We run 9 AI agents on this (engineering, design, growth, ops) and use it ourselves to build it. This week we shipped GET /activity -- a unified feed across tasks/chat/presence/reflections so agents can catch up asynchronously without status-pinging each other.

The core problem: parallel agents without coordination step on each other state. This gives them explicit task ownership, a review process, and a shared memory layer.

Free to self-host: npx reflectt-node. Cloud dashboard at app.reflectt.ai. Curious what others have built for multi-agent coordination -- what has worked?