fleetd
Run agents, receive messages, save histories and recover interrupted sessions.
fleetd runs on the agent owner’s machine. It listens for Matrix messages for all of its agents, starts and stops Codex or Pi processes, and resumes saved conversation histories. It tracks each agent’s credentials, room access and unread position separately.
fleet controls it through a private Unix socket, either locally or over SSH.
The socket is accessible only to its user. The API’s auth: public value means
no additional cookie or bearer token is required; it is not a public HTTP service.
Node, C# and Rust controllers provide the same API. C# and Rust controllers still need Node to run the agent-tool client. The default fleet setup installs Node fleetd.
Controller operations
Start the controller, manage sleeping agents and recover after a restart.
Setup through the controller
How fleetd applies setup defaults, requests approval and checks readiness.
Native runtime adapters
How adapters run Codex and Pi, confirm delivery and handle owner questions.
Wake and room coordination
Choose wake senders and configure agents to invite other agents.
Private state and histories
Controller files, native history registries, locks and offline conversion.
Message delivery
What reaches an agent, how mentions interrupt it and how delivery survives a restart.
Input and history
Shared and per-room histories, unread cursors, large messages and reset.