📦 mshumer/claude-of-duty

Claude of Duty

An entire multi-layered first-person shooter with over fifty-five thousand lines of code was built inside a web browser from a single prompt, using absolutely zero external art assets. Called claude-of-duty, this project generates every single texture, 3D mesh, character animation, and sound effect procedurally from pure code at load time. A fleet of coordinated artificial intelligence agents built the entire game, which features a custom physics engine, synthesized audio, and intelligent enemies that navigate and take cover. It is a fascinating look at what happens when automated developers are pushed to build a complex, fully self-contained three-dimensional world entirely from scratch.

🆔 @hackernewsgithubprojects

📦 ilbertt/bun-sqlgen

Bun-Sqlgen: Type-Safe SQL Without the ORM

You can now write raw SQL in your Bun applications with complete type safety, completely skipping the heavy object relational mapper. A clever tool called bun-sqlgen scans your source code, runs your migrations against an in-memory database, and automatically generates TypeScript types for every single query. Because it uses an in-memory engine, you do not even need a running database to generate your types. You just tag your queries, run the generator, and instantly get compile-time errors for any broken database calls. It is the perfect middle ground for anyone who loves writing raw SQL but misses the safety nets of a massive framework.

📰 https://news.ycombinator.com/item?id=48645390

🆔 @hackernewsgithubprojects

📦 marcossete/awesome-free-ai-books

Awesome Free AI Books

You can read the definitive textbooks on artificial intelligence and machine learning written by the world's leading experts completely for free and entirely legally. The GitHub repository called awesome free ai books acts as a curated index pointing directly to official downloads and author-hosted websites. Instead of hosting pirated files, it connects you to authorized releases of legendary resources covering deep learning, natural language processing, and computer vision. Because it links only to official sources, you are guaranteed to get the latest updates directly from the authors. It is the ultimate library for upgrading your skills without spending a dime.

🆔 @hackernewsgithubprojects

📦 bottle101/forceband

ForceBand

ForceBand is the low-cost wristband system that translates muscle signals directly into precise finger-by-finger force predictions. By combining a simple wearable sensor band with a smart transformer model, this technology reads the tiny electrical signals in your forearm to estimate exactly how hard each of your five fingers is pressing. This is incredibly useful for robotics and virtual reality because it lets machines understand physical touch and effort without needing heavy, expensive sensors on the fingertips themselves. Ultimately, it bridges the gap between human muscle movement and robotic control with just a lightweight band around the wrist.

🆔 @hackernewsgithubprojects

📦 prismml-eng/mlx

MLX

Run massive machine learning models right on your Mac without the usual memory headache. The mlx framework brings native, high-performance array operations straight to Apple silicon. Instead of constantly copying heavy data back and forth between your processor and graphics card, this library uses unified memory so both chips work on the exact same data instantly. It feels just like using familiar tools like NumPy or PyTorch, but it is built from the ground up for Mac hardware. Now you can easily train or run complex language models and image generators right on your laptop.

🆔 @hackernewsgithubprojects

📦 gary23w/nl-veil

nl-veil

The multi-agent orchestrator that finally runs an entire hive mind of artificial intelligence from a single local binary. Instead of dealing with heavy frameworks, clunky setups, and endless cloud subscriptions, nl-veil packages an entire server, desktop dashboard, and command line tool into one lightning-fast executable built with Zig. When you give it a task, a swarm of specialized local agents spins up to research, build, and debug together, while a single core consciousness integrates their findings into one clear, cohesive response. It is the ultimate way to run a smart, self-learning network of agents right on your own machine.

🆔 @hackernewsgithubprojects

📦 lightseekorg/smg

Shepherd Model Gateway

Route large-scale artificial intelligence traffic like a pro with a lightweight, high-performance tool built in Rust. It is called smg, and it acts as a unified traffic controller that sits between your applications and your AI models. Instead of tearing your hair out managing different APIs for hosting engines like vLLM, SGLang, or cloud providers, this tool gives you one single endpoint that is fully compatible with standard APIs. Even better, it uses smart routing that knows the exact cache state of your graphics processing units, sending requests where memory is already warmed up to drastically cut down on wasted processing. Grab it via Docker or cargo, point it at...

📰 https://news.ycombinator.com/item?id=49063132

🆔 @hackernewsgithubprojects

📦 vercel-labs/scriptc

scriptc

Compile your standard TypeScript files directly into tiny, self-contained native executables with practically instant startup times. This clever tool parses and typechecks your ordinary TypeScript code, converting it to machine code without needing Node or a bloated JavaScript engine in the final bundle. If you hit complex dynamic features or external packages, it automatically embeds an incredibly lightweight engine to handle the heavy lifting while validating your types at runtime. It means you get the developer experience of TypeScript with the raw speed and minimal memory footprint of a native application. Check out scriptc to start building incredibly fast binaries today.

📰 https://news.ycombinator.com/item?id=49063175

🆔 @hackernewsgithubprojects

📦 sentient-os-labs/sentient-os

Sentient OS

Sentient OS is the Mac application that quietly runs a local artificial intelligence model overnight to build a private memory of your digital life. While you sleep, the app securely wakes your computer to read your local files, screenshots, and messages, distilling the new information into plain text summaries. It completely filters out sensitive data on-device before passing anything to your own external frontier model, keeping your raw files strictly on your hard drive. By the time you wake up, Sentient OS presents you with ready-to-use draft replies and automated task suggestions, letting you handle your daily digital chores with a single click.

🆔 @hackernewsgithubprojects

📦 hesorchen/muselab

muselab

Run your own local artificial intelligence workspace without burning through expensive API fees. This clever project connects directly to your existing browser subscriptions, letting you reuse those accounts to power your own self-hosted setup. Instead of staring at an empty chat box, you get a fully integrated dashboard that combines file management, instant document rendering, and a real command terminal side by side with your model. It tracks your projects across multiple directories and syncs everything smoothly between your computer and phone. You get a unified, powerful playground for your files and code that actually keeps your data in your own hands.

🆔 @hackernewsgithubprojects

📦 experientiallabs/world-model-optimizer

World Model Optimizer

World Model Optimizer is the agent command-line tool that turns your real-world execution traces into highly optimized local agent runtimes. This tool ingests your historical agent activity logs to build a customized, interactive simulation of your software environment. Instead of testing your agents on slow, expensive real-world APIs, you run parallel evaluations directly inside this simulated world model. From there, you can easily find the best prompts, select optimal tools, and even distill smaller, lightning-fast models that match giant frontier models at a fraction of the cost. Start optimizing your agents today by checking out the world-model-optimizer repository on GitHub.

📰 https://news.ycombinator.com/item?id=49063454

🆔 @hackernewsgithubprojects

📦 chenxiachan/thoughtdag

ThoughtDAG

ThoughtDAG is the infinite canvas that finally lets you edit the memory of your AI assistant. Unlike traditional chat interfaces that hide your conversation history, this tool maps your reasoning visually using an acyclic graph. Its core principle is simple: wires are the context. By drawing or cutting lines between text nodes, you decide exactly what the AI remembers. You can import documents, highlight passages, and watch the canvas automatically cite page numbers. When you delete a connection, the AI forgets the noise and regenerates a cleaner answer. It is the perfect playground for researchers who want to structure deep thoughts visually.

📰 https://news.ycombinator.com/item?id=49057101

🆔 @hackernewsgithubprojects

📦 miaai-lab/qwen3.6-27b-nvfp4-vllm

Qwen 3.6 27B NVFP4 vLLM

Spin up a high-performance, private AI server on your own hardware in just minutes. The project qwen3.6-27b-nvfp4-vllm gives you a complete, ready-to-run setup for hosting a highly optimized twenty-seven billion parameter language model. It bundles everything into a simple Docker container, complete with automated start and stop scripts, so you do not have to struggle with complex configuration. Because the model is compressed to a highly efficient four-bit format, it fits comfortably on standard forty-eight gigabyte graphics cards while still supporting massive text histories and multi-modal image inputs. Grab the repository, run the launch script, and enjoy your own local, blazing-fast chat assistant.

🆔 @hackernewsgithubprojects

📦 miaai-lab/qwen3.6-27b-nvfp4-vllm

Qwen 3.6 27B NVFP4 vLLM

Spin up a high-performance, private AI server on your own hardware in just minutes. The project qwen3.6-27b-nvfp4-vllm gives you a complete, ready-to-run setup for hosting a highly optimized twenty-seven billion parameter language model. It bundles everything into a simple Docker container, complete with automated start and stop scripts, so you do not have to struggle with complex configuration. Because the model is compressed to a highly efficient four-bit format, it fits comfortably on standard forty-eight gigabyte graphics cards while still supporting massive text histories and multi-modal image inputs. Grab the repository, run the launch script, and enjoy your own local, blazing-fast chat assistant.

🆔 @hackernewsgithubprojects

📦 img2threejs/img2threejs

img2threejs

img2threejs is the image to three d generator that turns a single reference photo into a clean, animation ready, code only three js model. Instead of giving you a massive, messy mesh file or a blurry photo projection, this clever tool writes a lightweight typescript factory file that builds the object using clean geometric primitives and procedural shaders. It works in structured, quality gated passes, checking things like joint attachments and realistic materials to make sure the parts actually line up and move naturally. It is incredibly token efficient because clever python scripts handle the heavy lifting while the artificial intelligence only steps in to judge the visual quality. If...

🆔 @hackernewsgithubprojects

📦 songguoxs/seedance-prompt-skill

Seedance Prompt Skill

Design cinematic, production-ready video prompts for ByteDance's video generator right from your command line. The seedance-prompt-skill is a specialized custom tool for Claude Code that turns the AI into a professional prompt engineer. You simply describe your creative concept in natural language, and the tool outputs structured, multi-modal Chinese prompts tailored for the generator's advanced capabilities, including camera motion replication, storyboard continuation, and video editing. To get started, clone the repository, copy the skill folder into your Claude directory, and launch your workspace. Take control of your AI filmmaking process and bring your visual ideas to life instantly.

🆔 @hackernewsgithubprojects

📦 grp06/useful-codex-skills

useful-codex-skills

You can now completely automate how your coding assistant plans, pressure-tests, and executes complex software refactors. useful-codex-skills is a clever collection of modular workflows that teach your AI assistant to act like a seasoned software architect. Instead of writing messy, uncoordinated code, it uses these tools to run structured decision-tree interviews, turn vague tasks into highly detailed execution plans based on classic software design principles, and even publish clear explanations of code changes directly to your connected Notion workspace. It is a brilliant way to ensure your development agents always produce simple, well-tested, and predictable code on the first try.

🆔 @hackernewsgithubprojects

📦 laskco/mpvnova

mpvnova

Your Android TV can finally play high quality videos smoothly without stuttering or compatibility issues. That is because mpvnova is a media player built specifically for television screens, bringing the legendary power of mpv directly to your living room. Instead of struggling with sluggish interfaces designed for touchscreens, this app gives you a custom television user interface, advanced audio delay controls, and specialized decoder tools to handle tough video files. It even includes clever workarounds for popular streaming hardware, like fallback options for Nvidia Shield devices, ensuring your movies play perfectly every time. It is the ultimate way to make your smart TV actually smart.

📰 https://news.ycombinator.com/item?id=48085168

🆔 @hackernewsgithubprojects

📦 ancientjames/scanwheel

Scanwheel mechanical television

Build your own working mechanical television using 3D printed parts and a Raspberry Pi Pico. Scanwheel is a drum-style mechanical TV featuring five side-by-side viewing windows that share a single spinning drum. Tiny pinholes in the drum act as physical pixels, while separate, synchronized light-emitting diodes recreate images behind each window. By using the Pico's specialized hardware timers, the system achieves incredibly high horizontal resolution, making the twenty-line display look surprisingly detailed and sharp. It is a brilliant way to bring early twentieth-century television technology into the modern era using simple, accessible maker electronics.

📰 https://news.ycombinator.com/item?id=49047707

🆔 @hackernewsgithubprojects

📦 ahapbean/xhc

xhc

Scale artificial intelligence models wider and push their intelligence further by unlocking massive parallel pathways. This is made possible by xhc, a breakthrough design that expands the standard single connections inside neural networks into sixteen parallel streams. Usually, adding more streams slows computers to a crawl, but this method reads all sixteen streams while only updating four active ones at a time. It also uses smart temporal features to make sure the information stays rich and diverse. This clever shortcut delivers massive performance boosts with barely any extra computing cost. It is a brilliant way to make big models much smarter without needing more hardware.

🆔 @hackernewsgithubprojects