Skip to content
OverlordThe Docs

Overlord AI Documentation

The coordination and review layer for coding agents.
Get startedView on GitHub

Overlord is the durable work record around your coding agents. It keeps the objective, progress, questions, artifacts, delivery notes, and file-change rationale together—whether you work from the desktop app, terminal, or an agent’s own interface.

It works with the tools and subscriptions you already use. Overlord is not a replacement for Codex, Claude Code, Cursor, or your terminal; it coordinates their work across projects and over time.

How Overlord works

Section titled “How Overlord works”

You organize work as projects → missions → objectives, and each objective is handed to exactly one agent session. The agent attaches, works against your repository, posts progress, and delivers a reviewable result with per-file change rationale—all recorded on the mission so nothing is lost between passes.

One objective maps to one agent session. Amission gathers objectives plus the shared context they all read and write. Missions live inside a project mapped to a repository.

  1. Human → OverlordCreate a mission and its objectives, then launch an agent.
  2. Overlord → Agentovld protocol attach — the agent reads context, history, and artifacts.
  3. Agent ↔ RepoRead and edit files, posting update progress along the way.
  4. Agent → Overlordovld protocol deliver — summary plus per-file change rationale.
  5. Overlord → HumanReview the delivery; the mission advances to the next objective or is done.

The core model

Section titled “The core model”

Projects

A project connects a repository or other resource to the work you want to manage.

Missions

A mission is a durable ticket for a higher-level goal.

Objectives

An objective is one agent pass: plan, implement, review, or follow-up.

Deliveries

Review the result with the original request, activity, artifacts, and change rationale in one place.

Choose a path

Section titled “Choose a path”