If you’ve got 30+ repos, you know the pain: every “AI code tool” you try feels blind outside its little sandbox. RepoSwarm , a new open source tool, started as an experiment to fix that — a way to give AI agents full context across multiple repos, docs, and readmes. I’ll share what we built, how it actually works under the hood, and what surprised us when agents started reasoning across projects (both the good and the frustrating parts that still aren't great).

Topics:
* Why AI tools are blind in multi-repo setups
• RepoSwarm’s architecture: Temporal, Prompting, scraping, embedding, permissions
• What happens when agents “see” your whole org
• Practical takeaways for dev teams