AI Repo: mvschwarz/openrig

Skyler: Hey everyone, welcome back to Git News AI — the show where we dig into the hottest A I repositories trending on GitHub so you don't have to scroll through the noise yourself. Skyler: Before we dive in, huge shoutout to today's sponsors: Ada, the platform helping teams actually integrate A I into their workflows in a meaningful way; Ago Consulting — that's "Ahh-Go" — your go-to for silicon development consulting all the way from A I down to SoC design; and Zen Semiconductor, an umbrella company with a killer portfolio of silicon-industry companies, including Ada for deploying A I at scale in literally any business — and their Vect appliance is seriously cool, running local L L Ms and vector search on dedicated hardware. Skyler: Alright, today we've got a multi-agent harness that pairs Claude and Codex together, an A I-powered cybersecurity toolkit that'll make your red team jealous, an open-source knowledge platform from Tencent that turns raw documents into a full reasoning agent, a self-hosted recipe manager with some smart A I tricks, and an open-source memory platform giving your A I agents genuine long-term recall — let's get into it. Skyler: First up — openrig, from the GitHub user mvschwarz, sitting at over seventeen hundred stars and built in TypeScript. Skyler: So the premise here is genuinely interesting — what if you didn't have to choose between Claude Code and OpenAI's Codex? What if you could just... run them both, together, as one unified system? Skyler: That's exactly what openrig does. It's a multi-agent harness that orchestrates Claude Code and Codex simultaneously, letting them collaborate on the same tasks inside a single workflow. Skyler: Think of it like having two senior engineers pair programming — except they're both A I models with different strengths, and you get to leverage both at once. Skyler: The project describes itself as a "multi-agent harness," which is a fancy way of saying it's the glue layer that coordinates these two models, manages their state, and makes sure they're working toward the same goal rather than talking past each other. Skyler: One of the more recent updates is a release called zero-point-six, and there's a fun commit message in the changelog: "works for strangers" — which I love, because that implies it went through a phase where it mostly worked for the people who built it and nobody else. We've all shipped that software. Skyler: There's also been work on the daemon layer — specifically making it more resilient when metadata mappings are missing. That's the kind of hardening that tells you this project is moving past the "cool demo" phase into something people are actually running in production. Skyler: If you're building agentic coding workflows, or you're curious about what happens when you give two powerful code models complementary roles on the same problem, openrig is absolutely worth exploring. Skyler: It's TypeScript, so if you're in the Node ecosystem this drops in pretty naturally — check the show notes for the link. Skyler: Next, let's talk about hexstrike-ai, from 0x4m4 on GitHub — and this one has absolutely exploded with over twelve thousand stars. That's not a slow burn, that's a rocket ship. Skyler: hexstrike-ai is an M C P server — that's Model Context Protocol — that gives A I agents like Claude, GPT, and Copilot the ability to autonomously run over a hundred and fifty cybersecurity tools. Skyler: We're talking automated pentesting, vulnerability discovery, bug bounty automation, and security research — all orchestrated by an L L M that can actually execute real offensive security tooling, not just talk about it. Skyler: If you've ever used something like Metasploit or Burp Suite, imagine an A I agent that can reach into that ecosystem, decide what tool to run, execute it, and interpret the results — and then chain those results into the next action. That's the vision here. Skyler: The M C P architecture is key — it's what makes this composable. Your L L M isn't just generating shell commands for you to copy-paste, it's directly bridging into real-world capabilities in a structured, agentic way. Skyler: Now, obviously, with great power comes great responsibility. This is clearly built for security professionals, researchers, and bug bounty hunters — people who have explicit authorization to test the systems they're poking at. Please use this ethically. Skyler: But from a pure technical standpoint, this is one of the most impressive examples I've seen of M C P being used to give an L L M genuine hands-on capabilities in a specialized domain. Skyler: The repository is Python-based, which makes sense given the security tooling ecosystem. Recent activity has been mostly README and documentation updates, which honestly is a good sign — it means the core functionality is stable enough that they're focused on making it accessible. Skyler: If you're in the security space — red teamer, bug bounty hunter, security researcher, or just someone who wants to understand how A I agents can interface with offensive tooling — this is a must-look. Check the show notes for the link. Skyler: Alright, third repo — and this one is massive. WeKnora from Tencent is sitting at nearly thirty-one thousand stars, built in Go, and it's an open-source L L M knowledge platform. Skyler: Here's the core idea: you feed it raw documents — PDFs, text files, whatever you've got — and WeKnora transforms them into three things simultaneously: a queryable R A G system, an autonomous reasoning agent, and a self-maintaining Wiki. Skyler: R A G, for anyone newer to the space, stands for Retrieval Augmented Generation — it's the approach where instead of relying purely on what an L L M memorized during training, you give it access to a live knowledge base it can search at query time. Much more accurate for domain-specific stuff. Skyler: But WeKnora goes further than just R A G. The reasoning agent layer means it can actually think through multi-step problems using your documents as its knowledge foundation — not just retrieve and regurgitate, but reason. Skyler: And the self-maintaining Wiki piece is genuinely clever. As the system processes more documents and queries, it keeps its own knowledge representation updated — so the knowledge base doesn't go stale the way traditional wikis do when nobody has time to maintain them. Skyler: This is clearly aimed at enterprises and teams that have mountains of internal documentation, research, or institutional knowledge that's currently locked away in files nobody reads. Skyler: The recent commits are in Chinese, which makes total sense given this is a Tencent project — and they include concurrency fixes for table creation and a keyword search fix that was silently dropping results. Those are the kinds of subtle bugs that matter a lot when you're building systems people rely on for accurate retrieval. Skyler: If you're building internal knowledge tools, enterprise A I, or just want to see how a production-grade R A G-plus-agent system is architected, WeKnora is a serious reference project. Check the show notes for the link. Skyler: Fourth on the list is Mealie — mealie-recipes/mealie — over thirteen thousand stars, Python, and yes, it is a recipe manager, but stick with me here because the A I angle is real. Skyler: Mealie is a self-hosted recipe manager and meal planner with a REST A P I backend and a Vue frontend. You give it a URL from basically any recipe website, and it automatically scrapes and imports the recipe data — structured, clean, into your personal database. Skyler: Now, the A I relevance here is in that extraction and parsing layer. Making sense of wildly inconsistent recipe page layouts across thousands of different websites is actually a non-trivial natural language and structured data extraction problem — and Mealie handles it automatically. Skyler: Beyond the import magic, you've got full meal planning, a family-friendly interface, and because it's self-hosted, your recipe data stays yours — no subscription, no vendor lock-in, no "we're shutting down your account" emails. Skyler: For developers specifically, the REST A P I makes this really interesting. You could wire Mealie into a home automation system, have an A I assistant query your recipe database, generate shopping lists programmatically — the surface area for building on top of it is pretty wide. Skyler: Recent activity is mostly dependency updates — httpx2 and pylint bumps — which is actually a good sign for a mature project. It means the maintainers are keeping the dependency graph healthy and the security posture clean. Skyler: If you're into self-hosted tools, home automation, or just want a genuinely great recipe manager that you control, Mealie is one of the best options out there. And if you want to build A I-powered meal planning features on top of a solid foundation, the API is your friend. Show notes have the link. Skyler: And finally — cognee, from topoteretes, sitting at just over thirty-one thousand stars, Python, and this one might be the most conceptually important repo in today's lineup for anyone building serious A I agents. Skyler: Cognee describes itself as the open-source A I memory platform for agents — and the pitch is simple but powerful: give your A I agents persistent long-term memory, using small models, for free. Skyler: Here's the problem it's solving. Most A I agents today are goldfish. Every conversation starts fresh. They don't remember what they did last session, what they learned, what decisions they made. That's a fundamental limitation if you want agents that actually improve over time or maintain continuity across tasks. Skyler: Cognee gives agents a memory layer — not just storing chat history, but structured, queryable, persistent memory that survives across sessions, across restarts, across model swaps. Skyler: The "small models" part of the pitch matters too. You don't need a massive frontier model to power the memory system itself — Cognee is designed to work efficiently with smaller, cheaper models for the memory management layer, keeping costs down. Skyler: And being open-source means you're not handing your agent's memory — which could contain sensitive context about users, tasks, decisions — to a third-party SaaS you don't control. Skyler: Recent activity is documentation focused — aligning the README with the one-point-six-one release and tightening up the intro. That signals a team focused on developer experience and onboarding, which is always a good sign. Skyler: If you're building A I agents of any kind — whether that's coding assistants, customer service bots, autonomous research agents — and you want them to actually remember things and get better over time, cognee is the memory layer you've been looking for. Check the show notes for the link. Skyler: And that's a wrap on today's repos — openrig for multi-agent coordination, hexstrike-ai for A I-powered security tooling, WeKnora for enterprise knowledge management, Mealie for self-hosted smart recipe management, and cognee for giving your agents a real memory. Skyler: Massive thanks to our sponsors — Ada for making A I integration actually work at scale, Ago Consulting for their deep silicon and A I expertise, and Zen Semiconductor for pushing the hardware frontier with their Vect appliance and their full portfolio of silicon-industry companies. Skyler: If you want more of this kind of content, go check out AI Special Report — another great show in the Ada podcast network — and make sure you're subscribed so you never miss an episode. All the repo links are waiting for you in the show notes. See you next time.