# Agent Orchestrator for AI agents

This page is the machine-readable entry point for AI agents working inside Orchestrator (https://orchestrator.inc). Orchestrator is a desktop application and local daemon that run agents on your own machine; Orchestrator Cloud, which you sign in to from the desktop app, also runs worker and orchestrator sessions in hosted sandboxes. There is no documented public API for driving either from outside: inside a session, use the local `ao` CLI.

Orchestrator.inc (formerly Agent Orchestrator, AO), the open-source (Apache-2.0) orchestrator for 38 coding agents, including Claude Code and Codex.
Orchestrator.inc is operated by Untrivial, Inc., a Delaware corporation.

## When to use Agent Orchestrator

Reach for Agent Orchestrator when you need to:

- Hand a larger outcome to a project orchestrator agent that plans it, splits it into focused tasks, and spawns and coordinates worker agents.
- Take pull requests from first push to merge-ready without babysitting: CI failures, review comments, and merge conflicts are routed back to the agent that owns the branch.
- Get an independent, adversarial review of each PR from a built-in reviewer agent (`ao review trigger`), with its required changes sent to the owning worker.
- Run several coding agents (Claude Code, Codex, OpenCode, or any of the supported harnesses) at the same time on one repository without them stepping on each other: each agent gets an isolated Git worktree and its own branch.
- Supervise the fleet from the desktop app on macOS, Windows, or Linux, or from the iPhone, iPad, and Android apps.
- Schedule recurring agent runs (automations) that execute a prompt on a cron or RRULE schedule.
- Review diffs, PRs, artifacts, and browser previews, and monitor many concurrent agent sessions on one live board.

Agent Orchestrator is not a coding agent itself; it is the local workspace and orchestration layer the agents run in. If you are an AI agent inside an AO-managed session, use the installed `ao` CLI. To learn the product, start with the docs index at https://docs.orchestrator.inc/.

## Capabilities

Inside an Agent Orchestrator-managed session, an agent can use the installed local `ao` CLI to:

- **Sessions**: spawn worker or orchestrator sessions (`ao spawn`), list and inspect them (`ao session ls`, `ao session get`), message them (`ao send`), and claim a PR for a worker (`ao session claim-pr`).
- **Pull requests**: merge a PR (`ao pr merge`) and resolve its review threads (`ao pr resolve-comments`). CI failures, review comments and merge conflicts on a claimed PR are routed to the owning session automatically.
- **Code review**: start the built-in adversarial reviewer on a worker's PR (`ao review trigger`), then check or cancel it (`ao review ls`, `ao review cancel`).
- **Progress and artifacts**: report progress, blockers and outputs to the orchestrator (`ao report`).
- **Previews and browser**: show a URL or file in Orchestrator's browser panel (`ao preview`) and drive that page (`ao browser`).
- **Automations**: create and manage recurring prompts on a cron or RRULE schedule (`ao automation`).
- **Projects and agents**: inspect registered repositories (`ao project`) and the installed agent adapters (`ao agent ls`).

## How to operate

- Run `ao help` or `ao <command> --help` for the commands installed with the current AO version.
- The CLI is a thin client of the local daemon; do not bypass it by reading AO's SQLite state directly.
- AO's primary daemon listener is loopback-only. Do not treat this website as an API endpoint.

## What sets it apart

- **An orchestrator agent plans and delegates**: Every project gets a persistent orchestrator agent. Describe an outcome and it plans the work, spawns workers into their own branches and worktrees with the context they need, follows their progress, and coordinates follow-up.
- **The PR loop runs itself**: Orchestrator watches every pull request its workers open. Failing CI checks, review comments and merge conflicts are routed back to the agent that owns the branch, and new feedback keeps going to that agent until the PR is ready to merge.
- **Built-in adversarial review**: An isolated reviewer agent, on the reviewer harness and model you choose, reviews each PR head, posts an inline comment for every change it requires, and sends those comments to the worker that owns the PR.
- **38 coding agents, chosen per project**: Claude Code, Codex, Cursor, OpenCode, Aider, Goose, GitHub Copilot, Gemini CLI and more. Each project picks its own worker, orchestrator and reviewer agents.
- **Desktop on macOS, Windows, and Linux, plus mobile**: A desktop app for macOS, Windows and Linux, and free iPhone, iPad and Android apps to manage the fleet from your phone.

## Also shipped

- **Orchestrator Cloud**: Sign in from the desktop app, clone a GitHub repo to the cloud, and run worker and orchestrator sessions in hosted sandboxes, with Chat, terminal, Files and PR reviews. Agents keep working when your computer sleeps. Cloud agents start with Starter, from $20/month.
- **A live Kanban**: Every worker on one board, placed by its session, PR, CI and review state: working, needs you, in review, ready to merge. Cards show the latest artifacts a session produced.
- **Chat or the agent's own terminal UI**: Work with each session in structured Chat or the agent's native TUI, and switch between them without losing the session.
- **An agent-controllable browser**: Agents preview and drive the app they're building in a browser panel beside the session, with an isolated profile per session.
- **Artifacts**: An Artifacts page collects what agents produce, and each session gets an Artifacts tab. Charts, diagrams and HTML pages render inline in Chat.
- **Automations**: Schedule recurring prompts with a cron or RRULE schedule, then pause, edit and review their run history.
- **Agent management in Settings**: Install and update agent CLIs, choose which models appear, switch Codex accounts, and see each agent's health.
- **Conversation forks and side chats**: Fork a conversation natively across providers, and open independent side chats in the inspector.
- **Chat suggestions**: After a reply, Chat suggests your next message. Press Tab to use it; it never sends on its own.
- **Built for the keyboard**: 20 color themes in light and dark, and redesigned shortcuts with a full list on Cmd/Ctrl+/.

New in nightly (merged on main, not yet in stable v0.14.2):

- **Artifact viewer with version history** (new in nightly): Open artifacts full screen, edit Markdown in place, preview or restore earlier versions, and preview images, SVGs, audio and video in Files.
- **Multiple Claude Code and Codex accounts (preview)** (new in nightly): A preview behind Settings → General → Developer mode: add several subscription accounts, each kept in its own CLI config folder, pick a default for new sessions and see each session's account.
- **Mobile with the lid closed** (new in nightly): An opt-in mode keeps an Apple Silicon MacBook reachable from your phone with the lid closed. It also turns on Keep awake, and stops at low battery or high temperature.
- **Open-orchestrator shortcut** (new in nightly): Cmd+Shift+O on macOS or Ctrl+Shift+G on Windows and Linux opens the current project's orchestrator, starting it if needed.

## Orchestrator Cloud

Orchestrator Cloud is available now: sign in from the desktop app and run agents in the cloud. Cloud agents start with Starter, from $20/month. Choose Clone to the cloud to pick a GitHub repository, then run worker and orchestrator sessions in hosted sandboxes with Claude Code, Codex, Cursor or OpenCode. Each Cloud session has Chat, its terminal, a Files view of the remote workspace, and its pull request and reviews on GitHub. Plans: https://orchestrator.inc/#pricing. Cloud guide: https://docs.orchestrator.inc/guides/cloud/

## Pricing

Plans: https://orchestrator.inc/#pricing. Items marked (coming soon) are not built yet.

- Free ($0/month): Unlimited local agents on your machine; Any of the 38 coding agents, including Claude Code, Codex and Cursor; Orchestrator, workers and PR reviews; 2 devices total: any mix of iPhone, Android, MacBook, Mac mini or remote VMs; Connect on the same network, or from anywhere with Tailscale or the HTTPS tunnel; No cloud agents
- Starter ($20/month): Everything in Free; Agents keep working when your computer sleeps; Pull requests, CI fixes and reviews in the cloud; Cloud relay for seamless connectivity across devices (coming soon); Unlimited devices, including mobile (bring your own) (coming soon); 5 cloud projects; 15 simultaneous cloud agents
- Pro ($40/month): Everything in Starter; Bring your own Coder deployment; 20 cloud projects; 50 simultaneous cloud agents
- Enterprise (Custom pricing): Priority support with an SLA; Collaborate on cloud agents (coming soon); SSO and team management (coming soon); Unlimited cloud projects; Unlimited simultaneous cloud agents; No session-hour caps

## Cross-device and mobile

Manage 38 coding agents, including Claude Code and Codex, from your phone, laptop or desktop.

Platforms: Desktop app for macOS, Windows, and Linux. Mobile apps for iPhone and iPad (iOS 16.4 or later) and Android.

- [Orchestrator for iPhone and iPad (App Store)](https://apps.apple.com/app/ao-mobile/id6792552173)
- [Orchestrator for Android (Google Play)](https://play.google.com/store/apps/details?id=aoagents.dev)
- [Mobile overview](https://orchestrator.inc/mobile/): pairing, remote access and security
- [Connect Mobile docs](https://docs.orchestrator.inc/configuration/remote-access/)

From the phone you can:

- See every session on one board: A live Kanban of every worker session, grouped by working, needs you, in review and ready to merge.
- Get notified when an agent needs you: A push notification arrives when a session needs your input. Tapping it opens that conversation.
- Start new work: Kick off a new agent session from your phone, with a dictated prompt if typing is impractical.
- Chat with any agent and approve its requests: Full conversation history, streaming replies, approvals, model and provider switching, file attachments and voice input.
- Follow pull requests: Open and draft PRs with their checks, CI status and review state.
- Talk to the orchestrator: Plan with your project-level orchestrator agent and delegate new work from your phone.
- Open the live terminal: Attach to a session's terminal when you need the raw view.
- Preview running work: Open a worker's browser preview to see what it built.
- Manage every machine: Pair more than one desktop, name each one, pick a default and filter work across machines, so one phone covers the whole fleet.

How the phone connects:

- Same network (LAN): Every plan · iPhone, iPad and Android. Phone and computer on the same trusted network. The phone talks straight to your computer, so the connection stays on your network. There's no encryption: it's plain HTTP protected by the connection password, so use it only on networks you trust.
- Tailscale: Not on iPhone or iPad · Android not yet verified. Phone and computer on the same tailnet, from any network. When the computer is signed in to Tailscale, Connect Mobile adds its Tailscale address to the pairing QR. iPhone and iPad cannot use it. Android is allowed but not yet verified on a device, so test it off your home Wi-Fi before relying on it. Traffic goes through Tailscale's WireGuard tunnel between devices; the Orchestrator request itself is plain HTTP.
- Managed HTTPS tunnel: Starter and up · any network · iPhone, iPad and Android. Install cloudflared from Connect Mobile and your computer gets an HTTPS address the phone can reach from any network, after you scan the refreshed QR code. Traffic goes over HTTPS to Cloudflare's edge, and Cloudflare terminates TLS. The address can change when the tunnel restarts; scan the refreshed QR code if the old one stops working.

Your computer has to stay awake for the phone to reach it. On supported Macs, Connect Mobile can keep the Mac awake while mobile access is on. New in nightly: an opt-in mode that keeps an Apple Silicon MacBook reachable with the lid closed; it stops at low battery or high temperature.

Orchestrator doesn't have a built-in relay for reaching your computer from your phone yet. Away from home, use the managed HTTPS tunnel (Starter and up), or Tailscale on Android (not yet verified on a device; iPhone and iPad can't use it). If you'd rather agents keep running while your computer is off, Orchestrator Cloud, available now, runs them in hosted sandboxes; cloud agents start with Starter. Plans: https://orchestrator.inc/#pricing

## Learn more

- [Docs](https://docs.orchestrator.inc/)
- [Quickstart](https://docs.orchestrator.inc/quickstart/)
- [CLI reference](https://docs.orchestrator.inc/cli/)
- [Download](https://orchestrator.inc/download/)
- [Mobile apps](https://orchestrator.inc/mobile/)
- [Migrate from Conductor](https://orchestrator.inc/migrate/conductor/) ([Markdown](https://orchestrator.inc/migrate/conductor.md))
- [Site llms.txt](https://orchestrator.inc/llms.txt)
- [GitHub](https://github.com/OrchestratorInc/agent-orchestrator)