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/commandThe 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
| Panel | What it does |
|---|---|
| Stats | Live system readout (GPU and memory) from the /stats feed. |
| Chat | Talk to a local Ollama model. Each exchange is recorded as a run. |
| Launch | Start an agent: pick a name, model, mode, credential filter, and goal. |
| Agents / Missions | Live agents and the mission tree of managers and specialists. |
| Session console | When a session is open, live events with inline approve/deny. |
| Pending approvals | Tool calls waiting on your decision. |
| Activity | A 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.