LUMOVI_CONTEXTpicks the context to show; without it, the kubeconfig’s current context.- The cluster is called by its context’s name, unless
LUMOVI_CLUSTER_NAMEsays otherwise. - Lumovi listens on port 8080 (
LUMOVI_PORT), and answers/healthzfor health checks. WhenLUMOVI_PORTisn’t set, it listens onPORTif that’s set, as platforms that run containers set it.
Whose credentials it uses
That depends on how people sign in, exactly as in a cluster:- With tokens (the default), people’s requests carry their own tokens. The kubeconfig only says where the cluster is and how to trust its certificate.
- With single sign-on or a proxy, Lumovi uses the kubeconfig’s credentials to impersonate people. Give it a service account’s, with permission to impersonate users and groups, and nothing more.
*-data fields. The image runs as a non-root user (UID 65532), so make sure that user can read what you mount.
With single sign-on
-e LUMOVI_OIDC_CLIENT_SECRET without a value passes the variable on from your shell, so the secret stays out of your shell history. Put TLS in front of Lumovi with a reverse proxy, and set LUMOVI_URL to the address people open. See Single sign-on for registering Lumovi with your provider.
A fleet
To show many clusters, give Lumovi a kubeconfig whose contexts are the clusters, with credentials that may only impersonate, instead ofKUBECONFIG. It needs no Kubernetes of its own, and single sign-on or a proxy, as a fleet always does:
LUMOVI_FLEET_KUBECONFIG instead, as YAML or encoded in base64, as on Sevalla. Clusters it can’t reach join through an agent that dials it, listed in LUMOVI_FLEET_AGENTS. See A fleet of clusters.
Keep the audit log
Without a folder for it, Lumovi keeps its audit log in memory, from when the container started. Every event is also a line of JSON on the container’s output. To keep its history, give it a volume, andLUMOVI_AUDIT_DIR:
/var/lib/lumovi/audit belongs to Lumovi’s user (UID 65532), so a new named volume mounted there does too. A folder of your own on the host must be one that user may write to, or Lumovi doesn’t start, and says why. One container keeps a volume at a time: a second on it waits for the first to stop, and doesn’t start if it’s still running after about 40 seconds.
Behind a company’s proxy
Give it the proxy, and mount the certificate authority a proxy that inspects HTTPS signs with:proxy-url goes through that one instead. See Proxies and certificate authorities.
Pin a version
Without a tag, Docker pullslatest, the latest release. Pin a version to know what runs:
Keep people signed in
Without a folder of its own, Lumovi keeps sessions in memory, and restarting the container signs everyone out. Give it a volume, andLUMOVI_DATA_DIR, and a restart signs nobody out:
state.json there, which only its user can read: who’s signed in, and the AI assistants they allowed, each entry sealed with a key it makes beside it, in state.key. See Across restarts. In the same folder, it keeps people’s AI rules, what admins set, and, without LUMOVI_AUDIT_DIR, the audit history, in audit. The image’s /var/lib/lumovi belongs to Lumovi’s user, so a new named volume there does too. Whoever can write the folder can make Lumovi act as anyone: keep it Lumovi’s alone.
Configuration
Every environment variable.
Security
What Lumovi is trusted with, and how to contain it.