Transcription
Welcome back to GitHub Awesome, the channel where we explore the new projects lighting up GitHub every day. Today's episode, GitHub Trending Today number three, brings you 19 brand new open-source projects across AI, dev tools, and more. Let's dive into today's trending repos.
First up, Katakotti lets you run self-hosted secure VM sandboxes for AI compute at scale. Think of it as your own infrastructure to safely execute untrusted code whether via CLI API or Python SDK. Built on battle tested VM isolation with Kada, Firecracker, and Kubernetes. Katakate makes it easy to create, manage, and orchestrate lightweight safe VMs. Perfect for AI agents, custom serverless setups, hardened CI/CD runners, or blockchain execution layers. It's designed for maximum speed, minimal footprint, and security.
Next, check out Super Cedar, a fullfeatured Bit Torrent client that runs entirely in your terminal built in Rust. It's currently in late alpha, but already supports everything you'd expect from a serious torrent client. Trackers, DHT, PEX, magnet links, and even a rarest first piece selection strategy for optimal swarm performance. The real magic though, that gorgeous real-time terminal UI showing peer stats, network graphs, and download progress, all with buttery smooth performance. It's a fast torrent engine with full persistence, configurable speed limits, and network history tracking.
Moving into the world of motion and embodiment, meet Embody 3D, a massive new multimodal human motion data set released by Meta's CODC Avatars Lab. It captures over 500 hours of 3D motion from 439 participants across more than 54 million frames. Everything from expressive hand gestures and walking sequences to multi-person interactions, emotional conversations, and even co-living scenarios. EmBody 3D isn't just about animation. It's a breakthrough data set for AI, AR, and VR researchers, combining 3D body tracking, hand motion, text annotations, and synchronized audio.
Next up, Find PDFs, an open source data set project that's literally unlocking the hidden web inside PDFs. This repo contains the full endto-end pipeline to extract, OCR, clean, dduplicate, and classify text from millions of PDFs. A process so painful most teams just avoided it entirely. But fine PDFs went allin, liberating over 3 trillion tokens from 475 million documents across 700 plus languages. The result, a data set that rivals top tier web corpora like small LM3's Mixture and even boosts performance when combined with HTML data sets.
Next up, meet Wami, an independent 3D interactive AI companion library that brings your digital assistant or chatbot to life with visuals, sound, and personality. Wami Kwaami lets you create an AI blob, a morphing real-time reactive 3D avatar that moves with sound, responds to user interactions, and integrates seamlessly with LLMs, text to speech, and speech recognition. It's modular, fast, and visually stunning. Complete with shaderbased skins, real-time audio visualizations, and full WebGL performance.
Next up, a classic reborn. Tiny Tiny RSS or TTRSS is one of the most respected open- source RSS readers on the web and it's just been forked to continue its legacy after the original project sunsets in late 2025. TTRSS lets you self-host your own personalized feed reader, giving you full control over your news, podcasts, and privacy with features like folder organization, article dduplication, full content embedding, podcast playback, and even a JSON API for custom integrations. The new fork, led by one of the longtime contributors, ensures ongoing updates, Docker support, and plug-in compatibility, keeping this beloved RSS platform alive for years to come.
Next up, say hello to Lightly Studio, the brand new open-source data platform from the team behind Lightly. It's built to unify everything in your computer vision and multimodal ML workflow, from curation and annotation to embedding management. Rebuilt from the ground up in Rust, Lightly Studio delivers instant search, AI-driven filtering, and native multimodal support across images, text, and 3D data. It's like the ultimate control center for your data sets. Visual, intelligent, and blazing fast, even on a MacBook.
Next, let's talk about a tool every Mac OS developer is going to love, Devstrip. It's a sleek Rustpowered command line utility that helps you reclaim gigabytes of wasted disk space by cleaning up stale build products and language specific caches. Devstrip automatically scans your system for things like Xcode derived data, homebrew caches, node modules, Python builds, and other heavy project artifacts. Reports how much space you can recover and lets you safely delete them with a single command.
Next, let's get into the front-end engineering zone with Domoscope, an advanced HTML diff built in TypeScript that takes content comparison to a whole new level. Unlike basic text diffs, Domoscope actually understands DOM structure, intelligently matches elements across versions, and gives you precise structured insights into what changed from attributes to text content. It's perfect for CMS systems, version control interfaces, collaborative editors, or visual regression testing tools.
Next up, meet Fenox, your personal Fort Knox for secrets. It's a developer friendly tool that unifies how you store, encrypt, and access secrets across Git cloud providers and local environments, all from one simple interface. Whether your secrets live encrypted in Git or remotely in AWS, Azure, GCP or Vault, FFNOX makes them instantly accessible with a single command or automatic shell integration when you enter a project directory. It supports modern encryption with age, integrates with one password and Bit Warden and even your system keychain.
Next up, Deepseek OCR app. A quick vibecoded app for Deepseek OCR powered by React and Fast API. Deepseek OCR offers four powerful modes. Plain OCR for raw text extraction, describe for generating smart image captions, find to highlight specific terms visually, and free form for custom prompt-based OCR tasks. With its glass morphism UI, drag and drop uploads, animated transitions, and real-time bounding box visualization. This app feels as smooth as it looks.
Next up, chunk data. A tiny utility for anyone dealing with file uploads, streams, or progress tracking. Have you ever noticed your upload progress jumping straight from 0% to 100%? That's because your app is sending one massive chunk of data instead of smaller pieces. Chunk data fixes that by splitting buffers, iterables, and streams into consistent, manageable chunks, so your HTTP uploads report real progress updates. It's lightning fast thanks to zero copy views using subarray, meaning it can chunk hundreds of megabytes in under a millisecond.
Next up, autopage, a system that turns academic papers into beautiful interactive web pages for under 10 cents. Created by the codec avatars lab at Meta, Autopage tackles one of academia's biggest bottlenecks, turning dense technical papers into accessible web experiences. Instead of relying on static PDFs or clunky manual design, Autopage uses a multi-agent humanin in the loop pipeline that blends automation with intelligent verification and optional human feedback. The result, a polished publication ready project web page crafted collaboratively by humans and AI, all in under 15 minutes and for less than a dime.
How far are today's AI image editors from true physical realism? Meet Pikbench, a new benchmark and data set that challenges models to go beyond surface level edits. Developed as a complete framework, Pikbench combines three key components. Pika bench, a benchmark featuring physicsaware editing cases across eight fundamental laws. Pika evil, a rigorous evaluation metric that checks edits through region grounded questionans answer verification with humanlabeled ROIs and PIA 100K, a massive synthetic data set derived from videos to train models for better physical consistency.
Ever wondered what happens when two AI models start talking to each other with no human in between? Introducing Autopilot chatting agents, a streamlit app that lets you watch two local Alama models have a live unscripted conversation. You can pick which models you want to pair up, start or stop the chat anytime, and watch their dialogue stream in real time, like two digital minds debating, collaborating, or just going off on tangents. And when the chat's over, you can even save the full conversation as a markdown file for later analysis or sharing.
Meet Auto Steer, your personal co-pilot for managing multiple cla code workspaces with total control and zero context loss. Built with Electron, Auto Steer runs natively on Mac OS, Linux, and Windows via WSL, giving you a unified desktop native way to organize all your projects. Its workree first architecture keeps every project isolated, each with its own file system, context, and chat history. That means no more mixing prompts or losing track of sessions. With persistent sessions, you can stop working, close the app, and pick up right where you left off. Context, messages, and even token stats intact.
Meet Adabra UI, the GPU accelerated UI library built for the Z editor's GPUi framework. Inspired by Shadcen, Adabra UI brings the same developer-friendly polish to desktop applications with over 70 plus fully accessible production grade components all written in Rust. It's got everything you need to build stunning native interfaces, a complete theme system with semantic color tokens, responsive layout utilities. The built-in typography system ensures consistent text rendering, while a fullfeatured code editor component comes with syntax highlighting and keyboard navigation out of the box.
Meet animated glass tabs, a Swift UI component that brings liquid glass elegance to your iOS apps. Designed for iOS 26 Plus, it's a custom tab view with a sleek glass morphism effect and beautifully animated icons that scale, rotate, and shimmer with every tap. Built entirely with Swift UI native APIs, animated glass tabs delivers a smooth, modern look. No external dependencies, just pure Swift magic. It's fully customizable, so you can tweak colors, animations, and layout to match your app's vibe.
Introducing Open Threat, the open-source platform that's democratizing threat intelligence. With over 300,000 CVEes aggregated from trusted sources like NVD, SISA and BSIERT bund open thread brings all critical vulnerability data into one clear actionable dashboard. Track actively exploited vulnerabilities. Explore LLM powered summaries and prioritize threats using a smart scoring system that blends exploitation data, CVSS, and recency. Developers and analysts will love the powerful search 20 plus REST API endpoints and RSS feeds for real-time monitoring.