Where Agents Learn to Work as a Team
The hard problem in AI isn't building one good agent. It's getting twelve of them to not step on each other.
How It Works
Three steps from tribal knowledge to shared coordination intelligence.
Capture your AI coordination intelligence in a structured OOS file with confidence ratings, evidence types, and failure modes.
Start publishingSee what other organizations do differently. The diff engine shows what's unique, similar, and where you conflict.
Browse & compareImport proven coordination patterns. The Intelligence Graph surfaces patterns no single organization could see alone.
View the graphKnowledge as structured claims
Every operational lesson becomes a claim with confidence ratings, evidence types, and documented failure modes. No more tribal knowledge locked in one person's head.
Learn how to create yoursEvery agent writes to a shared state file. No agent reads data sources directly.
Why: Direct data source access creates race conditions and inconsistent views across agents.
Failure mode: Two agents read the same data source at different times, get different results, make conflicting decisions.
From the Blog
Building in public. Lessons from running 14 agents in production.
The Intelligence Graph
See how coordination patterns connect across organizations. Discover shared operational truths and unique approaches.
Explore the GraphBe one of the first 50
Founding Publishers get a permanent badge, early access to Phase 2 features, and shape the protocol itself.
Publish Your OOS