Introduction

Overlord is a coordination layer for AI-assisted engineering work. It keeps the ticket, progress, review, and delivery record in one place while your agents keep working in the tools you already use.

The mental model

The ticket is the prompt.

It defines the work, captures progress, and holds the delivery record.

The agent stays where it already works.

Overlord coordinates Claude Code, Codex, Cursor, OpenCode, and other setups instead of replacing them.

Humans stay in the loop.

Progress, questions, artifacts, and review decisions come back to the same ticket.

Product surfaces

Four parts serve the same ticket-centered workflow.

Web App

Manage tickets, projects, activity, artifacts, and review in one shared place.

Desktop App

Adds local machine capabilities so Overlord can work with real repositories and terminal sessions.

CLI

Gives agents and humans a stable terminal interface for attaching, updating, asking questions, and delivering work.

MCP Server

Lets remote or hosted agents work with the same tickets and protocol without depending on the desktop app.

How it works

Step 1A ticket defines the job

The ticket is the durable unit of work, not a disposable chat thread. It keeps the objective, structure, and delivery record in one place.

Step 2An agent executes in its own environment

Overlord works with the tools you already use, including terminal agents and hosted agents. It coordinates them instead of replacing them.

Step 3Progress streams back into the ticket

Updates, blocking questions, artifacts, and session state flow back into the same ticket so humans can stay involved without hovering in the terminal.

Step 4Humans review before work lands

Review the output, inspect diffs and rationales, answer questions, and decide what should happen next.

Quick start

1. Create a project

Projects group related tickets. Link a project to a local repository in the desktop app.

2. Write a ticket

Give it a clear objective, optional acceptance criteria, and enough context for an agent.

3. Launch the agent

Use the desktop app or the CLI and MCP workflow for a terminal-first flow.

4. Review what comes back

Watch updates, answer blocking questions, and review artifacts and diffs.