📦 danzer1xxxxchan/h3-world

H3-World: The First Interactive World Model

H3-World is the interactive world model that turns simple keyboard inputs into controllable video. Most AI video generators create static clips, but this system uses a large language model to interpret your movements and generate the next frames of a world that reacts to you. By learning from thousands of gameplay clips, it maps actions like moving forward or panning the camera to specific visual outcomes. It’s a fascinating step toward playable digital environments, showing how language understanding can drive real-time world simulation. This is a major shift from passive video generation to active, user-controlled experiences.

🆔 @hackernewsgithubprojects

📦 akalanka1337/neuratube

NeuraTube: The AI Tool That Reads YouTube's Hidden Data

NeuraTube is the browser extension that stops guessing at YouTube metadata by reading the actual data YouTube sends to your browser. Most tools just look at what is visible on the screen, which means they miss hidden tags and cut-off descriptions. NeuraTube intercepts the raw information stream from YouTube's own servers, so it sees the full picture exactly as the platform stores it. This lets it generate titles, chapters, and comments that are grounded in real facts rather than approximations. Because it runs locally with your own API key, there is no server in the middle and no data tracking.

🆔 @hackernewsgithubprojects

📦 shadcn-ui/cn

cn: The 30x Faster Tailwind Class Merger

A tiny library called cn just made the standard tools for cleaning up CSS class names thirty times faster. When you build web components, you often have conflicting style rules, like a default color and a specific active color, and you need code to decide which one wins. Previously, developers used two separate libraries, clsx and tailwind-merge, to handle this. The new cn engine combines both jobs into a single, tiny tool that does the exact same work, but it runs the heavy calculations once at build time rather than every time your app draws the screen.

🆔 @hackernewsgithubprojects

📦 khromov/gpuix-svelte

Svelte Apps That Are Not Websites

Gpuix Svelte is the Svelte renderer that turns your components into native desktop windows without a browser. Instead of loading your interface into a web view, this library sends your Svelte code directly to the graphics card using the same engine that powers Zed, the fast code editor. You get the familiar syntax you already know, but the result is a real application on your desktop. It is not a web page wrapped in a window; it is a native app that feels like a desktop tool. This is a new way to build software for Mac, Windows, and Linux. No more loading screens or browser overhead.

📰 https://news.ycombinator.com/item?id=49540580

🆔 @hackernewsgithubprojects

📦 facebookresearch/mhr

Meta's New Human Body Model

Sculpt a fully articulated digital human that moves with realistic muscle deformation. Momentum Human Rig, created by Meta, generates a detailed three-dimensional body from a handful of simple numbers. You control the skeleton, the body shape, and even specific facial expressions like a raised eyebrow or a pursed lip. The coolest part is the neural network that corrects how muscles swell and shift as the joints bend, making the motion look natural rather than robotic. Since it runs on standard machine learning tools, developers can easily integrate it into their graphics pipelines. It is a powerful new standard for creating lifelike characters in games and simulations.

🆔 @hackernewsgithubprojects

📦 zvec-ai/zvec-grep

Zvec Grep: Search by Meaning, Not Just Keywords

Zvec Grep turns your local files into a searchable knowledge base that understands meaning, not just keywords. Instead of forcing you to guess the exact spelling of a function or variable, it scans your entire workspace once and builds an index that links code structure with plain text. This means you can ask natural language questions and get back the specific lines of code or documents that actually answer them. Because it runs locally, your data stays on your machine, making it safe for sensitive projects. For developers, this cuts down the time spent hunting for context, and for AI agents, it provides precise evidence to work with.

📰 https://news.ycombinator.com/item?id=49541572

🆔 @hackernewsgithubprojects

📦 crabbuild/crab

Crab: Git for Big Files

Crab turns your cloud storage bucket into a Git repository for heavy files. Instead of clogging up your code history with massive datasets or AI models, this tool keeps tiny placeholder files inside Git while the real, huge data lives safely in your own Amazon S3 or Google Cloud Storage bucket. The standout feature is smart deduplication, which breaks files into unique chunks so you only upload the parts that actually changed between versions. This means you get full version control for multi-gigabyte assets without the lag or storage bloat of traditional methods.

📰 https://news.ycombinator.com/item?id=49527786

🆔 @hackernewsgithubprojects

📦 doordash-oss/agentic-orchestrator

Agentic Orchestrator: Parallelize Your AI Coding Workflow

Agentic Orchestrator is the AI workflow orchestrator that finally stops you from babysitting every single AI prompt. Instead of manually copying context, running tests, and fixing broken plans, you just describe the feature once and let the system handle the heavy lifting. It spins up isolated workspaces to research, design, and code in parallel, meaning you can supervise multiple big projects at once without mixing up your state. You step in only when a real decision matters, while the software builds a solid knowledge base and runs strict quality checks before you even see the code. It turns a vague idea into a clean, reviewable pull request.

📰 https://news.ycombinator.com/item?id=48727448

🆔 @hackernewsgithubprojects

📦 continuum-ai-corp/orcareplay

OrcaReplay: Time Travel for AI Agents

OrcaReplay lets you record any coding agent and replay it offline with zero cost, letting you test if a different model would have fixed a bug. This solves the problem of chasing unpredictable failures by giving you a perfect, repeatable copy of the run. It works by standing in front of the model to capture every step, so you can pause at any point and swap the brain without touching your code. You finally know exactly why your agent deleted a file instead of guessing in the dark. If you want to debug with confidence, this is the tool that turns chaos into a clear, editable timeline.

🆔 @hackernewsgithubprojects

📦 nominal-io/instro

Instro: One Codebase for All Your Lab Hardware

Instro lets you write a single Python script that controls power supplies, multimeters, and oscilloscopes from different manufacturers as if they were the same device. Instead of learning a unique command set for every piece of lab equipment, you use one unified, typed interface to turn them on, set voltages, and read measurements. It solves the tedious problem of juggling vendor-specific drivers by wrapping them in a consistent structure, so your test code stays clean and portable. If you build hardware, this library saves you from rewriting your entire test suite every time you switch brands.

📰 https://news.ycombinator.com/item?id=49542921

🆔 @hackernewsgithubprojects

📦 mercor-intelligence/apexagents-skyrl-recipe

How Apex Agents SkyRL Recipe Trains Real Work AIs

Take a massive language model and teach it to actually do professional work without breaking a sweat. That is exactly what the Apex Agents SkyRL Recipe achieves by turning open source tools into a complete training loop. It lets you take a standard model and train it to handle complex tasks like analyzing documents or running simulations inside secure digital sandboxes. The coolest part is that it uses an LLM judge to grade the agent on real results, giving it a taste of success or failure. You get a clear blueprint for building agents that can navigate messy, real world environments.

🆔 @hackernewsgithubprojects

📦 nourmtir0722/paperlab

Paperlab Makes Digital Paper Feel Real

Make your web pages feel like physical objects, not just flat screens. Paperlab turns a simple code snippet into a three-dimensional sheet of paper that actually bends, folds, and ripples in real time. It is not a video or a clever trick; the text and images are painted directly onto the mesh, so they curl with the material. You can watch a receipt unroll or a letter fold itself, all while keeping the code simple. It is a fascinating way to add tangible, tactile depth to digital interfaces without heavy assets. If you want your designs to have texture and weight, this is the tool to watch.

🆔 @hackernewsgithubprojects

📦 hibiscusconsulting/nectar-lang

Nectar: The Web Without JavaScript

Write web apps that run entirely in Web Assembly, completely bypassing the JavaScript engine. Nectar is a programming language that compiles your code directly into a tiny, safe binary, eliminating the need for virtual DOM diffing or garbage collection. Because the logic runs on machine code, updating a specific part of the screen takes constant time, making interfaces feel instantly responsive without the usual browser overhead. This approach solves the slow performance and memory leaks typical of modern web frameworks by keeping your application logic isolated and fast. It is a fascinating look at a future where the web is faster, safer, and lighter than ever before.

📰 https://news.ycombinator.com/item?id=48869006

🆔 @hackernewsgithubprojects

📦 zestfulpulse/ios-app-store-submit

I Ouch, Apple's App Review Is Now a Skill for Your AI

iOS App Store Submit is the release intelligence tool that finally stops you from guessing if your app is ready to ship. It plugs into your coding assistant to scan your project for privacy gaps, design flaws, and guideline violations before you ever click upload. If Apple rejects you, it maps the rejection to a specific fix, verifies the change, and drafts your reply, ensuring you never claim a problem is solved until the proof is actually there. It is less about speed and more about making sure your app is genuinely safe to submit.

🆔 @hackernewsgithubprojects

📦 wanshuiyin/hero-anti-overdefense

HERO: Stopping AI Over-Defense

When you ask a coding agent for a simple feature, it often returns a fortress of unnecessary checks instead of the actual code. The project called HERO solves this by giving your AI tool a plain-text contract that stops four specific habits: adding useless hashes, guarding against impossible edge cases, replacing judgment with rigid checklists, and building complex scaffolding for features nobody asked for. This repository contains a short block of rules you paste directly into your agent’s configuration file, along with a catalogue of real examples showing exactly what went wrong. It works with major tools like Claude Code and Cursor without any installation.

🆔 @hackernewsgithubprojects

📦 dutkiewiczmaciej/maklock

MakLock: Lock Apps with Touch ID

Lock any application on your Mac using your fingerprint or Apple Watch. MakLock is a free, open-source menu bar app that covers your screen with a blur whenever you try to open a protected program, forcing you to authenticate before seeing anything. It works better than paid App Store alternatives because it can detect app switches, not just launches, and uses a safe overlay that never freezes your computer. This means no one can peek at your private apps by simply switching windows. The project also includes a panic key to unlock everything instantly if you get stuck.

📰 https://news.ycombinator.com/item?id=47794248

🆔 @hackernewsgithubprojects

📦 human-agent-society/reef

Reef: The AI That Gets Smarter

Reef is the learning backend that makes your AI agents improve themselves automatically. Normally, you train a model once, deploy it, and hope it stays useful, but Reef changes that by constantly watching how your agent behaves in the real world. It takes that live feedback, figures out what went well or wrong, and quietly updates the agent's skills or internal logic while it is still running. You do not need to restart anything or manage complex data pipelines; the system just keeps getting better the more you use it. It is like having an AI that actually learns from its mistakes instead of repeating them.

🆔 @hackernewsgithubprojects

📦 chenglou/sotaku

Sotaku: The Tiny Sudoku Solver

Sotaku is the tiny neural network that solves almost every Sudoku puzzle ever created. It is a model with fewer than one million parameters, which is remarkably small for this task. The secret is its ability to think in loops. While training only sees sixteen steps at a time, the model can repeat its reasoning process over a thousand times during inference. This allows it to slowly refine its answers until the grid is perfect. It achieved over ninety-nine percent accuracy on a massive benchmark of twenty-five thousand hard puzzles. Instead of memorizing solutions, it learns the logical structure of the grid itself.

🆔 @hackernewsgithubprojects

📦 radixark/miles

Miles: The RL Framework for Massive Models

Miles is the reinforcement learning framework that lets massive AI models learn without crashing. It connects the generation and training processes so they run at the same speed, which stops huge jobs from falling apart. The standout feature is fault tolerance. If a piece of hardware fails during a run, Miles fixes it and keeps going right where it left off. There is no pause and no restart. This makes training frontier models far more reliable. It is a powerful tool for anyone building serious AI. If you are training large models, this is the stability layer you need.

🆔 @hackernewsgithubprojects

📦 player-yn/pawwork_zhuazhua

Paw Work: The Chrome Extension That Turns Web Pages Into Real Files

Paw Work is the Chrome extension that turns the text and images you see on a web page into actual, editable office files. It solves the annoying problem of copy-pasting messy content from a browser into a blank document by letting you highlight exactly what you need and describe the result you want in plain language. You point at a product table, ask for a comparison sheet, and it builds a live spreadsheet right there, ready for you to tweak. The whole thing runs locally on your machine, so your data stays private and you control the AI keys.

🆔 @hackernewsgithubprojects

📦 superdesigndev/treg

Treg: The Universal Key for Agent Tools

Treg acts as a universal gateway for artificial intelligence agents, solving the messy problem of connecting them to dozens of different data sources. Instead of every agent needing its own unique keys and code for each service, Treg provides one shared connection point. This means your AI can access tools like web scraping, ad tracking, and business data through a single, standardized setup. It removes the friction of managing scattered credentials, making it easy for agents to find and use the information they need. Think of it as a master key that simplifies how smart programs interact with the outside world.

🆔 @hackernewsgithubprojects