📦 castorini/cmass

Testing AI on a 400 Billion Token Library

A new benchmark called ClimbMix Agentic Search Suite forces AI agents to hunt for answers inside a library of four hundred billion tokens. This massive dataset contains five hundred fifty three million documents, creating a realistic environment where finding information is genuinely difficult. The project takes existing questions and breaks them into small pieces to ensure the answers actually exist in this giant archive. Researchers found that even top-tier models struggle here, needing far more search calls to succeed. This tool helps developers test if their AI can truly dig through real world data instead of just memorizing facts.

🆔 @hackernewsgithubprojects

📦 wangqinsi1/spyrl

SpyRL: The Game That Trains AI to Write Better

SpyRL turns AI training into a high-stakes game of Who Is the Spy. Instead of relying on strict math problems, this project pits language models against each other in social deduction tasks. One player is secretly the spy, missing pieces of information, while others try to spot the fakes. This clever setup creates automatic rewards without needing human labels, allowing models to improve at creative writing and summarization. By gamifying the process, it solves the problem of teaching AI on open-ended tasks where answers are subjective. The result is models that get significantly better at storytelling and reasoning, proving that competition can drive real progress.

🆔 @hackernewsgithubprojects

📦 lykelee/cldefocus

Cldefocus Fixes AI Blurring

A new tool called Cldefocus solves a hidden weakness in AI deblurring by creating fake blurry photos that look exactly like real ones. Instead of relying on limited camera captures, this pipeline uses physics-based math to simulate how light actually bends through hundreds of different lenses. The result is a massive collection of realistic blurry images paired with their sharp originals. Models trained on this data learn to fix blur better than ever, even when they face cameras they have never seen before. The takeaway is that synthetic data, built on real optical physics, helps artificial intelligence handle the messy, out-of-focus photos of the real world.

🆔 @hackernewsgithubprojects

📦 igmunv/cryptolayer

CryptoLayer: Hide Messages in Plain Sight

CryptoLayer treats any existing messenger as a completely untrusted wire, meaning the platform cannot read your data. It inserts a cryptographic layer directly between you and the app, ensuring that encryption and decryption happen only on your device. The most surprising part is its byte-to-word encoder, which converts raw data into ordinary dictionary words to slip past automated filters. By using standard messengers purely as a dumb transport pipe, the project gives you a private channel without needing to build your own server.

🆔 @hackernewsgithubprojects

📦 ronaldo-avalos/maya

Maya: Turn Screen Recordings into Polished Videos

Maya is the macOS app that turns raw screen recordings into polished, share-ready videos. It wraps your footage in a beautiful device frame, lets you add cinematic zoom and tap animations, and exports clips ready for social media or in-app tutorials. Built for creators and developers, it solves the problem of making raw demos look professional without complex editing software. With features like adjustable pacing and custom backgrounds, it makes creating high-quality content simple and fast. Whether you are making a quick demo or a detailed tutorial, Maya helps you present your work with style and clarity.

🆔 @hackernewsgithubprojects

📦 edgetype/better-deepseek

DeepSeek Just Got A File Generator

DeepSeek can now build entire multi-file software projects and hand you a single downloadable zip archive directly from the chat window. The open-source browser extension Better DeepSeek turns the standard AI interface into a functional workspace. It solves the frustrating problem of copying code snippets one by one by letting the AI generate PowerPoint, Excel, and Word documents, or run Python scripts right inside the browser. You can even feed it whole GitHub repositories or local folders as context. The extension keeps your data local and private, so your projects stay on your machine. It transforms a simple chatbot into a practical, offline-capable development assistant that actually delivers usable files.

🆔 @hackernewsgithubprojects

📦 air-embodied-brain/zetta-embodiment

Zetta: The Robot That Fixes Its Own Code

Zetta is the robot harness that fixes its own mistakes without retraining the brain. Instead of learning from scratch, it watches failed robot attempts, groups them by failure type, and writes small, safe recovery scripts on the fly. This lets a frozen policy improve in real-time, boosting success rates to over ninety percent while speeding up processing. It’s a clever way to make physical AI adaptable and precise. If you follow robotics, this self-correction loop is the big idea to watch.

🆔 @hackernewsgithubprojects

📦 swimmingbrain/texbrain

Write LaTeX in Your Browser Without a Server

Compile professional LaTeX documents into PDFs entirely inside your browser tab. Texbrain eliminates the usual setup friction by removing accounts, installs, and backend servers. It uses web assembly to run the PDF engine locally, so your code never leaves your machine unless you explicitly push it to Git. The killer feature is the on-demand package loading, which fetches specific LaTeX libraries from public mirrors only when your document needs them, then caches them for offline use. It is like having a portable, private writing studio that works offline, tracks your version history, and compiles in seconds.

📰 https://news.ycombinator.com/item?id=49441375

🆔 @hackernewsgithubprojects

📦 tencent/wemm-embedding

WeMM-Embedding: One Model for All

WeMM-Embedding is the multimodal model that finally treats text, images, and videos as one unified language. The WeChat Vision team at Tencent built this family of tools so you can search for a video clip using a simple text prompt, or find the perfect image with a sentence description, without needing separate systems for each file type. It’s like giving a computer a single, powerful brain that understands everything you throw at it, from a quick photo to a full video clip. The coolest part? It’s so flexible that you can shrink its processing power down to a tiny fraction of its size while keeping almost all its accuracy.

🆔 @hackernewsgithubprojects

📦 sentelabsai/openexecutive

OpenExecutive: Your AI C-Suite

Stop hiring a full C-suite and start using OpenExecutive. This open source project gives you a single, coherent virtual executive that actually sounds like a seasoned leader. Behind the scenes, it orchestrates eight specialized agents covering everything from finance and product to legal and marketing. The real magic happens when you feed it your specific company documents. It then answers your strategic questions with precise advice grounded in your actual business context, not generic textbook fluff. It even remembers past decisions so you get continuity across conversations.

🆔 @hackernewsgithubprojects

📦 byrencheema/airstats

AirStats: The Mac Monitor That Uses Almost Nothing

AirStats is the menu bar system monitor that finally proves you don't need a heavy app to track your Mac's health. It lives quietly in your menu bar, reading everything from your processor to your battery temperature, yet it sips just a tiny fraction of your computer's power. Most other tools hog memory and slow things down, but this one stays invisible by using only a few megabytes. It shows your data in a clean drop-down panel or even as a widget on your desktop. The best part is how lightweight it is, running smoothly on Apple silicon without draining your battery.

📰 https://news.ycombinator.com/item?id=49343879

🆔 @hackernewsgithubprojects

📦 conardli/garden-skills

Garden Skills: The Secret Sauce for AI Design

Garden Skills is the open-source toolkit that transforms generic AI coding output into polished, professional-grade digital assets. Instead of settling for standard AI responses, this collection teaches coding agents specific design languages and workflows to handle web interfaces, stunning visual presentations, and high-quality image generation. It acts like a master designer sitting next to your code, ensuring every pixel and layout choice reflects intentional aesthetics rather than random defaults. If you want your AI projects to look like they came from a seasoned creative agency instead of a basic script, this is your new secret weapon.

🆔 @hackernewsgithubprojects

📦 rizriyz/luvus

Luvus: Your AI Agent Mission Control

Luvus acts as a live dashboard that watches your AI coding agents, showing you exactly which ones are working, stuck, or finished without you needing to guess. It solves the problem of juggling multiple terminal sessions and unknown agent states by giving you a clear visual status for tools like Claude Code, Codex, and Copilot, right alongside your code. This is useful because it turns a chaotic collection of command lines into a manageable workspace where you can track costs and session progress at a glance, ensuring you always know what your automated helpers are actually doing in the background.

🆔 @hackernewsgithubprojects

📦 nwiizo/nippo

Nippo Turns AI Work Logs Into Daily Reports

A single command turns your scattered AI coding history into a polished daily report. Nippo is a Rust tool that reads local logs from Claude Code and Codex to automatically draft your work summary, highlighting key decisions and technical insights. It handles the tedious data gathering so you can focus on the meaningful parts, like personal reflection. By simply typing one prompt, you get a structured document of what you did and why you chose certain solutions. This saves hours of manual writing while keeping your professional growth on track. It is the perfect way to document your daily engineering progress without the busywork.

🆔 @hackernewsgithubprojects

📦 bryllim/workout-guide

Workout Guide: 302 Open Exercise Illustrations

The workout guide contains three hundred and two distinct exercise illustrations, with each one broken down into three specific visual frames for a total of nine hundred and sixty six transparent images. It is a free open source package that developers can install to instantly add clear movement diagrams to their fitness apps or websites. The artwork is consistent and high quality, allowing creators to build professional features without having to draw every single pose from scratch. This project solves the common problem of finding reliable and uniform exercise images that are easy to integrate into code.

🆔 @hackernewsgithubprojects

📦 nxcoreai/everroom

Everroom: The AI Workspace That Remembers Your Decisions

EverRoom is a local desktop workspace built to solve a specific problem for AI users: the loss of context. Most AI tools start every session from a blank slate, forcing you to re-explain your project goals, past decisions, and source materials every single time. This repository creates a persistent, evidence-based layer that sits between your data and your AI agents. It ingests documents, repositories, and conversations, then organizes them into manageable project rooms with a clear audit trail.

🆔 @hackernewsgithubprojects

📦 liyana31811/codexless

ChatGPT Gets Local Hands

Hand ChatGPT the local tools already sitting on your computer, so it can actually do the work instead of just talking about it. Codexless plugs into your existing setup, letting the assistant edit files, run commands, and even control your browser right from your chat window. The genius move here is the smart escalation: it handles routine tasks using free, local capabilities, only calling in the expensive Codex model when a task genuinely requires that heavy lifting. This keeps your expensive quota reserved for the hard stuff while letting you stay in one conversation for everything else.

🆔 @hackernewsgithubprojects

📦 laurentzuijdwijk/llama.cpp

Llama.cpp Hits 4.7x Speeds on AMD

Run large language models at four times the normal speed using adaptive speculative decoding on AMD hardware. This fork of llama.cpp solves a critical bottleneck where standard AI inference tools slow down significantly when they try to verify multiple text drafts at once. By letting the system dynamically adjust the number of predicted words based on real-time accuracy, it keeps the generation process fast and efficient. On AMD Strix Halo chips, this approach turns a standard dense model into one that generates structured output at sixty-five tokens per second.

🆔 @hackernewsgithubprojects

📦 devlensio/devlensoss

DevLens: Visualize Your Codebase

DevLens is the open-source codebase visualizer that turns your messy code into a living map. Instead of staring at files one by one, you get an interactive graph where every function and route has a summary and a security check. The coolest part is that it helps your AI coding agents understand your project instantly without burning tons of tokens, making onboarding and pull request reviews actually fast and clear. If you want to see what breaks before you change it, this is your shortcut to understanding your architecture.

🆔 @hackernewsgithubprojects

📦 sentryhostclamp/instagram-downloader-plus

Instagram Downloader Plus: A Surprising Video Editor in Disguise

Instagram Downloader Plus is actually a full video editing suite for Windows, not a simple image saver. It includes a multi-track timeline with real-time preview, letting you mix cameras, sync audio, and apply color grading right inside the app. The tool focuses on making professional post-production accessible to everyday users by bundling transitions, motion graphics, and export presets for cinema or web. You just download the build, unzip it, and run the main file to start editing. It is a free, active project designed to handle complex video tasks without needing expensive software.

🆔 @hackernewsgithubprojects