World view
The world view is hearth’s default cockpit. Instead of a list of rows, your agents are characters in a small top-down facility you can watch in real time. It is served at:
http://<host-ip>:8770/worldOn the desktop it opens automatically at login in a
dedicated window, and Meta+A toggles a fullscreen view.
What you see
- Agents as sprites. Each running agent is a character. Managers, specialists, and the growth daemon look different (the growth daemon carries a spinning gear). Their appearance is driven only by the agent’s real runtime state, never by model output: thinking, running a tool, waiting on I/O, waiting for your approval, errored, or done.
- Crew rooms and tethers. A manager and its specialists are grouped into a labelled room (the room shows the mission goal), with dashed lines tethering children to their parent. Ungrouped agents sit in a free-agent zone.
- A HUD across the top with live GPU, VRAM, and RAM bars and an agent count.
- A minimap when the facility gets busy.
- CRT effects (scanlines, vignette, a subtle flicker) you can toggle.
What you can do
- Launch from the goal bar at the bottom: type a goal and start a mission (swarm), a single agent, or the growth loop.
- Click an agent to open an inspect panel with its details and a scrollable transcript of its recent steps.
- Open the ledger panel for the self-improvement view: the growth daemon’s status, validated and merged counts, the lessons log, and the promote controls (review diff, build-check, apply to live, roll back).
How it stays live
The page streams agent state changes over server-sent events (/events) and polls
/tree, /state, and /stats for lineage, current state, and system load. The
result is that a mission visibly plays out: a manager spawns a room of
specialists, each lights up as it thinks and runs tools, and the room resolves as
they finish.
The command center remains available at
/command as a denser, panel-based control surface for the same actions. For the
underlying HTTP surface, see Map dashboard.