Stigmergy

stig is a version control system and communication layer built for teams of AI agents and people working on the same project at the same time. Git was designed for humans taking turns, so when several agents share a codebase they race each other: pushes get rejected as stale, rebases pile up, and one agent’s renamed function silently breaks everyone else’s build. stig fixes this by making every function, part or section of work owned by exactly one room. Agents publish their interfaces before they write code (or other "units" in a domain -- e.g. molecules for biochemistry, tensors for open-models, etc.), every save syncs instantly, and when something changes only the code that depends on it is notified, pointed at the exact line. Main only moves when the checker and tests pass, so there are no merge conflicts, and notes attached to the code become a persistent tribal-knowledge layer that carries context across sessions. Each project also gets a dashboard generated by Meta’s Muse, so you can watch exactly how the agents coordinate: who owns what, which contracts they published, which notices they received, and every land in order. stig is designed as an open ecosystem rather than a single tool. Language and domain packs plug in as small files, so it already covers 16 languages plus CAD, circuit boards, music and open-weight AI models, and anyone can add their own. We started out building a better git for TypeScript and, over 10 days for the Meta AI Hackathon, it grew into a general coordination layer for anything that can be written as text. We built stig with stig itself, with over 25,000 events in its own hub. In head-to-head runs, the same team of agents designed a drone and its circuit board for less money on stig than on git, about 25% less per accepted change. Writing a four-voice fugue, the stig team got 89% of its changes accepted versus 58% for git, with zero rejected stale pushes, and recovered faster when the theme changed mid-run. You can see a showcase of what agents did with stig here: https://try-stig.github.io/stig/showcase.html You can try out stig for yourself here: https://try-stig.github.io/stig/index.html#download