Live activity

Path: /live (also the dashboard's default landing page when boards are not configured)

A real-time view of every reachable daemon: which agents are online, what they're doing right now, and the last handful of tasks. Driven by an SSE snapshot from /api/live/stream.

What you'll see

• Top bar: peer chips (one per dashboard.daemons[] entry plus auto-discovered mesh peers). Click to filter.
• Agent grid: one card per agent, sized by recent activity. The card shows: status (idle / busy / error), current task preview, last message timestamp, queue depth.
• Recent tasks rail (right): a chronological feed of completed and in-flight tasks across the fleet.
• Channel badges on each card: which Telegram/Discord/Slack/WhatsApp/GitLab bindings the agent owns.

The view is "skim friendly" by design — scan the grid, spot the agent that's stuck, click into the per-agent page to investigate.

What you can do

• Click an agent card to open /admin/agents/<id> (the per-agent page).
• Click a task in the rail to open the same page scrolled to that task in the history.
• Filter by peer via the top-bar chips when running multi-node mesh.
• Watch the SSE stream in dev tools: every change pushes a delta — no polling.
• Stop a running task — click ✕ stop on the running-task card. Sends SIGTERM to the underlying claude subprocess (SIGKILL after 3 s) and the task surfaces errorKind: "cancelled".
• Update / append a message — click ✎ update, type the message. The current turn finishes, then your message dispatches as the next turn on the same chat session (same model as sending a new message to an ongoing Claude session). On persistentProcess: true agents the warm handle and prompt cache are preserved. If you also want to kill the current run, click Stop separately. See the HTTP API reference for the underlying endpoints.

Common tasks

You want to…	Do this
See which channel routed a message	Hover the channel badge on the agent card — tooltip shows the route + chat id
Know if a peer is unreachable	Peer chip turns red. Hover for the last error
Tail the daemon log instead	The CLI is faster: agentx daemon logs -f

Troubleshooting

• Empty grid. Check dashboard.daemonUrl resolves and the daemon is up (curl $DAEMON/health). If the daemon is on another machine, set dashboard.daemons[] (see config schema).
• Stuck "loading…" spinner. The SSE stream couldn't connect. Look in the browser console for CORS errors; if you bound the daemon to a Tailscale IP, the dashboard's daemonUrl must be the same scheme/host.
• Cards missing channel badges. channels.<name>.enabled is true but no routes[] reach the agent. Check agentx channel list or the Admin → Channels tab.

Implementation pointers

• Page module: src/daemon/ui/pages/live.ts (skeleton + JS)
• API: GET /api/live/stream (Server-Sent Events)
• The live snapshot is computed from the same data the agentx daemon status and agentx daemon watch CLIs surface — single source of truth.

HTTP API

Endpoint	Purpose
POST /api/tasks/:taskId/cancel { reason? }	Stop an in-flight task. 404 if no running task matches taskId (use GET /agents → runningTasks[].id to find it).
POST /api/tasks/:taskId/followup { message, replace?, sender? }	Inject a correction/update. replace=false (default) queues behind the running task; replace=true cancels the run first and dispatches the message immediately.
POST /api/task/action?node=<url>&task=<id>&kind=cancel|followup	Dashboard-side proxy used by the Live page's Stop / Update buttons — forwards to the originating daemon with the configured operator token.