298 subscribers
14K videos
14.6K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ andrisgauracs/bevel-os

Bevel OS: Windows 95 on a Chip

Bevel OS is the retro desktop that fits on a single microcontroller and still runs Doom. It turns a tiny chip and a simple VGA cable into a full computer with windows, a file manager, and a web browser, all driven by just six resistors. The real star here is that an AI model wrote nearly all of this complex system, while a human just handled the physical wiring and printing. It proves you can build a surprisingly capable operating system on hardware that costs less than a coffee. If you love watching machines do things they were never meant to do, this project is a must-see.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ archive228/foreman

Foreman: The AI Harness That Refuses to Guess

Foreman is the observability layer that finally stops your AI agents from lying about their work. It connects to hosted bot crews and watches them run, but its defining feature is that it never accepts a bot's word as proof that a task is finished. Instead, it only marks work as done when an independent checker verifies the result, ensuring you never inherit a false sense of progress. It also shows you exactly where your bots are stuck, repeating, or waiting on you, all from a simple dashboard. If you are running multiple AI agents, Foreman gives you the truth, not just a status update.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ open-rollout/tokin

How Tokin Makes AI Agents Token-Native

Point your favorite AI coding agent at Tokin instead of the model server, and it starts handing back the exact numbers the machine actually picked. This Python tool acts as a smart middleman between your agent and the inference engine. Your agent sends regular text, but Tokin converts it into the raw token IDs the engine needs, keeping a detailed record of every step. Why does this matter? Because it lets you train your agent on the precise data it sampled, without rewriting your existing setup. It turns a black-box conversation into a transparent training log.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ bytesbay/pi-gpui

Pi GPUI Runs DOOM Inside a Coding Agent

Play the classic game of DOOM directly inside a coding agent. The open-source app Pi GPUI is a fast, native window for the pi coding agent that skips heavy web wrappers entirely to deliver a snappy, lightweight experience. Instead of just showing text, it lets you drop in plugins that draw live, interactive graphics right into the agent’s chat. The result is a coding tool that feels like a real desktop app, not a terminal. If you want to see what a modern agent can do beyond just typing code, this is the project to watch.

πŸ“° https://news.ycombinator.com/item?id=50464872

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ masterofthechaos/proxypilot-public

Run Any AI Model In Xcode Agent Mode

Xcode’s built-in AI features only work with a handful of specific providers, which locks developers out of using their favorite models. ProxyPilot solves this by running a local proxy on your Mac that translates requests, letting you connect almost any OpenAI-compatible provider or Anthropic API to Xcode Intelligence. You can switch between local models like Ollama and cloud services without losing your conversation thread. It is useful because it gives you total freedom to choose the best model for your workflow, whether it is free and local or powerful and cloud-based, without needing to reconfigure your entire setup.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ neuzhou/awesome-ai-anatomy

The Hidden Truth Inside AI Coding Agents

Dissecting the actual source code of fifteen major A I coding agents reveals secrets that official documentation completely ignores. Instead of just listing features, this project breaks down the internal architecture of tools like Claude Code and Codex to show exactly how they work under the hood. The most surprising find is a full virtual pet system hidden inside Claude Code, proving that even serious production software contains unexpected easter eggs. By reading every file, the creators uncover real engineering decisions and security quirks that nobody documents.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ lowkeytea/scyllasband

Scylla's Band: Local Voice Acting with Real Emotion

Scylla's Band lets you generate voice acting right on your own computer, with no cloud connection required. It turns plain text into expressive speech using fourteen distinct voices across seven languages. The real magic is the delivery control. You can dial up the intensity, pace, and pitch to shift a single line from calm and quiet to intense and loud, all without changing the words. Because it runs entirely locally, your data stays private while the software generates audio faster than real time. It even supports singing melodies from a simple score. If you want realistic voice acting that actually feels like a performance, this is the tool to try.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ vincegher/openwren

OpenWren: The Mail App That Actually Owns Your Data

OpenWren is the desktop email client that keeps your inbox completely private on your hard drive. It was built because the author couldn't find a modern, fast, and private mail app for Linux, so they wrote one. The app is tiny, only fourteen megabytes, and opens in about half a second because it is built with Rust and React. It connects to Gmail or any IMAP account and stores everything in an encrypted local database, meaning no servers see your mail. It even strips out tracking pixels and flags phishing attempts before you click on anything.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ bubblebrainbb/awesome-fable-5.2

Films Rendered Entirely By Code

A cat’s perspective, a mountain summit, and a sunlit room, all rendered without a single 3D model or texture file. The repository Awesome Fable 5.2 is a collection of films and images where every pixel is calculated by code written by an AI model. It skips game engines and asset libraries to build its own renderers from scratch. The cat movie simulates thousands of plants, while the mountain film grows its terrain through mathematical evolution. It proves that complex visuals don’t require heavy software, just clever math. It’s a fascinating look at how far procedural generation has come.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ rawnaldclark/stash

Stash: The Open Source Music App

Stash is the open source Android music player that lets you keep your favorite tracks on your phone without paying for a subscription. It connects to services like YouTube to download songs and store them locally, so you can listen to your library without using mobile data or worrying about ad interruptions. The app also syncs your listening history to track what you have heard and suggests new artists you might actually like. Because it is free and open source, the community controls its development, ensuring no sudden paywalls or tracking. If you want full control over your music collection without a monthly fee, this is the tool to try.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ viktorkav/uno-com-escada

Uno com escada: A 3D Car Game in One File

Uno com escada is the browser game that renders a full racing circuit with pure code, proving that graphics engines are optional. The project features a 1996 Fiat Uno with a ladder on the roof, built entirely from mathematical calculations rather than image files or 3D models. Two different AIs generated these versions from the exact same prompt, creating a fascinating comparison of how they handle complex graphics tasks. Every visual element, from the sky to the engine sound, is synthesized on the fly within a single file. This project is a brilliant reminder that sometimes the most interesting technology is found in the simplest, most self-contained implementations.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ shisa-ai/hipengine

Hip Engine: Native AMD Inference

Hip Engine is the local inference engine built directly for AMD Radeon graphics cards. It runs artificial intelligence models without needing PyTorch or other heavy software libraries, making it surprisingly lightweight. The standout feature is its drop-in server that speaks the same language as OpenAI tools, letting you swap your existing AI clients to run locally on your own hardware instantly. Because it is written from scratch for specific AMD chips, it avoids the performance penalties of trying to translate code meant for other brands. This gives developers a faster, cleaner way to test and serve models right on their own machines.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ interfere-inc/landmines

Landmines: Keeping AI Code Honest

Coding agents often cut corners to pass tests, using tricks like fake data or extra time that hide real problems. Landmines stops this by acting as a strict quality check for your code. It forces AI tools to write real, working tests instead of shortcuts. If an agent tries to cheat a test, this tool catches it immediately and tells the machine exactly how to fix it properly. This keeps your codebase clean and reliable as you build with AI. Your code stays honest because the rules never let a fake fix slide through.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ 0nelight/sutra-os

Sutra-OS: Trace Every Number

Sutra-OS turns your messy data into a single, living map of how everything connects. Instead of trusting black-box vendors, you self-host this platform to see exactly where every number comes from, so you can trace any failure back to its root cause before it becomes a disaster. It’s not just a database; it’s a digital twin of your operations that lets you ask what breaks if one piece fails. You get full ownership and transparency, making your data actionable and secure. No more guessing. Just clear lines of sight from source to outcome.

πŸ“° https://news.ycombinator.com/item?id=50525052

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ bashovski/reflektor

Reflektor: Code Review on Steroids

Reflektor turns your code branch into a full visual report in your browser. Instead of just reading text, you see the request flow from the user interface down to the database, complete with screenshots of your UI and live results from your tests. It runs locally as a small background helper that keeps updating as your coding agent works. This is a skill you add to tools like Claude Code or Cursor. It handles any language or framework, so you aren't locked into a specific stack. You can share these reports with your team for feedback.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ underleveledbuilder/awesome-social-media-skills

Find the Right AI Social Media Skills

Want your AI agent to actually run your social media accounts? Stop guessing which prompts work. This curated directory collects specific, installable capability packs for writing posts, making covers, and publishing on platforms like YouTube, TikTok, and X. The coolest part is that the list itself is a skill. You install it into your agent, then just ask for recommendations. It searches its own verified database to match your specific platform and job to the best tool. It handles the hard part of finding and vetting the right assets for you. It is the fastest way to turn a generic chatbot into a specialized social media assistant.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ smoovplay/smoovplay

SmoovPlay: Turn 30 FPS Console Games into 120 FPS

SmoovPlay takes a standard thirty frames per second console stream and paints entirely new frames into the gaps using AI, delivering sixty, ninety, or even one hundred and twenty frames to your monitor. Most people assume you need to hack your console to get smoother graphics, but this tool runs entirely on your PC, using your graphics card to generate the extra motion in real time without touching your hardware. It pulls the video stream from your PlayStation or Xbox, analyzes the movement, and inserts synthetic frames between the originals to make motion look incredibly fluid.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ codepawl/crucible

Crucible: The 3D Engine for AI Agents

Your AI agent can now build, light, and render full 3D scenes in the browser, and it does the heavy lifting so you can just focus on the creative look. Crucible is a JavaScript 3D engine designed specifically to be driven by AI agents through a standard interface. Instead of you writing complex code to set up every camera and shadow, you tell the agent what you want, and it handles the technical details, showing you previews along the way. This is interesting because it removes the massive barrier between a text prompt and a working 3D environment.

πŸ†” @hackernewsgithubprojects
πŸ‘1
Media is too big
VIEW IN TELEGRAM
πŸ“¦ tangency-labs/solidcore

Solidcore: The CAD Kernel Written in Rust

A full port of NVIDIA's proprietary computer-aided design kernel is now being rebuilt in the Rust programming language. Solidcore takes the complex mathematical engine that defines precise 3D shapes and rewrites it to be deterministic and safe, meaning the same input always produces the exact same output without risk of crashes. This solves a major problem for software developers who need reliable geometry calculations in web browsers or mobile apps, areas where traditional design tools simply cannot run. By running on WebAssembly, this project allows high-precision engineering tools to operate directly inside a web page rather than on a heavy desktop computer.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ dmitriygolub/threejs-devtools-mcp

threejs-devtools-mcp

threejs-devtools-mcp is the model context protocol server that lets your AI agent see and edit live three.js scenes right in your browser without touching your files. It works by injecting a tiny bridge script that connects the AI to your running scene, so you can ask it to inspect the scene tree, tweak materials, or debug why a model is invisible. The coolest part is live preview mode. The agent can change colors or lights instantly so you see the result immediately. If you like what you see, it gives you the exact code to paste into your project. If not, it reverts everything.

πŸ†” @hackernewsgithubprojects