📦 soniqo/speech-android

speech-android

You can run a complete, private voice assistant entirely offline on your phone using just over one gigabyte of memory. The speech-android project makes this possible by packing a full audio processing suite into a local Android SDK. By combining lightweight speech-to-text, text-to-speech, voice activity detection, and noise cancellation, it processes everything on your device without sending a single byte of data to the cloud. It even hooks directly into Android's system-level speech services so other apps can use your local pipeline. It is a brilliant way to build ultra-private, fast voice applications.

🆔 @hackernewsgithubprojects

📦 greyhavenhq/greywall

Greywall

Lock down your filesystem, network, and system calls without the overhead of heavy containers. This tool lets you execute untrusted commands and AI coding agents inside a native, deny-by-default sandbox that restricts access to only your working directory, keeping sensitive folders like your SSH keys and environmental secrets completely secure. It acts as a lightweight, kernel-enforced barrier for Linux and macOS that blocks dangerous operations, routes network traffic through a transparent proxy with a live dashboard, and even lets you run a passive observation mode to map out exactly what a tool is trying to touch before you restrict it.

🆔 @hackernewsgithubprojects

📦 hoainho/img2threejs

img2threejs

Most image-to-3D tools spit out messy, massive mesh files that are absolute nightmares to edit, but img2threejs does something completely different. You feed it a single reference image of an object or character, and it writes clean, highly structured TypeScript that builds a procedural, animation-ready Three.js model from scratch using code. It guides an AI agent through a locked, step-by-step sculpting pipeline, rendering the progress and self-correcting until the code matches the visual details. You get a lightweight, interactive 3D scene built from clean primitives and math, perfect for fast-loading web applications.

🆔 @hackernewsgithubprojects

📦 unable12/codebase-posters

Codebase Posters

Turn your git history into beautiful generative art that you can actually print and hang on your wall. Inside any repository, running a single command opens a local exhibition with nineteen distinct visual recipes that paint themselves live on your screen. The software translates your commit metadata into elegant canvases, depicting your midnight coding sessions as clusters of stars, or years of additions and deletions as gentle waves of color. Everything compiles completely on your machine with zero external dependencies, letting you customize colors and layouts before exporting a high-resolution, print-ready file to bring your development journey to life.

📰 https://news.ycombinator.com/item?id=48885500

🆔 @hackernewsgithubprojects

📦 itsplk/ps5-y2jb-autoloader

PS5 Y2JB Autoloader

The automated modding utility that finally takes the tedious manual work out of starting up an exploited PlayStation 5 console. Instead of manually pushing payloads to your device every single time you boot, this tool acts as a clever automated chain loader that runs directly through the console's built-in YouTube app. It automatically handles the initial kernel exploit phase and immediately fires up your custom payloads. If no manual setup is found, it automatically boots into a friendly web-based payload manager so you can configure everything straight from your web browser. It is the ultimate hands-off startup routine your PlayStation 5 deserves.

📰 https://news.ycombinator.com/item?id=48967115

🆔 @hackernewsgithubprojects

📦 lopopolo/harness-engineering

Harness Engineering

Improve the output of your AI coding agents by shaping the environment around them instead of constantly switching models. The harness-engineering repository is a curated field guide and context bundle designed to feed your agents the exact business constraints, local rules, and repository knowledge they need to do their jobs. By treating the AI model as a stable, fixed worker, this project helps you feed agents just-in-time context, legible tools, and clear authority guidelines so they can operate real systems and verify their own work. It is a smart, practical approach to making AI agents more effective by giving them better local instructions.

📰 https://news.ycombinator.com/item?id=48963483

🆔 @hackernewsgithubprojects

📦 prakhar1114/ai_mime

ai_mime

ai_mime is the macOS automation harness that finally turns your raw screen recordings into reliable, reusable scripts. Instead of fragile pixel-matching tools, you simply demonstrate a task once. The app analyzes your actions and screenshots, builds a coordinate-free workflow schema, and writes a deterministic Python script or browser automation to handle the work. If a layout changes or a step fails, a built-in assistant automatically analyzes the error, patches the code, and heals the script. It brings the perfect balance of reliable local code and intelligent self-healing, giving you robust desktop automations that actually last.

📰 https://news.ycombinator.com/item?id=48480492

🆔 @hackernewsgithubprojects

📦 nossa-y/activity-frames

activity-frames

Give your local AI agents eyes by turning raw screen recordings into a structured, highly organized timeline of your day. While language models can read your code and search the web, they typically start every chat completely blind to your actual work context. This project solves that by locally tracking your screen snapshots and compiling those thousands of chaotic points into clean, readable activity blocks. It maps out your active minutes, the specific websites you visited, and your input volume, then serves it all securely through the Model Context Protocol. You get a private, token-free memory feed that helps your agent understand exactly what you did, without any cloud uploads...

📰 https://news.ycombinator.com/item?id=48824758

🆔 @hackernewsgithubprojects

📦 eblanshey/historyworkbench

History Workbench

History Workbench is the FreeCAD add-on that brings intuitive, visual version control directly to your 3D modeling workflow. If you want to track changes without wrestling with complex developer tools, this workbench integrates Git under the hood while keeping everything in plain CAD terms. You can visually compare modified geometry with color-coded 3D views, inspect exact dimension changes side by side, and check the document tree to ensure a quick edit didn't break another part of your design. It is the perfect way to review your iterations and design with total confidence.

📰 https://news.ycombinator.com/item?id=48054923

🆔 @hackernewsgithubprojects

📦 younesfdj/gitfut

GitFut

GitFut is a fun web application that turns your real GitHub stats into a customizable, World-Cup-style player card. It reads your live development metrics straight from GitHub to automatically calculate your overall player rating up to ninety-nine, along with specific attributes like pace, shooting, and passing. Your commit frequency determines your speed, your star counts rate your shooting power, and pull requests evaluate your passing skill. It even assigns you a custom playstyle and player tier. You can customize your country flag and easily embed your live-updating card into your profile page to show off your coding stats.

🆔 @hackernewsgithubprojects

📦 tugcantopaloglu/openclaw-dashboard

OpenClaw Dashboard

openclaw-dashboard is the secure monitoring hub that finally makes managing your AI agents feel like running a professional command center. If you are experimenting with autonomous assistants, you know how stressful it is to lose track of what they are actually doing, how much money they are spending, or what they are writing in their memory files. This lightweight Node.js interface brings order to the chaos by giving you a beautiful, live feed of your agent's thoughts, real-time API cost tracking, and secure system controls under one roof. It runs locally with built-in multi-factor authentication, so your agent's workspace stays completely private while you easily monitor their daily progress.

🆔 @hackernewsgithubprojects

📦 wildminder/awesome-ltx2

Awesome LTX-2

Awesome-ltx2 is the open-source hub that finally organizes the best tools and workflows for the LTX-2 video generator. This project brings together all of the top models, text encoders, and custom LoRAs in one place so you can stop wrestling with messy, fragmented setups. The coolest thing it offers is a set of master guides that teach you exactly how to write cinematic prompts for perfect camera control, like pulling a flawless rack focus without your characters turning into melting pixel soup. If you want to create stunning, consistent AI videos with clean camera motion, grab these workflows and start directing like a pro.

🆔 @hackernewsgithubprojects

📦 maxmiksa/auto-company

Auto Company

Run an entire software business on your computer completely on autopilot. This project lets you spin up a continuous background loop where fourteen specialized AI agents work together as a virtual startup twenty-four seven. Instead of generic prompts, these agents act as virtual experts modeled after real-world legends, like a CEO with a Jeff Bezos mindset and a developer thinking like the creator of Ruby on Rails. They dynamically form small squads to brainstorm ideas, run financial pre-mortems, write real code, and deploy products without you lifting a finger. You can easily steer their direction anytime just by editing a single markdown text file.

🆔 @hackernewsgithubprojects

📦 agentichealthai/awesome-ai-agents-for-healthcare

Awesome AI Agents for Healthcare

AI is already quietly reading your x-rays, predicting new drugs, and rewriting the messy medical notes your doctor barely has time to type. To see how fast this is moving, you have to check out awesome-ai-agents-for-healthcare. It is a massive, carefully organized collection of research papers and real-world tools showing how autonomous software is changing medicine. Instead of just searching a database, these smart digital assistants can coordinate with each other to solve complex clinical problems. It is the perfect cheat sheet to see where medicine is heading, so go check it out.

🆔 @hackernewsgithubprojects

📦 ognjengt/founder-skills

Founder Skills

The open-source project founder-skills allows you to run over twenty pre-packaged business and marketing skills directly inside Claude Code. This clever tool installs a collection of highly specific prompts and reference guides that instantly transform your terminal-based AI assistant into a specialized team of startup experts. Instead of writing custom prompts from scratch, you can simply type a quick command to trigger professional copywriting frameworks, landing page optimization audits, or complete cold outreach sequences tailored to your specific brand context. It is the perfect way to get expert marketing, pricing, and product strategies directly in your development workflow.

🆔 @hackernewsgithubprojects

📦 anthropics/claude-ai-mcp

claude-ai-mcp

claude-ai-mcp is the central communication hub that finally bridges the gap between Claude's web interface and your custom developer tools. Instead of hosting complex code, this repository acts as a direct line of communication for developers to track updates, report bugs, and shape how Claude connects to external data sources. The most interesting thing here is how it lets you follow protocol shifts in real time, like the transition toward faster streaming connections and streamlined security handshakes. It is a vital workspace for anyone building tools to make artificial intelligence more capable in everyday workflows, giving you a clear window into how the platform interacts with external servers.

🆔 @hackernewsgithubprojects

📦 themaxmur/rs-key

RS-Key

RS-Key is the open-source firmware that turns a standard Raspberry Pi RP2350 microchip into a highly versatile hardware security key. Written entirely in memory-safe Rust with zero external heap allocation, it transforms a basic development board into a physical authenticator capable of storing your passkeys, managing two-factor authentication, and acting as a smart card for secure encryption. What makes it truly fascinating is its support for experimental post-quantum cryptographic schemes, preparing your physical security for future computing eras on tiny, affordable hardware. It is a brilliant, open playground for security enthusiasts looking to explore hardware authentication from the ground up.

📰 https://news.ycombinator.com/item?id=48959214

🆔 @hackernewsgithubprojects

📦 jolt-lang/jolt

Jolt

Compile Clojure code directly into fast, self-contained native binaries that run completely free of the JVM. By taking Clojure source and compiling it down to clean Scheme to run on the highly optimized Chez Scheme engine, jolt bypasses Java entirely. It solves the massive startup lag and heavy memory footprint of traditional Clojure setups by letting you build standalone, single-file executables with full tree-shaking that drops unused code automatically. You get persistent collections, atoms, and the full standard library in a tiny native package. Check it out and start compiling lightning-fast native Clojure apps today.

📰 https://news.ycombinator.com/item?id=48414519

🆔 @hackernewsgithubprojects

📦 taesiri/fire

Fire Replica Lab

This project runs highly detailed physics-based fire simulations right inside your browser using raw WebGPU and WebGL2 power. It acts as a local graphics laboratory that replicates complex academic research papers on combustion, soot formation, and water suppression. Instead of relying on pre-rendered video clips, it calculates fluid dynamics, heat distribution, and chemical reactions live on your graphics card. You can customize the fuel type, spray water to watch the flames realistically contract, or trigger a deterministic offline renderer to generate perfect 4K movies frame-by-frame. It is a fantastic playground for anyone curious about how high-end virtual fire actually works.

🆔 @hackernewsgithubprojects

📦 rindig/icm-architect

ICM Architect

Turn your complex workflows and messy folders into organized pipelines that artificial intelligence can run on its own. With icm-architect, you can structure any process, idea, or vault into an intuitive layout where folders carry the sequencing and markdown files hold the state. By separating your stable rules from your active files, a single AI agent can easily navigate your project step-by-step, replacing bulky multi-agent frameworks with plain, simple folders. It is a brilliant way to make your workflows self-documenting and perfectly clear. Give it a try to build smarter, human-readable workspaces that practically run themselves.

🆔 @hackernewsgithubprojects