Run 30 AI agents in parallel.
One dashboard.
Agent Orchestrator spawns Claude Code, Codex, Cursor, Aider, and OpenCode in isolated git worktrees. Each agent gets its own branch, creates PRs, fixes CI, and addresses reviews autonomously.
You're running AI agents in 10 browser tabs. Checking if PRs landed. Re-running failed CI. Copy-pasting error logs.
Agent Orchestrator replaces that with one YAML file. Point it at your GitHub issues, pick your agents, and walk away. Each agent spawns in its own git worktree, creates PRs, fixes CI failures, addresses review comments, and moves toward merge. If you are new, start with the docs quickstart and configuration guides.
agent: claude-code tracker: github workspace: worktree runtime: tmux notifier: slack
Multi-agent execution
Run Claude Code, Codex, Cursor, Aider, and OpenCode simultaneously. Each agent gets its own git worktree, its own branch, its own context.
Autonomous CI + review handling
CI fails? The agent reads the logs and pushes a fix. Review comments land? The agent addresses them. You sleep, your agents ship.
7 swappable slots
Runtime, Agent, Workspace, Tracker, SCM, Notifier, Terminal. Use tmux or process. GitHub or GitLab. Slack or webhooks. Swap anything.
Real-time Kanban + terminal
Every agent's state in one view. Attach to any terminal via the browser. SSE updates every 5 seconds. WebSocket for live terminal I/O.
From issue to merged PR
What teams run with AO
Clear a bug backlog overnight
10 issues, 3 days of context-switching
10 agents, 10 PRs by morning
Ship a feature sprint in hours
5 feature tickets, 1 dev, 1 week
5 agents in parallel, PRs landing same day
Migrate an API across 20 files
Manual find-and-replace, missed edge cases
Agent rewrites, runs tests, fixes failures, opens PR
Three steps to orchestration
Configure & assign
Point Agent Orchestrator at your repo with a YAML config. Choose your agent, set up trackers and notifiers. One file, full control.
Agents work
Each agent spawns in an isolated worktree. They write code, create PRs, run tests, and fix failures. Monitor everything from the live dashboard, or let them run.
PRs land
Agents create pull requests, address review comments, fix CI failures, and get them to mergeable state. Your morning starts with merged PRs, not a backlog.
The only open-source, web-based agent orchestrator
Conductor, T3 Code, and Codex App are native Mac apps. AO runs in your browser, works on any OS, and you can self-host or extend it.
| Feature | AO | Others |
|---|---|---|
| Web-based dashboard | ✓ | Native Mac apps only |
| Open source (MIT) | ✓ | Closed source |
| Multi-agent (Claude, Codex, Aider, OpenCode) | ✓ | Single agent |
| Auto CI failure recovery | ✓ | Manual |
| Plugin architecture (7 slots) | ✓ | Fixed integrations |
| Git worktree isolation | ✓ | Shared workspace |
Trusted by builders
“Set up 12 agents on our backlog before lunch. By end of day, 8 PRs were merged.”
“The auto CI recovery alone saves me hours a week. Agents fix their own broken tests. I just review and merge.”
“We went from 3 PRs/day to 15 PRs/day. The plugin system means we swapped in GitLab and Linear without changing our workflow.”
Three commands to launch
Install
One command. No dependencies beyond Node.js.
Configure
Create an agent-orchestrator.yaml. Pick your agents, tracker, and notifiers.
Launch
Assign issues and watch agents spawn.