📦 licoy/strokemouse

StrokeMouse: The Mac Mouse Gesture Hack

Your Mac mouse button is now a secret remote control for your entire computer. StrokeMouse is a tiny menu bar app for Mac that turns the physical act of drawing shapes with your mouse into instant shortcuts. Hold down your right mouse button and scribble a simple curve, like an up arrow or a circle, and the app instantly recognizes that shape to trigger commands like opening Safari, minimizing windows, or running custom scripts. It’s like adding a hidden layer of automation to your hardware without any new buttons. The coolest part is that it feels like you’re hacking your own system.

🆔 @hackernewsgithubprojects

📦 lambdahumanoid/openlambda

Openlambda: Teaching Robots by Watching Humans

Openlambda turns everyday human video into a foundation model for humanoid robots that learn to walk and grab objects. Instead of expensive robot-only demos, it trains on first-person footage of people moving, using a three-stage learning process to map human motion to robot actions. This solves the data bottleneck for generalist humanoid control, letting machines learn real-world skills from cheap, open-world human experience. The result is a versatile base for building robots that can adapt to new tasks without starting from scratch. It’s a practical step toward robots that learn like we do.

🆔 @hackernewsgithubprojects

📦 oosexgang/hakios-xnu-kernel

HakiOS: The XNU Kernel Running on Android

HakiOS is the experimental setup that actually boots real iOS on an Android phone. It takes the open-source XNU kernel, adds a custom boot loader, and runs the standard iOS user space on a Qualcomm chip. The coolest part is that it does not flash anything to your device. Everything lives in memory during the session, so a simple reboot returns the phone to its normal Android state. This makes it a perfect research bench rather than a risky permanent change. You get to see the lock screen and apps working on hardware that was never meant for them.

🆔 @hackernewsgithubprojects

📦 sheltercosmo/rilldex-rl-studio

How to See Inside an AI Agent

Rilldex RL Studio is the observability suite that finally shows you how an AI agent actually works during training. It stops the mystery of the black box by turning raw training data into clear visual maps of every action, retry, and decision your model makes. You can watch exactly where the agent gets stuck or goes off track, linking those behaviors directly to the performance metrics from your training run. This connects the dots between the numbers and the actual work, helping you fix problems much faster than staring at loss curves ever could. It turns invisible machine learning processes into a clear, inspectable story you can actually read.

🆔 @hackernewsgithubprojects

📦 leggedrobotics/defm

DeFM: Depth Foundation Model

A neural network now learns to understand the three-dimensional world purely from depth sensors, skipping the usual need for colored photos entirely. DeFM is a vision model trained on sixty million depth images that teaches computers to recognize objects and navigate spaces using only distance data. This is huge for robotics because it allows machines to perceive their environment accurately in any lighting condition, even total darkness, without needing a separate camera for every task. It essentially gives robots a universal sense of space that works across different sensors and real-world settings.

🆔 @hackernewsgithubprojects

📦 diffuspace/clover-35b-a3b-preview

Clover: The Diffusion Model Breakthrough

Clover is the open-source language model that generates text using a completely different engine. While standard A.I. predicts words one after another, this model paints the whole sentence in blocks, refining it like a picture until it makes sense. That is a massive shift in how machines create language. It packs thirty-five billion parameters but only activates three billion per token, making it efficient enough to run on a single graphics card. If you have ever wondered what text diffusion actually looks like in practice, this is your chance to see it. It is fast, it is open, and it is built to be tinkered with.

🆔 @hackernewsgithubprojects

📦 datalab-to/omni_parse_bench

Omni Parse Bench: The Truth About Your Document Parser

Your fancy AI document parser just choked on a single spreadsheet. That’s exactly what Omni Parse Bench is built to catch. This project stops us from guessing how well these tools actually work by slapping over sixteen thousand specific yes-or-no checks onto nearly three thousand real pages. It doesn’t just ask if the text is there. It verifies if the table structure survived, if the math holds up, and if the layout stayed sane. You get one clear number per system that tells you exactly where it fails and what that failure costs you in real dollars.

🆔 @hackernewsgithubprojects

📦 emanuele-em/kanzo

Kanzo: The Kanban Board for AI Agents

Kanzo is the self-hosted kanban board that finally lets your coding agents work alongside you. It is a single four megabyte binary that runs on your own computer, meaning no servers and no accounts to manage. The coolest part is that it acts as a bridge between you and your AI tools. Your coding agents can grab tasks, move them to doing, and even comment on their progress directly on the board. You can watch them work in real time from your browser. It’s like having a shared to-do list where humans and robots actually collaborate.

🆔 @hackernewsgithubprojects

📦 typetheoryforall/fvjobs

FVjobs: The Job Board for Formal Verification

FVjobs tracks seventeen open roles in formal verification, a niche field where the most surprising fact is that it specifically records which proof stack each job uses, like Lean or Isabelle, so you know exactly if your skills match. The project solves a real problem: formal verification jobs are usually scattered across academic lists and general boards that have no idea what a theorem prover is. This board collects them in one place, telling you the proof tools required and whether you are legally allowed to take the role from your location. It is useful because it removes the guesswork from applying to a specialized job.

🆔 @hackernewsgithubprojects

📦 deadhead04/skilled

skilled: Test Your AI Skills

skilled is the testing tool that finally proves your AI agent actually works. It acts like a rigorous exam for your custom instructions, checking if your model handles edge cases correctly before you trust it. If you tweak the prompt or swap the model, this tool reruns the same tests to catch silent failures. It’s not just a checker; it’s an improver, guiding your agent to fix weak spots based on real evidence. Think of it as a safety net for your AI workflows. No more guessing if a change broke something. Just run the suite, see the results, and ship with confidence.

🆔 @hackernewsgithubprojects

📦 haydenschmitty/floorplan

Floorplan: Your Codebase, Mapped in Plain English

Floorplan is the code-mapping tool that turns your messy codebase into a living, plain-English floor plan. It works by letting an AI read your files and draw an interactive map where every part gets a human name, not a cryptic filename. The coolest part is that it connects your front-end buttons directly to the server code that handles them, even if they live in completely separate repositories. It also runs a health check to find forgotten passwords or security holes, giving you a one-click fix for each.

🆔 @hackernewsgithubprojects

📦 nvlabs/voloagent

VoLoAgent: The Robot Brain

Teach a robot to recover from its own mistakes by using a vision language model as a safety net. VoLoAgent acts as a smart middleman between a standard robot policy and the simulation, breaking complex physical tasks into smaller steps. The standout feature is its failure detection: if the robot drops an object or gets stuck, the system spots the error instantly and triggers a recovery, such as re-grasping the item, without human help. This turns a fragile automation script into a resilient agent capable of handling long, multi-step jobs. It is a clear look at how language models can make physical robotics more reliable and self-correcting.

🆔 @hackernewsgithubprojects

📦 hamsanders/soundbox

Soundbox: The Free Local Audio Studio

Soundbox is the local audio suite that finally lets you own your sound without the subscription trap. You know how every AI music tool wants your credit card and your data? This one just runs on your own graphics card, completely offline, with no account and no monthly fee. You describe a scene, like heavy rain or a breaking window, and it generates high-quality sound effects, music, and even full songs with lyrics right on your machine. It is built for people who want creative freedom without the corporate cloud, giving you a real desktop app where you can edit, mix, and export exactly what you need.

🆔 @hackernewsgithubprojects

📦 mcg-nju/autotavern

AutoTavern

A computer program now plays full games of Hearthstone Battlegrounds entirely on its own, without ever touching the live game servers. This repository, called AutoTavern, recreates the entire game in code so researchers can test how well artificial intelligence makes split-second tactical decisions. Instead of just watching a bot play, you can feed it specific tricky scenarios to see if it truly understands the rules or just guesses. It solves the problem of needing a real human player to practice against, giving scientists a safe, repeatable sandbox to teach machines complex strategy.

🆔 @hackernewsgithubprojects

📦 feimacode/open-design-agent-kit

Open Design Agent Kit: Professional UI for your AI Coder

Ditch the generic, flat code your AI assistant spits out by default. Open Design Agent Kit transforms your standard coding tools into professional-grade design engines, letting you generate interfaces that look polished and brand-ready. It works by feeding hundreds of specific design styles directly into your existing editor, so you get that high-end aesthetic without installing heavy desktop apps or juggling new API keys. The real magic is the variety: you can switch between the sleek look of Apple or the bold energy of Nike with a simple prompt. It’s essentially a massive library of visual rules that teaches your coder to think like a designer.

🆔 @hackernewsgithubprojects