composiohq/agent-orchestrator

Agent Orchestrator is a tool for running multiple AI coding agents in parallel on a software project. The idea is that instead of having one AI agent work through tasks one at a time, you can have many agents working simultaneously on different issues or tasks, each isolated in its own copy of the codebase. A central orchestrator coordinates what each agent is doing, and a web dashboard at localhost:3000 shows you the status of all agents at a glance. Each agent gets its own git worktree, which means it has its own branch and its own separate working directory so agents do not interfere with each other. When an agent finishes a task, it opens a pull request. If automated tests (CI) fail on that pull request, the agent receives the failure details and tries to fix them on its own. If a code reviewer leaves comments requesting changes, those comments are routed back to the agent automatically. The goal is that you only have to intervene when something requires a human decision, like approving and merging a pull request. The tool works with several AI coding agents out of the box, including Claude Code, Codex, and Aider. It supports running on macOS, Linux, and Windows. On Mac and Linux it uses tmux (a terminal session manager) to run agents, on Windows it uses a native process approach instead. It can integrate with GitHub for pull requests and issue tracking. Setting it up takes one npm install command and one start command. The start command points at a repository, opens the dashboard, and starts the orchestrator agent. Configuration lives in a YAML file that is generated automatically and can be edited to adjust which agent to use, how many retries to attempt on CI failures, and whether to auto-merge approved pull requests. The project is aimed at developers who want to run many AI coding tasks in parallel without manually babysitting each one. It is MIT licensed and available as an npm package.