KAIROS / COORDINATION FOR AGENT TEAMS

Coordinate multiple AI agents without duplicate work

Kairos gives Codex, Claude Code, and human teammates one durable view of the work. Agents take different jobs, results survive closed sessions, and human Review becomes context for whoever continues next.

START HERE Run two agents without giving them the same job

Run make quickstart from a source checkout, or follow the release-package installation guide for no-build startup. The example opens two tasks in parallel, then unlocks a final task when both results are ready.

Kairos Workflow showing two parallel tasks joining into a release plan
Two agents work in parallel. Their results meet in one final handoff.
01

Start with the problem you have

GUIDES
01Coordinate two Codex sessions without duplicate workGive each session different work, preserve both results, and open a final handoff with the shared context.↗ 02Use Task Claims for reliable agent executionKeep one clear executor, renew responsibility while work continues, and recover safely after interruption.↗ 03Choose a fixed workflow or an evolving planUse enforced dependencies for known processes and a Blackboard when the work must unfold as you learn.↗ 04Add human Review to multi-agent workPreserve decisions, feedback, evidence, and retry context without keeping an agent session open.↗
02

Go deeper

REFERENCE

Today you can coordinate work with Workflow or Blackboard, store it in SQLite or PostgreSQL, connect agents over MCP, and follow progress in the operations console. Agent Daemon includes continuous scheduling, an opt-in local Codex Adapter, and isolated execution examples. Release packaging includes Core and Daemon; additional platform/provider validation remains separate. See Current Status.