The control plane, checked the way it checks your servers.
Infraa's job is telling you the truth about your infrastructure, so this page holds itself to the same rule: every reading below comes from a real probe, and anything we don't measure isn't shown.
Measured for this page load.
Rendering this page probed the gateway from our servers — the same signal behind the badge in the footer. Results are shared across visitors for up to 30 seconds.
Five parts, one honest definition of healthy each.
Signs you in, serves the dashboard and API, mints the short-lived tokens your browser uses to talk to the gateway.
Self-evident: it rendered this page. A bad deployment rolls back without touching the running one.
Holds every agent and browser socket, runs the LLM worker and the playbook executor, signs approval tokens.
The probe on this page and the in-product badge both call its public /health from our servers.
The chat and analysis engine can actually reach its model provider — the part that can be down while everything else is green.
The gateway pings the active provider every 5 minutes with its cheapest real call and reports the reading on /health; the heartbeat records it.
Single source of truth — servers, proposals, runs, audit history. Point-in-time restore instead of dump files.
Every dashboard read exercises it; this page reads the gateway check log from it live.
Each agent holds an outbound connection to the gateway and reports metrics and module health every 60 seconds.
Per-server, in your dashboard. Your agents' health is your data — it's not aggregated onto a public page.
One request, no middleman.
When this page renders, our web servers call the gateway's public health endpoint directly and time the round trip. Up means the gateway answered correctly within five seconds — the same bound that keeps a cold start from being miscounted as an outage.
There's no third-party status vendor in the path and no cached verdict older than 30 seconds. What you see is what we measured.
When something breaks, it stays on the record.
During an incident we post a notice at the top of this page. When it's resolved, it moves here — permanently. The record began on Jul 17, 2026, when this page shipped; we don't claim clean history from before we kept one.
Numbers we won't show until we measure them.
Every number above comes with its methodology, and this section exists for the ones that don't have one yet. It used to be longer — uptime and incident history moved up once real measurement backed them.
The heartbeat measures from a single vantage point. Latency from your region needs a global measurement network we don't run — one probe location can't honestly speak for yours.
Your own servers' health lives in your dashboard — per server, per service, checked every 60 seconds.