# Chorus

A harness that coordinates a team of coding agents and their human reviewer through one pipeline, where agents propose work and humans verify it before it ships.

Chorus is a harness for coordinating a team of coding agents together with a human reviewer, built around what it calls the Reversed Conversation: agents propose, humans verify. It handles session lifecycle, task state, subagent orchestration, observability, and failure recovery for multi-agent, human-in-the-loop work, following an AI-DLC-style flow from idea to proposal, a document and task graph, execution, verification, and completion.

There are no fixed roles; instead, a five-by-three permission matrix grants each stage's actions (create, elaborate, draft, report, verify, close) to a human, an agent, or both, for any combination of actors. It connects to coding agents including Claude Code and others over its own daemon, and one daemon can serve several independent agents at once, each with its own key, working directory, backend, and permissions, letting agents hand work to each other by @-mention.

It is licensed under AGPL-3.0.

Website: https://chorus-ai.dev/