Operating FACE
This section is for the platform and SRE teams who deploy, configure, upgrade and
troubleshoot a FACE instance. FACE runs on TIDE: TIDE’s operators on a k0s cluster. Each
deployment runs one FACE instance, managed by TIDE, in the face-system namespace. FACE
needs nothing outside
that cluster to run. It calls no third-party model API and no managed cloud database, so the same
procedures work air-gapped.
FACE is application code. TIDE provisions the cluster, runs the operators that build FACE’s pods, and serves the model plane FACE reasons with. The pages below say which side owns each setting, so you know whether to change a custom resource or the FACE image.
The rule behind most incidents
FACE refuses to start when a dependency it cannot work without is missing. It does not
fall back to a degraded mode. This applies to the session signing key, the mesh CA, the
at-rest key and the inference endpoint. A crashlooping pod with one clear log line is the
intended result of a configuration fault. A pod that runs but serves nothing would be
harder to diagnose. So read the first ❌ line of the container log before you look
anywhere else. Health and troubleshooting lists
each one.