DS that lets you still access your Pokémon after the servers shut down, with a migration mode too; very narrow scope but a clear goal, practically useful to the 3DS preservation community. The barrier is needing a modded 3DS with a 3gx environment.
3gx plugin to use Pokemon Bank offline when the servers go down. Includes a migration mode.
A nntc A new work from the basis universal author: a non-neural PBR texture compression scheme that saves VRAM and bandwidth for game and 3D pipelines. The idea is clear and not overhyped, early but concrete; the limitation is no compression-ratio or quality comparison against BC and ASTC yet.
NNTC (Non-Neural Texture Compression) - PBR Material Compression
A S-MU2000 Runs Yamaha MU2000 hardware firmware directly inside a VST3, effectively a software recreation of an old sound module, with real value for retro MIDI arrangers. The author states it is unofficial and unaffiliated with Yamaha, which is honest; the limitation is that you must supply the firmware, putting it in a gray area.
Software sound source for the Yamaha MU2000. A VST3 that runs the actual hardware firmware as-is. Unofficial, not affiliated with Yamaha
A doom_epub Packs the entire 1993 DOOM shareware episode into an EPUB3 and runs it in e-book readers via WebAssembly; the goal is absurd but the path is clear and scope restrained, worth ten minutes for e-book and WASM tinkerers, with practical value ending at the stunt.
A self-contained EPUB 3 that runs the complete 1993 DOOM shareware episode inside ordinary ebook readers (WebAssembly, no plugins, no network)
A textlint-rule-preset-ai-words-ja A textlint rule for Japanese writing that catches words and sentence patterns frequent in AI-generated text. Japanese editors and doc teams using textlint can install it directly; the limitation is heuristic word lists cause false positives, and it only covers Japanese.
A textlint preset that finds words and phrases commonly appearing in AI-written Japanese
A lokv A small library from Tailscale implementing an append-only log on a create-only ordered KV store. It targets infrastructure developers needing a lightweight WAL/log layer, with restrained interfaces and boundaries; the limitation is that it is a pure library with a narrow audience, so adoption is hard to judge short-term.
lokv implements an append-only log over a sorted, create-only key/value store
A Any.ADB Pins the ADB debugging port on any Android device via Sui or Root, solving the problem of wireless debugging ports changing each time. Clear scope and ready to use, convenient for frequent device debuggers, provided the device already has Sui or Root.
Enable a fixed ADB remote debugging port on any Android device using Sui or root privileges. Lightweight, simple, and ready to use. Use Sui or Root privileges to enable a fixed ADB remote debugging po…
A sap-tui A SAP DIAG terminal viewer written in Go that lets you see the server screen from the command line without SAP GUI. A rare small tool for SAP automation and security testing; the limitation is it covers only a slice of DIAG, with stability and setup unclear.
sap-tui — a SAP DIAG terminal viewer: watch a SAP DIAG server's screens in your terminal, no SAP GUI required
A openeval It provides a typed prompt-and-judge SDK, CLI and evidence viewer for agent evaluation, consolidating scattered scoring scripts into a reproducible pipeline. 57 stars in 1.6 days reflect evaluation moving from demos to engineering. Open source plus cloud result hosting and CI integration subscriptions could monetize it; the risk is Braintrust, LangSmith and Inspect already occupying the space, and platforms shipping their own evals.
Typed prompt-and-judge SDK, CLI, and evidence viewer for agent evaluations
A Show HN: Graphify C# – Compiler-accurate Find Usages for coding agents Provides compiler-accurate Find Usages for C# coding agents, fixing LLMs' inaccurate symbol resolution when reading code, and supports C#15 syntax. Show HN's 41 upvotes plus 50 stars in a day show the C# agent-tool gap is real. It could be an MCP or LSP middle layer charged per seat, or be acquired by an IDE vendor. The risk is being outflanked by official Roslyn and LSP solutions like Serena.
A dearagent A self-hosted inbox API for AI agents running on Cloudflare Workers with an MCP server, explicitly positioned against AgentMail. Usage is clear and deployment cheap, so agent workflow builders can adopt it today, though similar products are multiplying fast.
Email inbox API for AI agents, self-hosted on Cloudflare Workers. Open source, with an MCP server. AgentMail alternative.
A ai-data-extractor It unifies export of conversation histories from Claude Code, Cursor, Aider and others, solving the problem of AI coding records being locked into platforms with no backup or migration. 163 stars in a day show real user anxiety over data sovereignty. It could extend into team conversation archiving and training data pipelines; the risk is frequent format changes and official export features appearing anytime.
Free open-source extractor for AI coding assistant chat histories. Supports Claude Code, Cursor, Windsurf, Aider, Cline/Roo Code, and more.
A Airport_RTW88 A native AirPort driver for Realtek RTL88xx on macOS, letting Hackintoshes and older Macs use third-party wireless cards. This area has long had only half-finished solutions, so the effort is rare; but there is only one sentence of description, no supported chip list or working status, and 8 stars in a day means nobody has verified it.
Native AirPort Wi-Fi driver for Realtek RTL88xx wireless adapters on macOS.
A ccompactor Compresses agent sessions from Claude Code, Codex and others into sourced, verifiable handoff documents for the next agent. It targets the real context-handoff problem in multi-agent work with a focused CLI and restrained claims; not yet proven that compression preserves key information.
TypeScript CLI: extract any coding-agent session (Claude Code, Codex, Pi) into a compact, verified, provenance-linked handoff another agent can continue from.
A kernwatch A terminal kernel observability panel written in Rust with Ratatui, for ops and performance engineers watching kernel metrics on servers. Its positioning is clear and not overstated, but bpftrace and btop already cover most of it, so whether its TUI interaction differentiates matters.
Linux kernel observability in your terminal. Built with Rust and Ratatui.
A ComfyUI-Olm-YuE2 A ComfyUI node that generates songs with YuE2 by style and lyrics, including score viewing and stem generation. It is a ready-made building block for those already doing audio experiments in ComfyUI, and the author's Experimental label is appropriately measured.
Generate songs from a musical style and lyrics in ComfyUI using YuE2. Optional score viewing and editing, separate generation stages, and stereo 48 kHz audio output. Experimental.
A toolhost Lets an LLM write its own tools at runtime: acorn parsing, SQLite versioning, execution in a supervised subprocess. Unlike agent frameworks with hardcoded tools, it seriously handles persistence and isolation of dynamic code, worth ten minutes for anyone building agent infrastructure. The permission model is still unclear.
Let an LLM write its own tools at runtime: parsed with acorn, versioned in SQLite, run in a supervised worker process
A svelte-annotate A small library adding hand-drawn style annotations to Svelte components, aimed at frontend developers doing presentation or teaching markup. The scope is restrained and the description honest, and it works once installed, though rough-notation exists and this merely brings it into the Svelte ecosystem.
〽️ Hand-drawn annotations for Svelte
A vd_agent Revives the deprecated freedesktop Windows SPICE guest agent for people running Windows VMs in QEMU/Proxmox. The problem is real and the origin and licensing are clearly stated, making it typical gap-filling maintenance, though its feature scope is narrow and follows upstream specs.
Windows SPICE guest agent. Community-maintained fork of the abandoned freedesktop.org vd_agent.
A flm Connects a frozen language model to the fully preserved fruit fly MaleCNS connectome for training and dialogue, the week's strangest research toy. It has experimental value for the neuromorphic-curious, but depends on connectome data and yields no reusable conclusions.
Train and chat with a frozen language model coupled to the full retained MaleCNS fly connectome.
A pg-circuit A native extension adding a circuit breaker to production PostgreSQL, deciding from live database state whether dangerous SQL passes, warns or is blocked, one step ahead of after-the-fact auditing. DBAs and platform teams will want to try it; policy configuration and false-block costs are on you, and it competes directly with proxy-layer whitelisting.
Circuit breaker for production PostgreSQL. Native extension that can ALLOW, WARN, or BLOCK dangerous SQL based on live database state. Apache 2.0.
A konpeito A self-hosted RSS reader running on Cloudflare Workers + D1, supporting only passkey login and compatible with the Google Reader API. Deployment is easy and the positioning is clear, but FreshRSS and Miniflux already hold this ground, so it must win on minimal ops.
A self-hosted RSS feed reader on Cloudflare Workers and D1. Passkey-only sign-in, Google Reader API compatible.
A ptcg-population-rl A self-built C++ battle environment training a Pokémon card AI with GBDT imitation learning, Transformer distillation, then PPO, aimed at card-game reinforcement learning people. The training pipeline is described concretely without hype and is more complete than most RL repos, but it lacks reproducible win-rate data and currently reads more like a personal experiment log.
A minimal Pokémon TCG learning pipeline: C++ arena, GBDT imitation, Transformer distillation, and PPO.
A tracecrate A local-first AI agent trace workbench that reads Claude Code, Codex and OTLP logs, compares runs and exports redacted reports, targeting agent debugging and compliance pain. It hit 98 stars in 1.6 days with no backend or keys needed, and could monetize via team subscriptions and enterprise audits; its main rivals are cloud observability platforms like Langfuse and LangSmith.
Local-first AI agent trace workbench. Inspect Claude Code, Codex and OTLP logs, compare runs, and export privacy-conscious reports. No backend or API keys.
A reburp Exposes Burp's entire Montoya API as a local REST interface with Swagger UI, letting scripts and agents drive Burp for automated testing, filling a real gap. Clear scope and an obvious starting path; concurrency and stability still need real-world testing.
A Burp Suite extension that exposes the full Montoya API as a local REST API, with Swagger UI
A crossmatch Drives both the iOS and Android versions of an app and records each UI difference as side-by-side video, filling the neglected gap of cross-platform consistency testing. Clever idea; the hard part is running both automation suites on real devices, and setup instructions are still lacking.
Find where your iOS and Android apps disagree, and get a side-by-side video of every difference.
A Toneferry Streams Windows system audio to a phone on the LAN for real-time listening—one sentence explains the use and it works once installed, good for people who want to hear their PC on their phone. Closed-source tools like AudioRelay exist, but it gives no latency or audio-quality data.
Windows system audio → real-time listening on a phone over LAN.
A askWAM A small tool that requests Entra ID tokens directly from Windows Web Account Manager, from a security researcher familiar with the mechanism, saving red teams and token forensics a lot of guesswork. Clear scope and restrained description; only meaningful in authorized testing scenarios.
Ask the Web Account Manager (WAM) for Entra ID tokens
A sctxx Extracts session records from different agents like Claude Code and Codex into compact handoff files with source links, so another agent can continue the work; cross-agent context migration is indeed a new problem. The concept is clear; compression quality and format changes across vendors are the main risks.
sctxx — coding agent Session Context Extractor. standalone, agent-agnostic CLI (plus a thin Agent Skill wrapper) that reads a coding-agent session transcript from disk — Claude Code, Codex CLI, or Pi …
A foldelight Drives a native macOS origami-glass animation from a MacBook's lid angle—tiny in scope and restrained in description, the kind of toy you want to tweak a couple of lines of. Beyond fun it has no real use, and it's locked to a specific model's sensor.
A native macOS folding-glass effect controlled by the MacBook lid.
A catalogflow A product catalog pipeline for WooCommerce that emphasizes human confirmation and safety, putting a gate on the risky task of AI bulk-editing product data. The scope is clear, but it's only useful within the WooCommerce ecosystem.
Safety-first, human-approved product catalog pipeline for WooCommerce
A agent-launcher Wraps existing coding agent CLIs like Claude Code and Codex into one desktop shell for unified configuration and launching, solving a real hassle for people running several agents on one machine. Restrained in scope without hype; the challenge is keeping up with frequently changing CLI arguments.
Configure and run existing coding-agent CLIs from one desktop app.
A Daub Renders brush stroke JSON directly to PNG and outputs layered kra/psd files plus drawing replay videos, running batch renders outside the Krita process, useful for drawing-process videos or asset pipelines. The repo is tagged Python while the text says pure Rust, an inconsistency that needs clarifying first.
Headless calibrated-brush renderer: strokes JSON to PNG + layered .kra + layered .psd + reveal video. Pure Rust, no Krita process needed.
A RomMMobile An Android client for people running their own RomM, downloading ROMs straight into the frontend's directory and saving a genuinely annoying step. The scope is clear and the description honest, but it's entirely tied to the RomM ecosystem, so the audience is narrow.
Android companion app for RomM: browse your self-hosted library and download ROMs straight into the right folder of the frontend you already use.
A DuoLauncher An experimental Android launcher for foldables: a right-side dock, native widgets, and stackable home screens when unfolded, addressing the long-standing gap in Android large-screen adaptation. Fairly complete, but stability, battery drain, and support for new models remain to be seen.
An experimental Android launcher for foldables, with a right-side dock, native widgets, and overlapping unfolded Home pages.
A windows-wesp-research Reverse engineering and hands-on validation of a Windows endpoint security platform preview, closely following Microsoft's new interfaces and directly useful to EDR and offensive/defensive researchers; the limitation is that it's research notes, not packaged as a reusable tool.
Reverse engineering and live validation of the Windows Endpoint Security Platform preview
A Show HN: Open-source simulation testing infra for voice agents Simulation testing and online monitoring for voice agents, turning repeated manual phone-call evals into batch-runnable infrastructure, with built-in STT/LLM/TTS keys and zero markup on inference. Only 12 likes but 112 stars show developer approval, a Langfuse-style open-source plus cloud-hosted model has clear monetization, and competitors are voice platforms' built-in testing.
Hey HN, we’re Nischal & Naman. We’re brothers, and together we’re building an open-source platform for simulation based testing of voice agents (try it out in 5 mins - https://docs.egma.ai/docs/get-st…
A gongwen-gbt9704-skill Generates editable DOCX official documents per GB/T 9704-2012, solving the high-frequency rigid pain of manually adjusting fonts and line spacing in government and SOE formatting, and the clear standard naturally gives it a "standard dependency" position. It can pursue template markets and enterprise editions, with the risk that office suites like WPS ship similar templates.
Cross-platform Chinese official document typesetting Skill that generates editable DOCX following GB/T 9704-2012
A Show HN: MultiMatte, a Promptable Image Background Removal Model A promptable background-removal model using alpha mattes for hair and motion blur, significantly beating SAM 3 on DIS5K and solving the "keep only the specified object" matting need. Open weights and the NoBg library funnel into a custom-model business, with the risk that SAM and commercial matting APIs catch up fast.
Hey HN, I'm Shreyash from Feyn. We help companies build custom models from their data. Today we're releasing MultiMatte, a background removal model you can aim with words. Name an object in your image…
A opensend A self-hosted transactional and marketing email layer built on AWS SES, addressing high sending costs and vendor lock-in, with the Resend model already market-validated; 100 stars in a day shows real demand. It can be open source plus cloud hosting or usage-based pricing, competing with Resend, Plunk and self-built Postfix; the risk is that deliverability and compliance operations are heavy.
self-hosted transactional & marketing email layer on top of aws ses
A TridentVision Puts YOLOv26m entirely on an XC7K325T in pure Verilog with no C, a rare and reproducible reference for FPGA edge vision; the specific LUT/DSP figures suggest it really ran. The catch is the high hardware barrier, so most people can only read it, not run it.
Implement yolov26m using a complete FPGA, without any C language code. Requires 160K LUTs and 386 DSP cores (based on Xilinx XC7K325T).
A sam3d.cpp Converts Meta's sam3d weights to GGML so 3D objects can be tracked and segmented from video locally or on edge devices, one of the few offline 3D segmentation options you can compile yourself. Currently only a demo; accuracy, speed, and docs remain unverified.
Track and extract any 3D object from video. C++/GGML convertion of Meta's sam3d PyTorch model plus a demo app
A Show HN: Ctrlb-decompose: Strip the noise from logs before sending to LLMs Solves logs drowning LLM analysis in noise by decomposing logs into a surface usable for reasoning. It gained 297 stars in 0.5 days, hitting an AI debugging pain point; commercialization means integration into observability or AI debugging tools or becoming an open-source component standard, but existing log platforms may build similar features or the project may stay too niche.
A Show HN: Type.com: Multiplayer Codex/Claude in the cloud for non-tech use cases A multiplayer AI agent cloud IDE for non-technical users, built on Claude Code/Codex and usable in Slack. Its 14 HN points come from a very early stage, but positioning is clear and commercialization prospects are good; risks are direct competition with Claude Cowork and costly VM infrastructure, and the path to a standard tool is team collaboration stickiness.
Hey HN, I'm Komran, the CTO and cofounder of type.com. We're launching today. I've been describing type as "the IDE for non technical people." It allows you to build really cool shit without having to…
A nervous-system-atlas A browser-based 3D clinical neuroanatomy atlas: 592 meshes, synchronized MRI slices, vascular territories and tracts, case modes and quizzes, with entries citing open literature, in English and Turkish. A genuinely usable tool for medical students and educators; the downside is heavy frontend and data with high maintenance cost.
A browser-based 3D atlas of clinical neuroanatomy in MNI152 space: 592 meshes with synchronised MRI slices, arterial territories, traced pathways, a syndrome and lesion mode, clinical topics, a glossa…
A dst-character-studio A character art workbench for Don't Starve mod authors: layered drawing, rig calibration, animation preview and export to build, complete around one task. The positioning and description are honest, without jargon; limited to the small DST circle, only a day old, and actual completeness needs installing to judge.
A local art workbench for Don't Starve Together characters: layered drawing, binding calibration, native animation review and build export.
A effect-clickhouse A ClickHouse client for the Effect ecosystem: type-safe queries, result decoding, plus reproducible benchmarks. The scope is restrained and the naming honest, filling Effect's gap in OLAP scenarios; the limitation is that it only matters to teams already using Effect.
Type-safe ClickHouse queries, result decoding, and reproducible benchmarks for Effect and TypeScript
A appeditions Batch-generates localized app store screenshots locally and tracks per-version, per-language progress; no account, no uploads. Such assets are mostly paid online services, so a local Go desktop app is a clear difference. The limit is it only covers the narrow iOS/Android store asset case, and template and review-size fit still need verification.
Turn raw app screenshots into store-ready App Store and Google Play assets, and keep tracking of the version you are in, in every language you ship in. A local Go desktop app — no account, no upload, …
A self-hosted-ci-runner Starts multiple GitHub Actions runners on one machine with a single command, directly addressing heavy CI filling disks and OOM: each runner gets its own swap and scratch plus cleanup scripts, with a Hetzner reference config. Teams running self-hosted CI can use it directly; the limit is it is more an ops script than a general solution.
Stand up self-hosted GitHub Actions runners on one box, provisioned so heavy CI jobs don't OOM or fill the disk (per-runner swap/scratch + reaper). Parameterized; Hetzner reference config included.
A wsl-reclaim WSL2 disks that only grow are an old problem: clean up inside the distro first, then compact the vhdx on the Windows side, and it also explains why diskpart silently fails on sparse disks. Directly relevant to WSL users on Windows. The limitation is that it is a pure shell script relying on manual confirmation, winning only by explaining the principle clearly.
Reclaim WSL2 disk space for real: clean inside the distro, then compact the .vhdx on Windows. Explains why diskpart silently does nothing on sparse disks.
A mmd-mcp Exposes MikuMikuDance modeling and MME settings as a stdio MCP server on Windows, letting LLMs drive MMD directly — a rare interface. The scope is clearly stated; limited to MMD users on Windows, and how much it actually covers needs testing.
Windows stdio MCP server for MikuMikuDance authoring and MME settings
A polymo An electronic pet: ESP32-S3 firmware plus an Android app, with all 'thinking' done locally on the device; the hardware plus on-device inference combo is concrete enough for DIYers to follow. The limit is that it's only a virtual pet toy, and reproducibility and long-term maintenance need more days to judge.
A physical digital pet you keep alive — ESP32-S3 firmware and an Android app that does all the thinking on-device.
A bankmcp Lets AI read bank accounts via MCP, self-hosted and read-only, solving financial data access; 165 stars show interest in open banking plus AI. Could become a financial MCP server standard. Risks are bank API limits and security, with Plaid and others as competitors.
BankMCP™: your AI can now read your bank. Self-hosted, read-only MCP server for your own bank accounts via open banking (Enable Banking). Standard MCP; tested with Claude and Ollama.
A short-video-generator-AI Automatically turns long YouTube videos into short clips with highlight detection, subtitles, translation and voiceover. Its 47 stars in one day show real demand; commercialization could be a self-hosted content automation tool or a media company pipeline, but the biggest risks are competition from commercial AI video tools and API costs.
Free open-source project designed for turning youtube-viedos into viral short videos. Highlight detection, subtitles, translation, voiceover, all in one for your content.
A Show HN: Engrim – A universal, local-first SQLite memory engine for AI CLIs This provides a local-first SQLite memory engine for AI CLIs, solving lost session memory, and 90 Hacker News points show AI developers approve. Commercialization could be a standard dependency or hosted service; risks include SQLite performance bottlenecks, and competitors are Mem0 and LangGraph memory.
A pcb-skill Lets an AI agent take a concept to a manufacturable PCB by driving EasyEDA Pro; 86 stars show interest in hardware plus AI. Could be a subscription PCB design service. The main risk is EDA tool dependence, with Copperhead as a competitor.
Agent skill for taking a hardware idea to a manufacturable PCB: concept, schematic, sourcing, layout, routing, verification, and a purchase staged to the pre-payment page. Drives EasyEDA Pro over MCP …
A maintainable-update It addresses codebase maintainability by automatically sweeping the whole repo with copying, clarity, line breaks, splitting, renaming and tidying. 45 stars in a day shows developer interest. It could commercialize as a standard code quality CLI or CI integration tool; the biggest risk is being too opinionated and producing noisy changes, plus competition from existing formatters like Ruff and Black.
Repo-wide maintainability sweep: copy, clarity, wrap, width, split, rename, tidy
A Omni-Interaction-Agent An open interactive agent for full-duplex voice collaboration, audiovisual perception and long-horizon tasks, positioned at the frontier; 24 stars in 1.8 days shows research interest. Could become a multimodal agent development base or standard framework. Risks are inference cost and competition from OpenAI Realtime and Agora.
Open-source end-to-end omni interaction agent for natural full-duplex voice collaboration, continuous audio-visual perception, and asynchronous long-horizon agentic execution.
A embedflow Solves the downtime from full recomputation when upgrading embedding models in LLM/RAG applications; 31 stars in 1.2 days shows a real pain point for vector retrieval developers. Could become an index migration middleware sold to enterprise RAG platforms. Risks are needing to support many vector stores, with competitors like Weaviate's built-in reindex and pgvector rebuilds.
Zero downtime embedding upgrades
A france-mcp-servers A directory of French data, legal, and service MCP servers addressing French-speaking users' need for localized AI tools; 79 stars show growth. It could become a regional MCP standard entry, but the regional market is small and official data interfaces may be unstable.
A curated catalog of Model Context Protocol servers for French data, laws, and services.
A Show HN: Wg-admin – web UI for an existing WireGuard host wg-admin gives existing WireGuard hosts a web UI, reading and writing /etc/wireguard and applying changes with wg syncconf to avoid interface restarts. HN's 32 points show network engineer approval; risks are WireGuard's own simple configuration and security exposure.
Reads /etc/wireguard, lets you add/edit/remove peers, applies with wg syncconf (no interface bounce). Does not install WireGuard or rewrite your PostUp/NAT. That's all
A lynx An automatic Core Erlang to Lean translator that generates proofs for Elixir/Erlang programs, by the creator of Elixir—a rare and clean approach. Worth a look for formal verification or concurrent service work; the limitation is it's a research prototype not yet pluggable into CI.
Proofs for Erlang/Elixir programs through LEAN (via automatic translation of Core Erlang)
A dkyj-director Uses an iPhone as a viewfinder for camera previs and editable blocking in Blender, with MCP so agents can take part, far lighter than traditional previs workflows. Attractive to indie animation and live-action previs people; it depends on Blender plus Safari, so stability remains to be verified.
DKYJ Director / 大开眼界导演台 — Local Blender camera previs with an iPhone Safari viewfinder, editable blocking, reusable scenes and optional MCP agent integration. macOS / Windows.
A qalam Focused Arabic PDF extraction that preserves word order logic, restores base letters and ligatures, and judges per page whether OCR is truly needed. These language-specific pitfalls are long ignored by general parsers, making it worth trying for RAG or corpus cleaning; currently Arabic-only.
Arabic PDF text extraction that doesn't quietly corrupt your corpus: logical word order, base letters not presentation forms, ligatures intact and a per-page verdict telling you which pages genuinely …
A huashu-mac-use This lets agents control native Mac apps without APIs via accessibility features while keeping forensic logs, filling a gap in agent desktop automation; 53 stars indicate demand. It could become a standard Skill for tools like Claude Code, but competes with AppleScript and UI automation frameworks and must handle permissions and stability.
Let any agent control native apps on Mac that have no API: read the background, write without disruption, keep evidence at every step · Agent Skill for macOS computer use
A BreezeTTS2_Mac_Streaming Targeting redundant acoustic code computation in the mlx-audio deep decoder, it reuses intra-frame KV and does one CPU readback per frame, cutting Breeze-TTS-2 RTF on Apple Silicon from about 4 to about 1. Written for Mac developers running local speech, with concrete numbers and no hype; the limit is that it serves only one model and one hardware.
Real-time Breeze-TTS-2 streaming on Apple Silicon — RTF from ~4 down to ~1. mlx-audio 0.5.1's depth decoder recomputes each frame's existing acoustic codes instead of reusing its KV cache. Adds intra-…
A llm-api-speedtest A Go tool for reproducible load testing of LLM gateways and proxies, outputting TTFB, token speed and success rate as JSON reports, runnable on scheduled Actions. Useful for vendor selection; it only measures speed, not quality.
Benchmark LLM APIs, OpenAI-compatible gateways, and API proxies by token speed, TTFB, latency, and success rate. Compare multiple providers with reproducible JSON reports and scheduled GitHub Actions.
A picowallet A solder-free stablecoin hardware wallet built from three Amazon parts: a Pico 2 W, an LCD, and an ATECC608. The scope is restrained and the bill of materials clear, suiting people who want to verify their private keys never touch the network; the limitation is DIY hardware with no audit or firmware update commitment.
A stablecoin hardware wallet from three Amazon parts, no soldering. Pico 2 W + LCD + ATECC608.
A lium-localmaxxing A single script reproduces inference results for 7 benchmarks and 20 verified models on rented Lium GPUs, plus measured cost per million tokens. Useful for those building their own inference and comparing GPU rental value; the numbers are more credible than empty benchmark talk. The limitation is being tied to the Lium platform and a fixed model list.
Reproduce Lium's LocalMaxxing leaderboard runs (7 boards, 20 Verified) on a rented Lium GPU in one script — vLLM/llama.cpp recipes, measured cost per million tokens
A mikrotrick-poc A local reproduction environment for the RouterOS SSH public-key authentication bypass (CVE-2026-67276), by a security researcher with a track record, with scenarios and impact scope described with restraint. Ops and red teams can use it to verify their own devices; the limitation is that it stops at PoC, with no before-and-after patch comparison.
CVE-2026-67276 RouterOS SSH public-key authentication bypass lab PoC
A Show HN: We Beat MLPerf: Modern Storage for KV Offload and LLM Training Claims leadership in MLPerf storage benchmarks, using modern storage to solve the IO bottleneck of KV offload in LLM training. 35 HN points show infrastructure community recognition; the business path is clear as high-performance storage for AI training clusters, with risks that MLPerf results may be questioned and competition from NVIDIA and other major vendors.
A Show HN: Moadim.io – A scheduler for agents A local Rust daemon agent scheduler supporting Git compatibility, agent agnosticism, multiple runners, MCP/UI/HTTP, with 1,000+ users already. 32 HN points show real demand; monetization is team agent infrastructure subscription, with the risk of fast-changing agent ecosystems against cron and workflow engines.
Why can't we get an agent scheduler that supports all of the following: - git compatible - agent agnostic - 100% open source - os and system-agnostic - multi-runner support - support mcp/ui/http - unl…
A wechat-intelligence-hub Turns WeChat chat history into a local-first data source for agents, with a read-only CLI, Codex skill and daily briefings, bridging personal data and AI workflows. 393 stars in 1.4 days show real demand; monetization is subscription knowledge management/CRM or enterprise compliance, with risks of WeChat bans and privacy compliance against existing export tools.
Local-first WeChat intelligence system with a read-only CLI, Codex skills, searchable chat history, daily briefings, follow-ups and opportunity tracking.
A omapop A selection popup for Omarchy (Arch + Hyprland) that is interface-compatible with extensions for the paid macOS tool PopClip, filling a gap Linux has long had. Heavy text users and Mac migrants benefit most; it is written in QML and scoped to one job. The limitation is that it is tied to a specific desktop, so other environments need their own adaptation.
Select text and act on it: an Omarchy popup with PopClip-compatible extensions.
A expo-healthkit It uses the Expo Modules API to unify Apple HealthKit and Android Health Connect behind one JS interface, eliminating duplicated work for cross-platform health data access in RN; the scope is clear and very practical for mobile health apps, though platform permission reviews need attention.
Apple HealthKit and Android Health Connect for Expo and React Native, built on the Expo Modules API. One JavaScript API on both platforms.
A bgt A Python background thread wrapper from Hynek that adds crash restart and fault tolerance to threads, filling exactly the gap threading never handled. Python developers writing long-running services and background tasks can use it directly; stating the boundary in one line is a plus. The limitation is that similar supervisor wheels are plentiful, and 9 stars is too early to tell whether it will be maintained.
Fault-tolerant background threads for Python
A bitcoinfuzz Differential fuzzing across multiple Bitcoin protocol implementations and libraries—a rare and concrete low-level security tool that helps implementers find consensus or parsing differences. But you must prepare your own corpus and build environment, so it leans toward research and auditing.
Differential Fuzzing of Bitcoin protocol implementations and libraries
A BootViewer It parses Windows measured-boot TCG/TPM event logs, explains each entry, and cross-checks PCRs against the real TPM to detect tampering or Secure Boot anomalies; this is practical for security operations and forensics, with a clearly defined scope.
Windows Measured Boot viewer & TPM verifier. See exactly what your PC measured at boot. Decode the Windows TCG/TPM event log, explain each entry, verify PCRs against the real TPM and spot tampering, S…
A raycast-homelab A Raycast extension that brings 23 self-hosted services such as Jellyfin, *arr, Immich, and Uptime Kuma into the command palette, each optional; it is a genuinely useful efficiency tool for homelab users, though it depends on the stability of each service's API.
Your whole homelab in one Raycast extension: 23 commands and a Home hub for Jellyfin, Jellyseerr, the *arr apps, Audiobookshelf, Nextcloud, Immich, Uptime Kuma, Firefly III and more. Every service opt…
A dlssdl It queries NVIDIA's NGX update servers directly, pulls DLSS SR/RR/FG and Streamline, and renames them to consumer filenames, effectively adding a "fetch from source" step to DLSS Swapper. Useful for DLSS tinkerers and mod authors, and the single-purpose egui plus D3D12 positioning is clear. The limitation is that it only serves Nvidia users, and 10 stars is too few to show real usage.
Check NVIDIA NGX update servers for DLSS SR / RR / FG and Streamline updates and download them with proper consumer file names (egui, D3D12 renderer)
A Snap64Recomp Uses static recompilation to turn N64's Pokémon Snap into a native Windows program, explicitly requiring your own cartridge dump and shipping no game data—an honest boundary. Valuable to the recompilation and N64 porting scene, but the entry barrier for ordinary players is not low.
Native Windows port of Pokémon Snap (N64) by static recompilation. You supply your own cartridge dump; no game data is included.
A image-to-css-art Offline-traces reference images into single-file HTML+CSS illustrations, with contour extraction, local gradients, and output auditing—more restrained than typical image-to-code tools. Good for image-free web decorations, but complex images and browser compatibility remain hurdles.
An Agent plugin that offline-traces reference images into pure HTML + CSS single-file illustrations, with contour extraction, local gradients, and output auditing.
A FlashRec A CUDA Graph inference engine for generative recommendation supporting wide beam search and FP8 serving, solving recommendation latency and throughput bottlenecks. 85 stars show AI infrastructure interest; a recommendation service or open standard could commercialize, but NVIDIA Merlin and in-house systems compete, and ecosystem lock-in is the risk.
FlashRec is a CUDA-graph engine for generative recommendation: wide beam search (3–5 SID steps, n=50–512+) over a trie-constrained catalog, in-process FP8 serving, and ranked beams on /v1/chat/complet…
A Show HN: TERMy – A fast terminal assistant that does not use LLMs It replaces LLMs with a lightweight NLU pipeline for a terminal assistant, responding in milliseconds on a Raspberry Pi and solving the cost, slowness and connectivity needs of LLM assistants. 87 HN points proves developers want fast offline assistants. It could commercialize as an enterprise intranet security assistant or embedded terminal tool, but capability ceilings are obvious, competing with LLM assistants like Copilot.
I love research and development, you may have heard of me because of PJON (Padded Jittering Operative Network). It is a network protocol I started developing in 2010, which was recently implemented in…
A vulnometry Replaces abstract CVSS scores with a business exposure index integrating NVD, EPSS, KEV and other data, solving vulnerability prioritization. 55 stars show security practitioner approval; SaaS subscriptions could commercialize, but RiskSense and internal systems compete, and delayed data source updates are the risk.
Measure what a vulnerability is worth to your business, not how severe it is in the abstract. Business Exposure Index scoring over NVD, EPSS, CISA KEV, OSV and GHSA.
A nopasswd-sudo Time-limited passwordless sudo for AI agents: a systemd timer revokes it on expiry, visudo validates before install, and sudoreplay keeps an audit trail. The scope is restrained and the description honest, filling the security gap around letting agents run system commands; the limitation is it only covers Linux plus systemd.
Time-boxed passwordless sudo for agents. Grants expire on a systemd timer, validated with visudo before install, audited with sudoreplay.
A usque-custom-pro Visually generates Cloudflare WARP/MASQUE configs and converts them to Clash, sing-box and other clients, simplifying proxy setup. 173 stars in two days show strong demand from censorship-circumvention users; monetization could be paid config services, but Cloudflare terms and regional blocking apply, against the official WARP client.
A visual registration, config generation, and multi-client conversion tool based on Cloudflare WARP / MASQUE. Generation, Clash/Mihomo smart routing, Shadowrocket, sing-box, local VLESS bridging, and …
A psd2live Automatically converts layered PSDs into Live2D models with layer recognition, mesh subdivision, physics and animation, solving time-consuming manual Live2D rigging. 39 stars show VTuber community demand; a subscription tool could commercialize, but official Live2D Cubism competes, and unstable generation quality is the risk.
PSD2Live is an automated Live2D model generation pipeline and desktop application. Given a layered PSD file, the system automatically performs layer semantic recognition, connected-component bilateral…
A Show HN: Reactor Atlas A nuclear reactor map and intelligence layer integrating operational, commercial, news and geological data, solving fragmented nuclear information. HN 63 points reflect attention amid the nuclear revival; institutional subscription alerts could commercialize, but data accuracy and IAEA official data are the biggest risks.
Hi HN, I'm a nuclear engineer and tech entrepreneur. After working for 3 years in the National Atomic Energy Commision (CNEA) in HPC software for nuclear simulations, I co-founded ratherlabs.com, a so…
A halo-kmp Brings iOS Live Activities and Android 16 live notifications under one Kotlin Multiplatform API—a genuinely missing off-the-shelf solution for cross-platform status bars. The boundaries are clear and it suits teams already on Compose Multiplatform; but it's one day old with no release artifacts or examples, so read the source before integrating.
Halo: Kotlin Multiplatform Live Activities. ActivityKit on iOS, Android 16 live-update notifications, one API
A data-deletion-skill Turns personal data deletion into a Claude skill: 328 data brokers, 13 privacy-law jurisdiction templates, and escalation paths for 21 regulators. People cleaning up their data get an actionable checklist rather than a privacy manifesto; the limit is that the list's accuracy can't be verified, and results depend on whether the other side cooperates.
A Claude skill for running structured personal-data erasure campaigns — 328 data brokers, OSINT tools and breach-search engines with verified privacy contacts, request templates for 13 privacy regimes…
A pronote-digest Aggregates a Pronote iCal subscription feed daily into next-day classes and homework, pushed to parents and students by email or other channels, with both a GitHub Action and a CLI. The niche is small and clear, and French school users can use it right after installing; the limitation is tight binding to the Pronote ecosystem.
Chaque soir, le planning ou les devoirs du lendemain depuis les flux iCal Pronote, par mail ou autre canal. GitHub Action et CLI.
A coffer A Rust client that natively reads Apple Passwords/iCloud Keychain on Linux, addressing the real pain of cross-platform password access with a restrained scope. The risk is reliance on Apple's private protocols, so stability and compliance need your own assessment.
Native Linux client for passwords and other credentials stored in Apple Passwords (formerly iCloud Keychain)
A ATHER-OBD-READER Reads the CAN bus directly from an Ather electric scooter's BMS, exposing cell balancing and health data the official app doesn't provide. A genuinely useful small hardware project for owners and mechanics of this niche model; limits are it covers one vehicle and requires your own wiring.
A open-source OBD data reader for Ather electric scooters. This device reads real-time CAN bus data directly from the scooter's BMS (Battery Management System), allowing owners and technicians to moni…
A DogLegs Code accompanying an IROS 2025 paper, using multi-leg-mounted IMUs for legged-robot proprioceptive state estimation, with a concrete problem and a citable method. A rare open implementation for quadruped state-estimation researchers; limited by its specific sensor layout, hard to port short-term.
[IROS 2025] DogLegs: Robust Proprioceptive State Estimation for Legged Robots Using Multiple Leg-Mounted IMUs
A cc-readback A read-only MCP server that reads session logs from ~/.claude into Claude Desktop, with secret redaction. Immediately usable for anyone running both Claude Code and Desktop, and it stays within bounds; the risk is dependence on local file formats that break when upstream changes.
Read your Claude Code sessions from Claude Desktop. A local, read-only MCP server over ~/.claude with secret redaction.
A niubigeo An AI tool generating brand visibility and competitor reports for marketers. 160 stars in a very short time show strong momentum; SaaS subscriptions or report services could commercialize, but SimilarWeb and Brandwatch compete, and data sources and privacy compliance are the risks.
Open-source AI brand visibility and competitor reports
A CVE-2026-65343 A PoC for an out-of-bounds read in iOS 26.6 AppleKeyStore that can be used to break KASLR, with the version number and specific component clearly stated. For iOS kernel security researchers this is rare substance, and whether it is worth reading depends on whether you follow this line; the biggest limitation is that it is only a PoC with no full exploit chain.
CVE-2026-65343 PoC — AppleKeyStore OOB read → KASLR defeat (iOS 26.6 / 23G71)
A superlocal Provides a self-hostable email client and gateway letting developers customize their email experience. 105 stars in two days; hosted services or an enterprise edition could commercialize, but Thunderbird and Mailspring compete, and email protocol compatibility and security maintenance are the risks.
An email client & provider gateway so you can build your best email experience.
A lan-drop A process on your LAN lets phones on the same WiFi transfer 10GB files via browser, with resumable transfers and no app install; similar to LocalSend and croc, but winning on zero dependencies and clean scope. Limited by only 12 stars, single maintainer, and unverified cross-platform details.
Run it on a machine in your home network and phones on the same WiFi can transfer huge files (10GB+) straight to it — no cloud, no app install, resumable, zero dependencies
A sim-foundry It virtualizes a whole industrial site on one computer, PLCs, cameras, sensors and all, targeting the pain of being unable to integrate until hardware arrives. The target users, industrial gateway, host software, and IoT platform developers, are clearly stated, making it a rare practical simulation tool; maturity and protocol coverage still need trial confirmation.
A single computer can virtualize an entire industrial site's devices, with no need to purchase physical hardware such as PLCs, cameras, or sensors; aimed at developers and testers of industrial contro…
A Skala A one-click Fusion 360 plugin that zooms the viewport to true 1:1 physical size on screen, a hassle anyone doing product enclosure design has hit. The function is single-purpose and the description restrained, exactly what such a small tool should be; the limitation is that it requires Fusion and manual screen DPI calibration.
One-click Fusion 360 add-in that scales the viewport to true 1:1 physical size on screen.
A AgentCapture It inverts the honeypot idea for AI agents: fingerprinting, disguised recruitment, conversational C2, and SSH session replay, a fairly novel attack-defense angle. The claim that all five mainstream agents were countered in testing is self-promotional and needs your own verification; worth reading for agent security people.
Embedded honeypot and AI Agent countermeasure platform: 23 kinds of Agent fingerprinting + functional camouflage recruitment + conversational C2 + interactive SSH honeypot and session replay; in testi…
A wisense It uses WiFi CSI from a single ESP32 for presence detection, fall and breathing monitoring, and activity classification, with no camera, cloud, or Torch at runtime. A practical alternative for privacy-sensitive settings, with boundaries clearly stated; the limit is that accuracy depends on environmental calibration and still trails commercial mmWave solutions.
WiFi CSI human sensing for Python presence, fall detection, breathing rate, activity classification, and occupancy counting from an ESP32. No camera, no cloud, no PyTorch/CUDA at runtime.
A pytrace It wraps SEGGER J-Link/J-Trace ETM instruction-level tracing, code coverage, and RTT/SWO into a Python SDK, usable today by people doing embedded timing and coverage analysis. The scope is restrained and the description accurate; the limit is dependence on commercial probe hardware, so the audience is naturally narrow.
Python SDK for SEGGER J-Link and J-Trace: ETM instruction trace, code coverage, RTT, SWO and target control
A vocalwriter A user-space PowerPC interpreter that reruns VocalWriter 2.0's singing synthesis code, plus a decoded private file format and an accessible multitrack editor. Rare for old Mac audio software and accessibility users alike; very narrow audience, but real engineering.
VocalWriter 2.0's own PowerPC synthesis code, running again -- a user-mode PPC interpreter, its file formats decoded, and an accessible multi-track singing editor built on top.
A Candela OLED burn-in protection for macOS plus external monitor brightness, contrast, and HiDPI control, a layer OLED Mac users genuinely lack. The feature set is broad and overlaps BetterDisplay in places, but its burn-in history is its own angle and worth installing to try.
Candela protects your displays from OLED burn-in and wear: panel health history, a checkup for defects, and the controls macOS leaves out for external monitors (brightness, volume, contrast, custom re…
A appstore-forge Turns app screenshots into App Store and Google Play assets in one click, processed locally for privacy with no account or upload. It targets a real developer pain point, and its 66 stars show approval; a paid standalone app is a clear path, but platform rules may change and Canva-style tools compete.
Turn raw app screenshots into store-ready App Store and Google Play assets. A local Mac app — no account, no upload, no server.
A Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x Compares cost differences across 9 AI harnesses on the same model, finding a 17x gap and hitting a real developer selection pain point. Its 68 HN points prove genuine motivation, and a paid benchmark service could monetize, but harnesses update fast, maintenance is costly, and broader evaluation systems may replace it.
A consulting-pptx-skill Generates consulting-standard PPTX decks with 38 built-in SlideSpecs and a checking pipeline, hitting a real consulting pain point. Its 63 stars show market interest, and it could be a paid Claude Code skill or enterprise knowledge-base integration, but output quality depends on the model and general agents compete.
A Claude Code skill that makes AI produce serious PPTX — slide conventions + 38 SlideSpec types + generation pipeline + machine checks
A motion-bricks.cpp Ports NVIDIA MotionBricks to GGML/C++ for neural network-generated animation usable in games and robotics. The performance optimization is technically substantial, and at 38 stars it is early but promising; it could become a game-dev library or robotics animation component, but NVIDIA provenance may raise licensing issues.
NVIDIA MotionBricks ported to GGML/C++; dynamically generates animations from keyframes using a neural network. Can be used in games or with robots
A eslint-plugin-slop An ESLint plugin that targets AI-generated junk code, addressing code-quality drift in the AI era. From well-known developer antfu, it is growing fast at 37 stars; monetization could be enterprise CI integration or paid rule sets, but rule accuracy is limited and other lint extensions may absorb it.
ESLint rules for guarding AI slops in code.
A meta-ads-skill Teaches AI coding agents to run Meta ads, covering the API, browser automation, and conversion tracking. Its 37 stars validate marketing-tech demand, and it could ship as a standalone service or plugin, but Meta's ad rules change often and results depend on model capability.
Teach your AI coding agent to run Meta ads: API setup, programmatic campaigns, browser control, and conversion tracking that's actually proven to work.
A grail Uses clingo to solve version ranges over the nixpkgs historical index, answering which commit satisfies a set of dependency constraints. Rare and on-point for the Nix ecosystem, and the author is an insider; the limits are that it only serves Nix users and the documentation is thin.
Version ranges for nixpkgs, solved by clingo over the nixpkgs-multiverse history index.
A m3e-canvas Draws Material 3 interfaces in the browser and generates vibe-coding prompts to speed Android UI prototyping. Its 61 stars show designer-community approval, and it could be a standalone tool or plugin, but it is tied to one design style, the vibe-coding trend shifts fast, and existing design tools compete.
Sketch Material 3 Expressive screens in the browser and turn them into vibe-coding prompts.
A image-prompt-reverse Provides a high-fidelity image prompt reverse-engineering skill for Codex, addressing the pain of recovering prompts from images, and its 122 stars confirm social buzz. Monetization is unclear but could be a paid plugin or workflow integration, with risks of dependence on the Codex ecosystem and unstable reverse results, competing with tools like CLIP Interrogator.
High-fidelity AI image prompt reverse-engineering skill for Codex
A ApexAgents-SkyRL-Recipe Provides a recipe for training knowledge-work agents on APE-Agents data with SkyRL and Harbor, lowering the RL training barrier, and its 81 stars show researcher interest. Monetization could be an RaaS platform, but the user base is limited, with risks around data provenance, training complexity, and competition with traditional RL frameworks.
RL-training knowledge-work agents on APEX-Agents data with SkyRL + Harbor
A ComfyUI-DLSS5-Enhancer Treats NVIDIA DLSS 5 neural rendering as a ComfyUI node to enhance images and video, one of the few attempts to repurpose new hardware capabilities, testable with an N card and runtime libraries. The author is the developer of Fully Kiosk, which adds credibility; the limitation is a strong dependency on closed-source DLLs.
NVIDIA DLSS 5 Neural Rendering (NGX feature 18) as ComfyUI nodes. Enhance image batches and video files with the real neural renderer, with optional upscaling.
A papergraph-mcp Parses arXiv/LaTeX papers into theorem dependency graphs that agents can query via MCP. Math-literature agents genuinely lack such structured upstream data, unlike the flood of RAG wrappers; the limit is that parsing quality depends on LaTeX style, and it is currently more a proof of concept.
Turn arXiv and LaTeX mathematical papers into theorem dependency graphs for AI agents through MCP.
A gflow-engine An embeddable approval workflow engine for Go that describes flows in a JSON DSL, covering all-sign and any-sign semantics plus multi-tenancy, and it is a real part of the RuleGo ecosystem rather than a shell. Suited to teams wanting embedded approvals without Camunda's baggage; the downside is deep ecosystem lock-in, and the docs and examples still need verification.
Embeddable approval workflow engine for Go, built on RuleGo — JSON DSL, full approval semantics, multi-tenant.
A agent-deck A self-hosted Android remote console plus Windows bridge letting you take over Codex / Claude Code sessions from your phone. A real need for people who leave their desk but want agents running, and a rare form; the limit is you must set up the bridge yourself and read the code before trusting the permission boundaries.
Self-hosted Android remote console and Windows bridge for Codex and Claude Code.
A umbrik A Rust implementation of the Estonian CDOC2 encrypted container, with a CLI, library, Python bindings and ID-card PKCS#11 support. It fills a real regional need and the description is honest and restrained; the limit is that its audience is almost entirely in Estonia, and it is unusable outside that ecosystem.
CDOC2 encrypted containers in Rust — the Estonian .cdoc2 format. CLI, library and Python bindings, with ID-card support via PKCS#11.
A Show HN: Running 104GB Qwen3.8-Flash-Next on 48GB Mac with at ~12 tok/s Streams experts from SSD to run 125B models on low-memory Macs. At 130 HN points momentum is strong, and it could become a local inference engine or be integrated into LLM apps. Risk: speed bottlenecks and hardware dependence.
I built slotstream, a way to run Qwen3.8-Flash-Next 4-bit on a low-memory mac starting from 16GB, a 125B parameter model that would need 100GB+ memory/RAM, thanks to expert-offloading/ssd-streaming. E…
A DreamX-Creator Enables 2K native audio-video generation, lowering the production barrier. It reached 100 stars in 0.9 days in a hot field, with strong monetization potential as SaaS or model licensing, but it competes with Sora and other big players and compute costs are high.
Democratizing Native Audio-Video Generation at 2K Resolution
A Show HN: Orthogonal – One integration for agents to discover and pay for APIs Offers unified API discovery and payment for AI agents, similar to OpenRouter. At 11 HN points it hasn't taken off, but commercial potential is large as a fee-charging platform. Risk: competition from API providers and agent frameworks, plus trust issues.
Hey HN, we're Bera and Christian, and we're building Orthogonal. Orthogonal gives AI agents one integration to discover, access, and pay for APIs. Think OpenRouter for the APIs that AI agents use. We …
A kvm-automation An MCP server letting agents drive IP-KVMs like GLKVM, PiKVM and TinyPilot, with a CLI and batch config, a rare practical "give AI hands" direction. The catch is you need the hardware first, and the products it targets are building this themselves.
Agent automation for IP-KVMs: one MCP server, CLI, fleet config, and skills driving GLKVM, PiKVM, and TinyPilot. Open-source alternative to TinyPilot Automation.
A Currents A gesture-first Android Reddit client that uses its own Client ID to sidestep the old third-party client ban. The UI is clearly polished and it installs and works, but older clients like RedReader already fill this niche.
A polished, gesture-first open-source Reddit client for Android. Bring your own Client ID.
A PC2Sonos It pushes any audio from your computer to Sonos speakers with low latency, a real need for Sonos owners who want to play local sound; the description is restrained, it does one thing, and releases include ready-made packages. The limits are a single platform, overlap with AirPlay solutions like shairport-sync, and unknown long-term maintenance.
PC audio to Sonos, in sync. Downloads: see Releases.
A GeoNeXt The official implementation of an ECCV paper that treats a video generation model as a geometry learner for 3D reconstruction, directly reproducible code for researchers in multi-view reconstruction and 3D vision. The scope is in the title with no hype; the limit is that it is pure research code ordinary developers cannot install today, and 17 stars in two days is hardly validation.
Official implementation of GeoNeXt: Video Generative Models as Geometry Learner (ECCV 2026).
A fantasy-football-draft-room A browser-based fantasy football draft assistant that gives round-by-round advice against real Sleeper drafts, with scoring-format-specific live ADP, custom rankings, and trade picks, no account or API key needed. Scope is stated modestly and clearly. Competitors are paid sites like FantasyPros; the audience is narrow, and 21 stars in a day says nothing about retention.
A fantasy football draft tool with two modes: a mock draft against a room you can tune, and an assistant that follows your real Sleeper draft pick by pick. Live ADP for your scoring format, your own r…
A Volumix Per-app volume, mute and output switching from the macOS menu bar, using native APIs without a virtual sound card. Handy for people who take calls while listening to music, though Background Music already does the same and is more mature.
Native macOS menu bar app for per-application volume control, mute, and output routing — no kernel extension, no virtual driver.
A modist A widget adding interactive distribution plots to marimo notebooks, following the wigglystuff approach and aimed at Python data analysis users; the scope is restrained and the docs clean, a tool that works once installed, limited only by serving the marimo ecosystem alone.
Interactive distribution widgets for marimo, in the style of wigglystuff
A blip Uses a Mac as a gateway to send and receive iMessage from a Linux Omarchy status bar, with group chats and blue bubbles. A must-have for Mac-plus-Linux users; BlueBubbles does the same, this is lighter but tied to a specific desktop.
iMessage in the Omarchy bar. Read, send, groups, blue dots — your Mac is the gateway, Linux is the client.
A testpilot Runs E2E tests written in natural language via a browser agent, then has an LLM judge pass or fail from recordings and screenshots, aimed at testers burdened by Playwright script maintenance. Judging by evidence is honestly stated. browser-use and Magnitude exist; the difference is the judge step. 10 stars in a day, no real project adoption yet.
Agentic E2E testing platform — natural-language test cases executed by a browser agent in real Chrome, judged by an LLM with video/screenshot evidence
A fragment-api-dev It solves Telegram Stars payment integration, providing a Python SDK and REST examples with no API key to lower the integration barrier. 148 stars in 1.1 days shows strong demand. It could become the standard library for the Telegram payment ecosystem with a clear monetization path (charging for access); the risk is reliance on unofficial APIs that may change anytime, competing with the official SDK and other payment libraries.
Telegram Stars and Premium API with GRAM/USDT payments, Python SDK, REST examples, and no API key.
A hqtui A zero-dependency TypeScript terminal UI library that creates btop-level dashboards in one step. It solves the tedium of building terminal app UIs. 92 stars in 1.8 days shows fast growth, and it could become a standard dependency in the Node/TS terminal ecosystem, monetized through sponsorship. The risk is mature competitors Ink/blessed, and performance remains unproven.
High Quality Terminal UI for TypeScript — btop-grade terminal dashboards with a one-import API, dark by default, zero runtime dependencies.
A ontology-driven-dev It drives business system development with ontology modeling, proposing a standardized three-step method to reduce requirement ambiguity. 83 stars in 1.4 days fits the AI Agent development paradigm. It could become a standard Agent development methodology, monetizing through consulting or a template marketplace; the risk is a high learning curve, competing with traditional low-code platforms.
Ontology-driven business system development skill: a three-step method of requirement exploration → ontology modeling → application building, based on seven-model ontology YAML and the code-paas techn…
A Show HN: Linux server management over SSH – written in Rust and Tauri An SSH server management interface built with Rust/Tauri; 47 points on HN shows real demand. It solves the problem of switching between multiple terminals for server management. It can be commercialized (subscription or one-time purchase), but competitors like Termius and Baota are mature. The risk is fierce competition in ops tools and the need to maintain multiple platforms.
A scientific-agent-skills An agent skill library for scientific research with 165 skills and database connections, addressing scientists' AI coding and analysis needs. It claims 190,000 users but has only 62 stars, inconsistent evidence; if real, it could become a research AI standard. Risk: unclear data authenticity, competing with discipline-specific tools.
Turn any AI agent into an AI Scientist. The #1 Agent Skills library for science, used by 190,000+ scientists worldwide. 165 ready-to-use validated skills plus 100+ scientific databases covering biolog…
A aec-geometric-bench A small benchmark set extracting geometry and symbol information from construction-drawing PDFs, with 312 drawings scored, 15 released with ground truth, and a runnable scoring script; AEC document parsing has long lacked public benchmarks and it honestly states its sample size, though the data is too small and it's more of a starting point.
AEC-Geometric-Bench: geometric and symbolic data extraction from PDF construction drawings. 312 sheets scored, 15 released with ground truth and a runnable scorer.