Skip to content

Command center

The command center is hearth’s web cockpit. From one page you can see live system state, chat with a local model, launch a sandboxed agent, and watch runs happen on the map. It is served by hearth-mapd at:

http://<host-ip>:8770/command

The denser, panel-based surface lives at /command. The default cockpit is now the visual world view at /world; the command center is the same controls in a more compact dashboard.

What is on the page

PanelWhat it does
StatsLive system readout (GPU and memory) from the /stats feed.
ChatTalk to a local Ollama model. Each exchange is recorded as a run.
LaunchStart an agent: pick a name, model, mode, credential filter, and goal.
Agents / MissionsLive agents and the mission tree of managers and specialists.
Session consoleWhen a session is open, live events with inline approve/deny.
Pending approvalsTool calls waiting on your decision.
ActivityA feed of recent runs and state changes.

Launching an agent

The launch panel posts to /run. You choose a permission mode (plan, auto, or bypass), an optional credential allow-list, and one of the run types:

  • Open session for an interactive run you steer turn by turn.
  • Run in background for a one-shot background worker.
  • Launch mission for a swarm.
  • Marathon to loop until done, optionally with Telegram check-ins.
  • Self-evolve or grow hearth for the self-improvement modes.

Approvals and the kill switch

In auto mode, a dangerous tool call pauses and surfaces in the session console or the pending-approvals panel; you approve or deny it inline. The stop-all control halts every session and worker and clears the queue at once. See Permission modes & approvals.

Chatting

The chat panel posts to /chat, which calls Ollama’s chat API and records the exchange to the audit store like any other run, so chats show up in hearth-runs and the activity feed.