298 subscribers
14K videos
14.6K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ scottduncanai/public-chemical-manufacturing-datasets

Real Plant Data for AI

Government permit filings are quietly exposing the engineering blueprints of real industrial water treatment plants. This repository collects those public records, giving developers actual piping diagrams and live process data that were previously locked away in private files. It solves a major gap for anyone building AI to monitor chemical processes, offering real-world examples of faults and equipment behavior. You get the ground truth needed to test your models against physics, not just math. Now you can build reliable tools that actually understand how factories run.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ jasonpersinger/ghoulbox-fm1-dungeon-synth

Ghoulbox FM1 Dungeon Synth

A tiny keyboard can compose entire medieval soundscapes by itself. Ghoulbox is custom software for a pocket synthesizer that turns it into a dedicated dungeon music machine. It strips away modern dance beats and focuses entirely on dark, medieval, and fantasy tones. The coolest part is a feature called Dream. You just pick a mood and press a key, and the device writes a slow, evolving piece of music for you. It handles drums, strings, and drones automatically, creating a unique track every time. It is perfect for people who want instant, atmospheric music without complex programming.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ tthanh2k6/turbo-browser-mcp

How Turbo Browser MCP Saves AI Tokens

Turbo Browser MCP is the browser automation server that makes AI agents faster by drastically cutting down the amount of data they process. It replaces dozens of small, slow commands with a few powerful tools that handle complex tasks like filling entire forms or managing multiple tabs in a single step. The real magic happens when it shows the AI only the parts of the page that actually changed, rather than reloading the whole screen every time. This tiny tweak saves a massive amount of processing power and keeps your chat history clean. It is a clever way to build smarter, quicker web automation without overwhelming your system.

πŸ“° https://news.ycombinator.com/item?id=50503930

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ olovalabs/kathamo

Kathamo: The terminal for AI agents

Kathamo is the native terminal that finally tells you when your AI coder is done. You know that awkward feeling when you spawn an agent to fix a bug and then just sit there, wondering if it crashed, finished, or is stuck? Kathamo fixes that. It is a fast, built-in terminal that runs multiple coding agents side by side, but the magic is the attention system. It tracks exactly when an agent is working, blocked, or finished, and only pings you when you actually need to look. It even remembers your exact workspace when you come back, so your tabs and code are right where you left them.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ davnords/loma

LoMa: The Smarter Way to Match Images

LoMa, a new image matching tool, spots identical details across photos even when they look completely different to the human eye. It was built by researchers at several major Swedish universities to solve a common problem: computers struggle to link points in one picture to the same points in another if the angle or lighting changes. By finding these invisible connections, LoMa helps robots map new rooms or allows apps to find a specific building in a photo. It is essentially a drop-in upgrade for existing vision systems that need to know exactly where they are in the real world.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ guanmo-ai/design-style

Design Style: The Ultimate Visual Reference

Learn the visual language behind iconic design movements by exploring a library that breaks down styles like Bauhaus, Memphis, and Liquid Glass into short, high-quality motion films. Instead of just showing you a picture, this project hands you the entire recipe, including editable source code, practical design rules, and even the synthesized music used in the clips. It is essentially a masterclass for humans and A I alike, letting you see exactly how a specific aesthetic moves and feels. Whether you are a designer looking for inspiration or a developer building an interface, this repository gives you the exact tools to replicate the look.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ modalitydance/awesome-agent-as-a-judge

The AI Judge Is An AI

A recent survey reveals that the best judges for AI output are actually other AI agents, not humans. This repository, called awesome agent as a judge, catalogs the latest research on using language model agents to evaluate everything from code to medical reasoning. It solves the problem of scale, as human graders are too slow and expensive to check every single AI output. The coolest part is that these agent judges don't just give a simple yes or no. They use tools, debate each other, and plan out their evaluation steps to catch subtle errors. This is a game changer for anyone building reliable AI systems.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ starslingdev/skills

StarSling Skills: Why is your CI slow?

Diagnose exactly why your build is slow by analyzing real run history instead of guessing. StarSling Skills audits your GitHub Actions workflows to pinpoint specific bottlenecks using a catalog of over ninety optimization patterns. It measures actual wall clock wait times and runner minutes spent, then generates a prioritized report with ready to use prompts for your coding agent. This turns vague performance complaints into concrete, measured fixes you can apply immediately. Stop guessing what slows down your pipeline and start seeing the data. Grab these skills to make your CI faster and more efficient today.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ ttillberg/codevember-2026

Codevember 2026: Paint a City

Scribble roads, bridges, and roundabouts onto a digital canvas to build a living city that reacts to your strokes in real time. Theo Tillberg’s Codevember 2026 challenges creators to build one small project a day in November, and this interactive map turns simple line drawings into a working traffic simulation. As you paint, the streets snap into place, cars navigate the new lanes, and the environment updates instantly, making it a playful way to explore urban design. It is a creative space for seeing how infrastructure changes traffic flow, offering a unique blend of art and logic.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ link-ding/image-to-psd

Image to PSD: The Magic Layer Splitter

image to psd is the agent skill that finally turns a single flat character illustration into a fully layered project file without you touching a single brush. It takes one image, whether it is a photo, a scan, or a drawing, and breaks it down into nineteen distinct parts, from the hair behind the head to the earrings on the ears. The coolest part is that it doesn't just cut out what you see. It actually guesses what is hidden underneath, filling in the face beneath the bangs or the hair behind the neck so every layer is a complete, usable piece.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ mit-han-lab/pear

Pear: Teaching Robots to Think

PEAR is the training engine that teaches computer vision models to physically control real robots without needing any human video examples. Instead of showing an AI thousands of hours of footage, this system uses reinforcement learning inside a digital sandbox. The model practices picking up objects and navigating spaces in a simulator until it understands the physical world well enough to drive actual hardware. It solves the massive data bottleneck of robotics by letting the algorithm learn through trial and error in code rather than relying on expensive, manual demonstrations. This approach allows a single brain to operate different bodies, making autonomous machines easier to scale.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ miaai-lab/claude-fable-5.2-html-files

Claude Fable 5.2 HTML Files

Browse seventy-six original visual studies generated by a language model, complete with the exact prompts used to create them. This gallery lets you see the final result and then read the specific instructions that produced it, turning abstract AI outputs into a clear, searchable archive of creative experiments. It works like a digital scrapbook where you can inspect the connection between human intent and machine interpretation. This transparency is rare, offering a direct window into how these tools actually think and render. You get the final image and the recipe used to make it, which helps anyone understand the process without guessing.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ tuff-app/tuff

TUFF: Run Massive AI Locally on Mac

TUFF lets you run massive artificial intelligence models directly on a Mac, even when the model is far bigger than the computer's memory. The secret lies in how it handles these giant networks. Instead of trying to fit everything into the fast memory, it keeps the parts it doesn't need right now on the hard drive. As the model thinks, it pulls in only the specific pieces required for each word. This trick allows a basic laptop to power a model that would normally need a supercomputer. It is a fascinating look at how hardware limits can be bent with smart software design.

πŸ†” @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
πŸ“¦ gamingwithgravy/optishade

OptiShade: One App, Better Flight Sims

OptiShade is the Windows app that finally unifies your flight simulator graphics settings. Most players juggle separate tools for frame generation, color grading, and driver tweaks, which is a recipe for broken configs and lost time. This single launcher handles all of it by managing specific components for Microsoft Flight Simulator and X-Plane in one place. The standout feature is how it handles updates. Instead of wiping your hard-earned custom presets every time a new version drops, it tracks exactly what it changed. That means you can update your performance drivers and image effects without losing your personal look.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ ajbats/daytona-cce95

Daytona USA Ultimate: The Ultimate Saturn Racing Mod

Daytona USA Ultimate takes the newer Daytona USA Circuit Edition and replaces its wobbly handling with the precise, original feel of the 1995 classic. This Sega Saturn mod keeps the extra tracks and two-player mode from the Circuit Edition, but swaps in the classic car bodies, engine sounds, and road physics from the first game. You can even toggle between the old and new car designs during play. Because the handling data is fixed, the car no longer rocks in banked turns, sliding exactly like the arcade original. If you have the two discs, you build the final image yourself to preserve the authentic hardware experience.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ yasakani-dev/yasakani

Yasakani runs a massive A.I. engine on a single graphics card

A single graphics card can now run one of the world's largest local A.I. models with astonishing speed. The project known as Yasakani is a specialized engine built specifically for Windows users who own an N.B. Video R.T.X. five oh nine zero. It takes a huge language model called Qwen and compresses it so it fits entirely on your desktop computer. The surprising part is that this engine writes code faster than any other local option available right now. It does not just run the model; it is designed to outpace every other competitor on that exact hardware. This is not a general purpose tool.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ unharnessed-agent/unharnessed

Unharnessed: One Tool, Infinite Power

Run your code agent with exactly one tool: a persistent Python shell. That is the entire bet behind Unharnessed. Instead of giving the model a dozen rigid commands to read files or edit text, it hands over a live kernel where it writes its own scripts. The model reads a massive log, filters it, and stores only the errors it needs in a variable, keeping the rest out of its head. It even writes helper functions it can reuse later, all inside that same running process. It is a radical simplification that lets smart models do the heavy lifting. The result is faster, cleaner, and way less noisy.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ pydantic/typing-inspection

The Secret Inside Your Python Type Hints

Python’s type hints usually vanish the moment your code starts running, leaving your validation libraries blind to what you actually wrote. Typing Inspection changes that by letting your programs peek inside those annotations while they are alive. The truly surprising part is that it can unwrap nested layers like a Russian doll, revealing the core type, the special markers, and any attached metadata all at once. This matters because many tools, including the popular Pydantic framework, rely on this precise visibility to catch errors before your application breaks. Instead of guessing what a developer meant, your code can now read the exact intent hidden in the syntax.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ bigpizzav3/claudemom

ClaudeMom: Full Features Without an Account

ClaudeMom lets you run the full desktop application of Claude without needing an official account. Instead of signing in with a standard login, the software uses a local service to make the app believe you are already logged in. This means you can use your own API key to access all features, including voice input and computer control, which are usually missing in key-only modes. The interface is fully translated into Chinese, and the app runs as a native desktop program rather than a web wrapper. It is a clever workaround that gives you the complete user experience while routing requests through your own provider.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ khazp/librecorsa

LibreCorsa: Browser Racing Sim

LibreCorsa is the open-source racing sim that finally brings hardcore physics to your browser. It runs a full-blown motorsport engine that calculates tire temperature, aerodynamic drag, and wet grip changes five hundred times every second. You do not need to install anything or maintain a server to play. This project proves that a website can handle the complex math of a six-degree-of-freedom vehicle model while letting you race against AI or friends peer-to-peer. It is a wild technical display that turns your simple web page into a serious cockpit experience. If you love cars and code, this project is definitely one to keep on your radar.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ nicobailon/pi-better

Pi Better: Code-Aware Prompt Rewriting

Pi Better takes a simple request and rewrites it into a detailed professional instruction based on your actual code files. Instead of guessing what you need, it scans your project to find the real details, then writes a precise prompt for you to review before anything runs. This solves the problem of vague AI responses by grounding the task in the specific files and logic you are actually working on. You keep full control because the improved text lands in your input box for you to edit or discard.

πŸ†” @hackernewsgithubprojects