MAO vs Orchestrator.inc (formerly Agent Orchestrator): which should you use to run coding agents in parallel?

Orchestrator.inc, the new name for Agent Orchestrator (AO), whose GitHub repository used to live at ComposioHQ/agent-orchestrator, is a free, open-source (Apache-2.0) desktop app for macOS, Windows and Linux: an orchestrator agent plans the work, starts a worker for each piece in its own Git worktree, and routes CI failures and review comments back to the worker that owns the pull request. MAO is a desktop canvas for the same three systems where each agent runs as a real terminal node, and agents you wire together hand tasks to each other and drive a browser or a phone. Choose Orchestrator.inc if you want a free, open-source path from task to merged pull request with phone apps; choose MAO if you want to see and wire the agents yourself, test on Android or the iOS Simulator, or run a workspace on your own server.

What is Orchestrator.inc (formerly Agent Orchestrator)?#

Orchestrator.inc is a desktop app for supervising AI coding agents, known until its rename as Agent Orchestrator or AO.

Its docs describe it as an app that gives each worker an isolated workspace, keeps its conversation or terminal available, and connects Git work to pull requests, CI, reviews and merge conflicts. Each Git-backed session gets its own worktree and branch, based on the repository's default branch. The basic loop is: add a project, start a session, work in an isolated worktree, open a pull request, go through CI and review, merge, clean up.

You tell a project's main agent what you want built; it plans the work and starts a worker for each piece. Sessions sort themselves into lanes on a board (Working, Needs you, In review, Ready to merge). Failed checks, review comments and merge conflicts go back to the worker that owns the branch, and a reviewer agent can read each pull request before a person does. Each session has an inspector with its pull request, reviews, files and a browser preview that agents can drive.

It supports 38 coding agents, including Claude Code, Codex, Cursor, OpenCode, Aider, Goose, GitHub Copilot and Gemini CLI, each through the CLI and sign-in already on your machine; worker, Chat and reviewer support varies by agent. A session runs either in the agent's own terminal interface or in a structured Chat. Mobile apps for iPhone and Android pair with the desktop to answer agents, approve requests and review pull requests, and paid Cloud plans run workers in hosted sandboxes. The local app is free and needs no account to run agents locally.

How do MAO and Orchestrator.inc differ?#

Both run on macOS, Windows and Linux and can isolate agents in Git worktrees; they differ on how an agent runs, how work moves between agents, and where it ends.

When does Orchestrator.inc fit better?#

When the work should end as reviewed pull requests, and you want a free, open-source tool to follow them there.

• Free and open source. The local app is free under Apache-2.0, with unlimited local agents and no account needed to run them.
• A pull-request loop. Failed checks, review comments and merge conflicts go back to the worker that owns the branch until the pull request is approved, and a reviewer agent can review each pull request first.
• Planning by an agent. A project's main agent splits the task and starts the workers, so you describe the outcome instead of placing each agent.
• Many agents, chat or terminal. It lists 38 coding agents, and agents with a Chat driver can switch between Chat and their own terminal interface without changing the session or worktree.
• Phone apps. The iPhone and Android apps send a push notification when an agent has a question, and let you reply, approve, start a worker and review pull requests. MAO does not document a phone app.
• Cloud agents. On a paid plan, workers run in hosted sandboxes and keep going with your laptop closed.

When does MAO fit better?#

When you want to see the agents and the connections between them, test on phones, or keep the work on your own machines.

• Agents you wire yourself. Each agent is a real terminal on a canvas next to notes, files, Git changes and portals. You draw the connections, and a connected agent can send another a task with mao ask and get the result back by task ID.
• Phones and simulators for agents. Besides a browser, a wired agent can drive an Android emulator or phone, and on a Mac an iOS Simulator. Orchestrator.inc documents a browser preview but no device testing.
• Your own server. A MAO workspace can live on your Linux server over SSH, or inside WSL on Windows, with its terminals, agents and layout on that machine. Orchestrator.inc's docs say not to expose its daemon on a remote server.
• Landing without a pull request. A MAO layer lands into Ground on your machine after a check for conflicts and uncommitted changes. Orchestrator.inc's loop is built around pull requests on GitHub or GitLab.
• No telemetry by default. MAO sends no telemetry unless you opt in. Orchestrator.inc's production builds send usage telemetry, including your GitHub username, until you turn it off in its settings.

How do I compare them on a real task?#

Run the same two-part task in each, in a repository on GitHub so Orchestrator.inc can follow its pull requests.

1. 1Pick a task with two parts, such as an API change and its tests.
2. 2In Orchestrator.inc, add the repository, describe the task to the project's main agent, and watch it start workers in their own worktrees and open pull requests.
3. 3In MAO, create a layer for each part, place an agent on each, connect them, and have one send the other its part with mao ask.
4. 4Have each agent check a web change in the built-in browser: the session's browser preview in Orchestrator.inc, a browser portal in MAO.
5. 5Compare how a failure reached the right agent, how you followed the work, and what it took to merge it.

Frequently asked questions#

Sources#

• Orchestrator.inc README on GitHub: the orchestrator and workers, worktrees, the board's lanes, CI and review feedback, the reviewer agent, supported agents, the browser per worker, the mobile apps, Cloud Agents, downloads, telemetry and the licence.
• Orchestrator.inc LICENSE: Apache License 2.0.
• github.com/ComposioHQ/agent-orchestrator: answers with a 301 redirect to github.com/OrchestratorInc/agent-orchestrator.
• Orchestrator.inc website and its llms.txt: "the new name for Agent Orchestrator (AO)", 38 coding agents, free and open source, no proxying of API calls, cloud agents.
• Orchestrator.inc pricing: Free, Starter, Pro and Enterprise plans; limits apply only to cloud agents.
• Orchestrator.inc docs: Introduction: the basic loop, Chat and Terminal UI, local data, supported agents.
• Platforms and Runtime: desktop packages, native terminal hosts, ConPTY, tmux as a fallback, the mobile companion, remote and server use.
• Agent adapters: Terminal UI, Chat and reviewer support per agent.
• Git worktrees: one worktree and branch per session, based on the default branch.
• Dashboard: the inspector and the browser preview that agents can drive.
• CLI reference: ao send and ao preview.
• Cloud sessions: hosted sandboxes and Clone to the cloud.
• FAQ: not exposing the daemon on a remote server, cost, telemetry and how to opt out.
• MAO home and pricing: platforms, the one-time price and the 7-day free trial.
• Run a shell or agent and Switch an agent node to chat mode: terminal nodes and chat mode.
• Send a task and collect its result and Manage agents as an Orchestrator: mao ask, mao send, mao result, recruiting agents.
• Create a worktree layer and Land a layer into Ground: layers as Git worktrees and the checks before landing.
• Browser and device portals and View an iOS Simulator portal: browser, Android and iOS Simulator portals.
• Remote AI agent workspaces: workspaces on a Linux server over SSH and inside WSL, and the licence they need.
• Opt in to diagnostics: MAO's diagnostics are off unless you turn them on.

Checked on 2026-10-28.