Core concepts

Overlord keeps the request, the agent activity, and the final review record together. A few durable objects hold everything in place.

• Workspace: The shared home for people, projects, permissions, and execution targets.
• Project: The place where related work is organized and where repositories or other resources are linked.
• Mission: A durable goal that can contain multiple steps. It carries status, priority, shared context, history, reviewers, and delivery artifacts for a feature, bug fix, investigation, or review thread.
• Objective: One focused step in a mission, usually handled by one agent run. It is the unit of work an agent actually receives: the prompt for this pass, the chosen agent and model, attachments, and the checkpoint.
• Delivery: The agent’s handoff, including its summary, artifacts, changed files, and rationale.

The normal flow

1. Create a mission for the outcome you want.
2. Add the first objective to describe the next concrete task.
3. Choose an agent, attach any files, and launch or queue it.
4. The agent works and reports progress back to Overlord.
5. Review updates, answer questions, and evaluate the delivery.
6. The mission stays as the durable record, and you add another objective when a follow-up pass is needed.

Start with a project, create a mission, launch an objective, and review the delivery. Add another objective when the work needs a follow-up.

Missions contain objectives

Think of the mission as the durable container and each objective as the next instruction in that same work thread. When you launch an agent you send the current objective, not just a free-floating title — that is why objectives are the unit of work while the mission stays the shared-context record for the higher-level goal.

Keeping planning, implementation, review, and follow-up passes on one mission means you never have to open a new chat or restate the full history. Each objective can use a different agent or model, and every pass inherits the same context.

What this model gives you

• a shared system of record for humans and agents
• less prompt drift across tools
• clearer handoffs between human and agent work
• a reliable surface to review before anything lands

Next: Missions and objectives