Documentation

Moa is a coding agent that runs on the machine where your code already is. These pages cover getting it running, driving it from each of its three interfaces, and deciding what it is allowed to touch.

New here? Start with Quickstart, then read the security section before you let anything reach it over the network.

Start here

• Overview What Moa does, how the agent loop works, and where it keeps its state.
• Quickstart Install Moa, authenticate against your provider, and run the first session.

Using Moa

• Web UI Run the browser and mobile interface: panes, attachments, voice, and how to expose it safely.
• Owners A standing agent per codebase: keeps the project book, starts and directs its sessions, and answers what the book already answers.
• CLI reference Every flag, subcommand, and model alias accepted by the moa binary.

Reference

• Configuration Global and per-project config: permissions, sandboxing, limits, models, and MCP servers.
• Tools The built-in toolbox, memory, subagents, custom script tools, and the sandbox.

Automation

• Automation API Start and steer agent runs over HTTP: authentication, idempotency, callbacks, and the security model.
• Recipe: Linear Wire Linear issues to Moa runs end to end with the Automation API.

Under the hood

• Architecture Package map, event bus, agent loop, and the design constraints behind them.
• Releases Versioning, the release checklist, update checks, and distribution channels.

Every page here is the markdown that ships in the Moa repository, published from the latest release tag. Spotted something wrong? The “Edit page” link at the bottom of each page goes straight to its source.