Health & monitoring

Daedalus reports on itself. A first-party health dashboard lives at /health — no external monitoring stack to stand up, keeping the platform self-contained.

The dashboard

Open /health (behind the same access gate as the console). It polls every few seconds and shows, at a glance:

• Overall status — healthy or degraded, with uptime, turns in flight, and errors in the last hour.
• Host basics — CPU, memory and disk for the control plane, plus a top-process list.
• Backends — up/down and latency for inference (Ollama), the executor sandbox, and the BAS server, so a failure is obvious at a glance.
• Workload — agents by status, active turns, tasks by column, 24-hour spend.
• Mode — the autonomy, local-inference and auto-dispatch switches at a glance.

Endpoints

For scripts and uptime checks, three endpoints are open (behind the edge gate):

Endpoint	Purpose
/healthz	Liveness probe — returns ok and whether the scheduler is enabled.
/metrics	Prometheus-format gauges — agents by tier/status, tasks by column, active turns, spend, rate-limit utilization.
/api/health	The JSON behind the dashboard — status, uptime, host stats, backends, workload.

Self-contained by design

The platform exposes these so you can scrape them, but nothing external is required — the built-in /health page is the intended view, so Daedalus doesn't depend on a separate monitoring deployment.