📦 clausy9/evo-wam

evo-wam: Robots That Teach Themselves By Watching Videos

This robot learns to pick up objects by watching videos of itself, without any human teacher. The project, called evo wam, solves the problem that robots usually need thousands of hours of human recording to learn basic tasks. Instead, the system generates its own attempts, checks if the actions match the video, and keeps only the successful parts to train on. It improves its skills round after round, boosting success rates from twenty percent to nearly eighty percent on real robot tasks. It proves machines can improve by verifying their own imagination, not just by following orders. That is how robots get smarter without a human in the loop.

🆔 @hackernewsgithubprojects

📦 valeriodolci/capy-flash

Run Big AI on Cheap Hardware

capy-flash is the inference engine that runs massive AI models on just two cheap graphics cards. Most people assume you need expensive data centers to run powerful language models, but this tool splits the heavy math across two sixteen gigabyte consumer GPUs. It even moves parts of the calculation to your computer's regular memory so the smaller cards do not run out of space. It turns affordable hardware into a capable local server for serious AI tasks. No more waiting for a server cluster. Your desktop can do this right now.

📰 https://news.ycombinator.com/item?id=50118153

🆔 @hackernewsgithubprojects

📦 perrylink/dsh-data-quality

The AI That Doesn't Guess Your Data

This plugin performs data checks without letting the AI guess. Instead of letting a language model estimate your numbers, it runs strict code to profile, clean, and verify your files. This means it finds duplicates, fixes messy dates, and checks if your numbers actually match your records. It matters because it gives you a clear, repeatable scorecard on data quality, so you can trust your results. It stops the model from making things up. It just tells you what is actually there. This is the real takeaway: you get hard facts, not AI guesses.

🆔 @hackernewsgithubprojects

📦 nvidiagameworks/rtx-remix-minecraft

RTX Remix Minecraft: Realistic Lighting

RTX Remix Minecraft is the Fabric mod that brings realistic light bouncing to your favorite voxel world. It replaces the simple, flat lighting you usually see in Minecraft with a system that calculates how light actually travels. A torch no longer just lights up the immediate blocks around it. Instead, the light bounces off nearby walls, creating soft shadows and subtle color shifts in the dark. It turns a blocky game into something that looks surprisingly lifelike. It is still in beta, so expect some tweaking. But for anyone who wants to see their builds look amazing, this is the mod to check out.

🆔 @hackernewsgithubprojects

📦 wquguru/cicada

Cicada: Your Mac's AI Interpreter

Turn any video call into a bilingual experience where your Mac listens to the system audio and instantly displays live subtitles and structured meeting notes. You can even hold down a key to speak, and the app broadcasts your voice back into the call in the other person's language, making it feel like you have a professional interpreter sitting next to you. It is the perfect tool for developers and creators who need to follow global streams without missing a beat. The best part is that the app keeps all your recordings and transcripts safely on your computer, so you never have to worry about your data leaking.

🆔 @hackernewsgithubprojects

📦 mille-okws/pestinha

Pestinha: Your Desktop Pet with a Brain

A tiny character on your screen actually learns from your actions using its own local neural network. Pestinha is a desktop pet for Windows, Mac, and Linux that lives entirely on your computer without sending data to the cloud. The most interesting part is that it has genuine emotions and a personality that evolves based on how you treat it. When you pet it, it gets happy and remembers the kindness. If you ignore it or put it in time out, it gets sad, throws tantrums, or even goes on strike against your cursor.

🆔 @hackernewsgithubprojects

📦 traccia-ai/traccia-py

Traccia: The Control Plane for AI Agents

Your AI agent is about to spend fifty dollars on a single API call, and nobody is watching. Traccia is a Python SDK that acts as a real-time control plane, letting you see exactly what your agent does and, more importantly, stopping it before it breaks your budget or security rules. It plugs into your favorite frameworks like LangGraph or CrewAI automatically, tracking every token and cost without rewriting your code. But the killer feature is runtime policy enforcement. You can set hard limits, like a spend cap or a loop cap, and if your agent tries to go rogue, Traccia blocks the action instantly.

📰 https://news.ycombinator.com/item?id=50357443

🆔 @hackernewsgithubprojects

📦 kayraucklnc/beluga

Beluga: The IDE Built for AI

Beluga is the Mac IDE that treats your AI assistants as real teammates. It is a full application where multiple agents, like Claude or Codex, work side by side in shared workspaces, chatting with each other and coordinating their own code changes. This is not just a chat box; it is a complete environment where the AI agents handle the heavy lifting, like running tests and managing branches, while you just watch the dashboard. You get a view of the entire workflow, seeing exactly what each agent is doing and how they are collaborating on your project. It turns your development process into a visual, multi-agent collaboration.

🆔 @hackernewsgithubprojects

📦 okcontract/oksolc

oksolc: The Solidity Compiler Built for Speed

oksolc is the Zig-based Solidity compiler that cuts compile times by five times. If you are writing smart contracts, you know how long you wait for the code to build. This project rewrites the compiler in a faster language, allowing it to process your code in parallel. It caches the parts that have not changed, so you only wait for the new edits. The goal is not to replace the original compiler for final deployment, but to make the daily development loop feel instant. You get the same output, just much faster.

📰 https://news.ycombinator.com/item?id=50056963

🆔 @hackernewsgithubprojects

📦 kadobot/comfyui-h3-continuity-studio

H3 Continuity Studio Keeps Characters Consistent

Directing multi-shot videos with artificial intelligence usually means losing track of who your characters are and what they look like after a few scenes. H3 Continuity Studio solves this by giving your video editor a long-term memory. It remembers the exact appearance of a person or place in the first shot and carries that visual truth through every later clip, even if you change the camera angle or move to a totally different location. This keeps your final movie feeling like one cohesive story instead of a patchwork of random images. You get a full editing workspace to plan your narrative, manage these visual memories, and assemble your final cut.

🆔 @hackernewsgithubprojects

📦 vicpitic/the-brains

The Brains

Tap into the collective wisdom of top founders by connecting their talks, essays, and posts directly to your AI tools. The Brains packages this vast library of insights into a searchable, cited resource, so your AI assistant doesn't just guess but answers with verified facts from industry leaders. This means you can query specific strategies or ideas instantly, getting straight to the source without digging through hours of video. It turns scattered online content into a reliable, queryable knowledge base, making it easy to tap into expert advice whenever you need it. This project is a powerful way to make sure your AI is grounded in real, high-quality founder knowledge.

🆔 @hackernewsgithubprojects

📦 minimax-ai/minimax-code-miniapps

MiniMax Code MiniApps

MiniMax Code MiniApps is the plugin collection that turns a chat interface into a full interactive workspace. It is an open community repository where developers share self-contained software packages that run directly inside the MiniMax Code desktop app. The most surprising aspect is how these mini apps let you build and edit actual 3D models right on your screen. You can point at a specific part of a digital object, type a request to change it, and watch the agent instantly update the geometry. This approach keeps you in control of every creative decision rather than relying on blind text prompts.

🆔 @hackernewsgithubprojects

📦 bobliuuu/open-flapping-airplanes

open flapping airplanes

Launch a single command and watch one GPU kernel train a language model, stream live text responses, and play Doom at thirty-five frames per second simultaneously. This project, called open flapping airplanes, recreates a viral demo by writing one persistent worker that never stops. Instead of the computer constantly starting and stopping tiny tasks, which wastes time, this single worker stays active and pulls work from a queue. The most surprising part is that it actually runs the classic Doom engine right on the graphics chip while training. This proves that modern hardware can handle mixed workloads without constant overhead.

🆔 @hackernewsgithubprojects

📦 color-ide/coloride

Coloride: The Local-First IDE for Mac

Coloride is a desktop editor that refuses to call home, meaning your code stays on your Mac without needing an account or cloud connection to start editing. It bundles a file explorer, live terminals, and even a built-in browser so you can test your work in one window. The most interesting part is the strict local-first design, where every action runs on your own machine by default. You get a complete development setup that respects your privacy, proving that a powerful tool does not require handing over your data to a server. If you value keeping your code private, this is the one to try.

🆔 @hackernewsgithubprojects

📦 ethany14/eregion-job-finding-agent

Eregion Job Finding Agent

Eregion Job Finding Agent fills out real job applications on your computer and stops right before you hit the submit button. It searches company boards and LinkedIn, reads every posting in full, and sorts the results by how well they fit your resume. The most surprising part is that it actually opens your browser, types in your details, and navigates through the forms page by page. It handles the tedious typing so you can focus on the decisions that matter. If a CAPTCHA or password field appears, it pauses and waits for you to take over. Since everything runs locally on your machine, your data never leaves your hands.

🆔 @hackernewsgithubprojects

📦 lxe/dsx

Run a 552B AI Model on Your Gaming PC

Run a massive five hundred and fifty billion parameter AI model on your own gaming desktop. dsx is a custom engine built in Rust and CUDA that squeezes this huge model into two twenty-four gigabyte graphics cards and one hundred twenty eight gigabytes of system memory. Usually, software this big needs a datacenter, but this project spreads the model’s experts across your graphics cards and processor so they can all work at the same time. The coolest part is that it actually follows your conversation in real time.

🆔 @hackernewsgithubprojects

📦 openselena/omniget

OmniGet: The One-App Download Suite

Your browser is quietly forgetting every video, article, and song you ever saved, and OmniGet fixes that by becoming your personal, offline library for everything you find online. It is a desktop application that acts like a universal grabber, pulling high-quality video, music, and even entire paid courses from over a thousand websites right onto your hard drive. The coolest part is that it isn't just a download button; it is a full workbench that lets you clean up audio, organize your files, and even keep a private chat client running all in one place. You stop worrying about links breaking or platforms changing their rules.

🆔 @hackernewsgithubprojects

📦 kkkkhazix/agent-autoloop

Agent Autoloop: The Silent Worker

Stop babysitting your terminal. Agent Autoloop turns your command line agents into a background worker that picks up tasks, does the heavy lifting, and then quietly waits for your input only when a human decision is required. The truly surprising feature is the zero cost wait. When the agent hits a critical choice, it shuts down completely. No process runs and no tokens are spent while you take your time. The system stores its state in a local database, so if you reply three days later, the agent instantly resumes exactly where it left off.

🆔 @hackernewsgithubprojects

📦 saos-3624/dsh-see-now

See Now: The Screen Watcher That Saves Money

See Now is the visual plugin that makes artificial intelligence watch your screen without burning through your budget. It solves the problem of feeding constant screenshots to a model, which is wasteful and often misses important details. Instead, this tool measures how much the picture actually changes on your display. It only sends a new image when something real happens, like a window switch or a video scene change. If your screen is still, it sends almost nothing, saving you from paying for blank frames. You get a clear picture of what your computer is doing, but only when it matters.

🆔 @hackernewsgithubprojects

📦 balanov/skein

Skein: The Emergency Stop for AI Swarms

Skein is the fail-closed orchestration tool that ensures your local agent swarm actually stops when you tell it to. It executes task graphs with a persistent epoch fence, meaning if you hit the emergency stop, it doesn't just ignore the signal. It actively rejects any late results from the old run and kills the underlying processes immediately. This is huge because it prevents zombie tasks from finishing work after you've already called it off. You get a single static binary with no database, just pure control. If you are running local AI agents, this is the safety net you need.

🆔 @hackernewsgithubprojects