CMU and Fleet AI Open-Source SwarmChat, a Slack for Agent Swarms
A CMU team (Pranjal Aggarwal, Jing Yu Koh, Sean Welleck, Ruslan Salakhutdinov, and others), together with Fleet AI, has released SwarmChat—an open-source, Slack-style chat server designed for AI agent swarms rather than human teams, installable via pip.
Confirmed
- SwarmChat offers channels, DMs, threads, kanban boards, and file sharing; each agent connects through a purpose-built command-line client, and the whole swarm can be observed running in a browser.
- The project was open-sourced jointly by CMU and Fleet AI, with several authors including Sean Welleck.
- One motivation is that agent swarms have already produced hard-to-predict behaviors, such as the HuggingFace incident, so tools are needed to observe how swarms collaborate collectively.
- Relay posts mention the team burned enormous numbers of tokens studying swarm scaling laws, with inconsistent figures across posts ("five trillion tokens" vs. "500 billion tokens"); the exact scale should be verified against the project's original materials.
Why it matters
- SwarmChat turns multi-agent collaboration from researcher-hand-orchestrated pipelines into persistently observable, message-based interaction, providing infrastructure for studying emergent behavior and scaling laws in agent swarms.
2026-10-29 ~ 2026-10-30 · 6 related posts
Primary sources
- Team burns half a trillion tokens to study agent swarm behavior with SwarmChat — wellecks · 2026-10-29
- [source] CMU Team Open-Sources SwarmChat, a Slack for AI Agent Swarms — wellecks · 2026-10-30
- CMU Releases SwarmChat, an Open-Source Slack for AI Agent Swarms — niloofar_mire · 2026-10-30
- CMU open-sources SwarmChat and burns half a trillion tokens to map how agent swarms scale — wellecks · 2026-10-30
- SwarmChat: Open-Source Slack for Agents, 500B Tokens Shows Swarms Scale 6x Faster — wellecks · 2026-10-30
1 near-duplicate retellings: wellecks