📦 samplaman/owmb

OWMB 3D Sample Cloud

OWMB is the audio tool that maps your entire sample library into an interactive three-dimensional constellation. Instead of scrolling through endless folders, you see thousands of sounds floating in space, clustered together based on how similar their timbre actually sounds. It turns a messy disk full of audio files into a navigable visual map where you can fly through clusters, click a node to audition it instantly, and drag sounds straight into your digital audio workstation. For anyone managing a large collection, this spatial view makes finding that specific snare or synth patch far more intuitive than typing search terms.

🆔 @hackernewsgithubprojects

📦 leekt/skim

Skim: Sync Only What You Need

Skim is the Ethereum state client that syncs only the specific data your application actually uses. Instead of downloading the entire blockchain to answer a single question, it tracks exactly which accounts and storage slots you care about and keeps just that piece current. This means your app runs queries locally in milliseconds without paying for an expensive hosted server or managing a massive full node. It pulls updates straight from the peer network, verifies every piece of data, and ignores everything else. The result is a lightweight tool that lets developers run their own trusted infrastructure without the heavy hardware overhead.

🆔 @hackernewsgithubprojects

📦 mxhm11cr/linuxrunner

Linux Desktops on Your iPhone

LinuxRunner runs a full desktop environment on your iPhone without jailbreaking the device. It interprets ARM64 instructions directly inside an app, allowing you to boot Debian or Alpine Linux and launch real graphical applications like KDE Plasma and Firefox. This solves the problem of mobile devices being locked to a single operating system by creating a sandbox that mimics a standard computer. You get a working terminal, a file manager, and even 3D graphics rendering through a specialized software bridge. It is a fascinating example of how far emulation has come, turning a phone into a capable workstation for development and browsing tasks.

🆔 @hackernewsgithubprojects

📦 zhuoyangliu2005/agentic-wam

Agentic WAM

The robot actually writes its own code. This project, called Agentic World Action Models, makes a robotic arm plan its next move by generating a tiny computer program alongside its physical actions. While standard robots just guess the next motor command, this system imagines the future video, writes a symbolic plan, and moves its gripper all at once. It uses a shared skill library to handle complex tasks like folding towels or using tools, making the robot far more flexible than simple pre-programmed bots. If you are into robotics, this is the bridge between seeing and doing. Check out the Agentic World Action Models repository.

🆔 @hackernewsgithubprojects

📦 haohervchb/turing-unlock

Turing-unlock

A single ASUS RTX 2080 Ti just doubled its floating-point speed without any hardware changes. This project, called Turing-unlock, is a deep dive into a volatile, pre-operating-system trick that pokes specific firmware controls on this exact card. It is not a permanent mod or a generic installer, but a highly specific experiment that runs before your computer even boots. By tweaking internal settings, the chip’s compute units shifted from half speed to full speed, effectively doubling its tensor performance for a single run. It is a fascinating, if slightly scary, look at how far we can push hardware limits before the system even wakes up.

🆔 @hackernewsgithubprojects

📦 charliegerard/overgrown

Overgrown: The Chrome Tab that eats itself

Reclaim your forgotten browser tabs by slashing through the vines that grow over them. Overgrown is a Chrome extension that turns your neglected open tabs into wild, untamed gardens the longer you ignore them. It solves the problem of tab hoarding by making your forgotten digital clutter physically obstructive; if you leave a page unread, plants invade the screen until you can’t use it. To get your work back, you have to swipe your mouse like a machete to chop the vegetation down. It’s a clever, playful way to force yourself to clean up your digital workspace without any extra effort.

🆔 @hackernewsgithubprojects

📦 null0526/dario-genesis

The AI Film Made Entirely From Code

Dario Genesis is the AI short film that proves every pixel and sound can be created purely by code. It is an eleven minute story about human civilization, where visuals are drawn with web graphics and audio is synthesized from scratch using Python. No stock assets or recordings were used. The project is a collaboration between different AI models, with one directing the visuals and another handling the music. The source code is open for anyone to read, modify, or learn from. It shows a unique workflow for making art entirely through programming.

🆔 @hackernewsgithubprojects

📦 lambdatest/rook

Rook: Test Your AI Agents Like Real Software

Stop guessing if your AI agent actually works. Rook walks through your project code to automatically build test scenarios that check what your agent really does, not just what it claims to do. It runs these checks against your live tools to catch broken workflows or security holes before you ship. The standout feature is how it handles uncertainty: if it cannot verify an action, it reports that gap instead of guessing a pass. This gives you a clear picture of exactly what works and where your blind spots are. You get honest, evidence-based results that save you from embarrassing bugs in production.

🆔 @hackernewsgithubprojects

📦 toon-format/toon-java

JToon: 30% Token Savings for Java LLM Apps

JToon cuts LLM token costs by thirty to sixty percent by converting standard JSON into a compact, human-readable format called TOON. If you are building Java applications that talk to large language models, this is the missing piece for your context window. It looks like a mix of YAML and CSV, using indentation and tabular rows to squeeze out every possible space without sacrificing readability. The library handles encoding, decoding, and validation out of the box, so you can swap formats with a single function call. Because it is purely community-driven and built for Java, it fits right into your existing toolchain.

🆔 @hackernewsgithubprojects

📦 gminustwo/primerecomp

PrimeRecomp: Metroid Prime Remastered on PC

PrimeRecomp is the native port that finally lets you play Metroid Prime Remastered on your PC without an emulator. Instead of slow emulation, it statically recompiles the Switch game code directly into C, translating it for your specific hardware so it runs at full speed. You need your own game files, but the result is a smooth, modern experience with mouse aiming and high frame rates. It’s a clever, hardcore approach to PC gaming, proving you can make console ports feel native.

🆔 @hackernewsgithubprojects

📦 asherjn/recall

Recall: Overwatch on Apple Silicon

Recall is the Mac app that finally makes Overwatch feel native on Apple Silicon. Instead of wrestling with Windows emulation layers, it runs the game directly on your hardware for a massive speed boost. The real magic is in the aim. It bypasses the operating system’s mouse handling to read your input directly, meaning every flick lands instantly without lag. It is completely free and open source, so you just download it, sign in, and play. No paid licenses, no complicated setup. If you have an M-series Mac and want smoother matches with sharper aim, this is the one to try.

🆔 @hackernewsgithubprojects

📦 openvetta/open-vetta

Open Vetta: Your Local AI Agent

Open Vetta is a desktop AI agent that runs entirely on your computer, meaning your files and data never leave your control. It is not just a chat bot; it is a workspace that can actually inspect your projects, run commands, and generate real files like code or spreadsheets while you watch. The most interesting part is how you can bring your own models, connecting local or private AI endpoints directly to your workflow. This solves the problem of trusting a black box with sensitive work, giving you a transparent, reviewable trail of every action the agent takes.

🆔 @hackernewsgithubprojects

📦 microsoft/scope

Scope: How to Test Your AI Agents

Microsoft just released Scope, a platform that puts your AI coding agents on a timed obstacle course. Instead of guessing if your bot is helpful, you define specific tasks and let agents like Copilot or Claude attempt them. The system watches every single action, from typing code to debugging errors, and scores the results against your own custom checklist. It is like hiring a very strict, objective manager who never gets tired and never takes bribes. You can watch the agents work in real time, see exactly where they got stuck, and compare their performance side by side.

🆔 @hackernewsgithubprojects

📦 odiecay/android-tv-ps4

Android TV for PS4

Turn your PlayStation 4 into a full Android TV box with a single installation package. This project boots your console directly into the Android operating system, running it from the internal drive while keeping your apps, accounts, and settings safe between sessions. It is built on existing Linux work, making it possible to run modern streaming services and applications on gaming hardware. The developer is actively testing various console models to fix display issues and ensure compatibility across different firmware versions. It is a fascinating experiment in repurposing hardware for new uses.

🆔 @hackernewsgithubprojects

📦 blakeandersonw/ai-app-playbook

AI App Playbook: Build Your Next Hit App

The AI App Playbook is a complete roadmap for building profitable mobile apps in twenty twenty six. It gives you a specific tech stack and a step by step guide to go from idea to revenue. The author built five apps making over one hundred thousand dollars a month using this exact method. You can even clone the repo and point your coding agent at the included instructions to start building immediately. It skips the fluff and gives you the real tools to make money with AI. Stop guessing and start building your next successful app today.

🆔 @hackernewsgithubprojects

📦 dk23lhl/termius-mcp

Termius MCP: Your AI's New Terminal Remote

Termius MCP is the bridge that finally lets your AI assistant actually read and type in your terminal. It connects to the Termius app you already use, so your agent can run commands and upload files without you copying and pasting everything. The coolest part is that it tracks real file uploads, so you know exactly when they finish. No more guessing if your AI is stuck. If you use AI to help with your coding, this is the tool that makes it feel like it has hands.

🆔 @hackernewsgithubprojects

📦 ekzhang/resin

Resin: The LEGO Kit for AI Speed

Resin is the inference toolkit that lets you build custom AI speed engines from scratch. It is not a finished product; it is a library of high-performance building blocks written in Rust and Python. The problem it solves is that off-the-shelf tools often waste power on tasks you do not need. Resin handles the slow, messy scheduling work so you can focus purely on writing fast math for your specific hardware. The coolest part is the proof of concept: developers used these blocks to build standalone engines for major language models in just twelve hours, matching the speed of much larger, complex systems.

📰 https://news.ycombinator.com/item?id=50247183

🆔 @hackernewsgithubprojects

📦 zgx949/ai-pojia

AI Pojia: The Fixer for AI Tasks

AI Pojia is the configuration tool that finally makes AI agents deliver verified results instead of just talking. It solves the frustrating problem where AI models often refuse to touch actual code or fail to prove their work. By installing a specific skill and a tweaked prompt, you turn your AI assistant into a diligent worker that runs real tests and keeps copies of every change. This means you get a tangible record of what was done, not just a vague summary. It is a clever, hands-on way to ensure your AI actually does the job, making your projects safer and more reliable.

🆔 @hackernewsgithubprojects

📦 motionsites-ai/claude-skill

Claude Skill: Cinematic Landing Pages

Claude Skill is the website builder that makes AI look like a human designer. It stops the model from guessing at layouts and instead uses real cinematic video, precise design rules, and forty proven examples to guide the build. The most surprising part is that it actually looks at its own work. It takes screenshots, finds what looks broken, and fixes it until the page looks polished. This turns a rough draft into a high quality landing page. You get a site that looks like it took weeks to design, not minutes of typing.

🆔 @hackernewsgithubprojects

📦 minimax-ai/minimax_search

MiniMax Search: Give AI Real Web Eyes

Hand an AI agent the ability to hunt down facts and digest entire web pages in one go. MiniMax Search plugs directly into your coding tools via a standard protocol, letting the model run multiple Google searches at once and then read the resulting links to answer your specific questions. It handles the messy work of fetching content and turning it into clean text, so your assistant can actually understand what is on the page instead of just guessing from search snippets. You get a smart helper that can verify information across sources without you manually pasting links into a chat window.

🆔 @hackernewsgithubprojects