Health endpoints
GET /api/health reports whether the daemon is answering. Its response depends on caller authentication.
Public liveness
Section titled “Public liveness”An unauthenticated caller receives ok, a process instanceId, and safe Slack connection state (status and connected). It does not receive the gateway path, workspace identity, authentication configuration, detailed errors, or engine login information.
For a local operator check:
curl --fail http://localhost:4747/api/healthUse the configured host and port. A successful response with connected: false means the daemon answered but Slack is not connected; do not treat those as the same health state.
Authenticated readiness
Section titled “Authenticated readiness”An admin session receives revision, sanitized update state, engine health and enablement, warm sessions, dropped-write counters, full platform snapshot where exposed, and container-runtime status. The detached updater can obtain readiness through a narrower same-machine authenticated path; that does not make detailed health public.
instanceId changes with the daemon process, allowing the UI to detect a restart even if it occurs between two successful polls. droppedWrites can reveal best-effort accounting/event persistence failures that otherwise make totals incomplete.
Network and security
Section titled “Network and security”Health still follows the deployment’s Host/Origin guard. Configure Settings → Connection → Public URL for the trusted reachable origin when using a tunnel or reverse proxy. Public URL does not open a port or create a tunnel. The HTTP run API token is not a general admin credential and does not authorize all detailed health/admin reads.
Limits
Section titled “Limits”Liveness alone does not prove chat admission, a model entitlement, connector authentication, or successful engine execution. Test the applicable workflow separately. Beta platforms have their own connection and delivery state; public Slack health is not a universal transport report.