Agents and connectors

Overlord coordinates agent work; it does not replace your agent harness, terminal configuration, repository permissions, or subscription. Connectors let supported agents attach to missions, report progress, and deliver results back to the same work record.

Install connectors

Use the desktop settings or run the CLI:

ovld agent-setup
ovld agent-setup codex
ovld agent-setup claude
ovld agent-setup cursor
ovld agent-setup pi

ovld agent-setup by itself lists installable connectors. Add --dry-run to see the intended changes without applying them, or all to prepare every supported connector.

Launch from Overlord

Once an agent is configured, choose it for an objective in the app or launch it from the terminal:

ovld launch codex --mission-id <mission-id>

You can select a model or thinking level when the agent supports it. Overlord records that choice with the objective, so later review shows how the work was run.

Let agents report work

Connected agents use the Overlord protocol to attach to their mission, post updates, ask blocking questions, and deliver. You normally do not need to run those protocol commands yourself; they are how the connector maintains the mission record during an agent session.

If you use an agent outside Overlord, you can still create or inspect missions with the CLI and keep the resulting delivery in the same project history.

Slash commands

Installing a connector also registers a set of slash commands you can use mid-session inside the agent. Each is a thin wrapper around an ovld protocol call. The exact names depend on the agent — in Claude Code they are namespaced, for example /overlord:attach:

• attach <mission-id> — establish a persistent session with an existing mission and begin executing it.
• connect <mission-id> — route the current session onto another mission without loading its full context.
• load <mission-id> — read a mission’s details, history, and artifacts without creating a session.
• discuss-objective <mission-id> — mark a draft objective as submitted (in active discussion) without starting execution.
• create <objective> — create a draft mission from the current conversation.
• prompt <objective> — create a mission in execute and attach the current session immediately.
• add-objectives — add ordered follow-up objectives to an existing mission.
• record-work — record work already completed in chat as a mission in review, with objective, summary, and per-file rationales synthesized from the conversation and local diff.

Keep connectors current

After updating Overlord, rerun ovld agent-setup <agent> (or all) to repair or refresh the local connector files. Use ovld doctor when a connector or agent binary is not detected as expected.