Hosting guide
This handbook is for the people who run Labtagon AI OS: they install it on a Kubernetes cluster with
its Helm chart (deploy/helm/ai-os), keep it running, back it up and upgrade it. What
administrators set inside the console is in the Administration handbook.
What you need
Section titled “What you need”| Kubernetes | 1.30 or later. The chart says so in kubeVersion. |
| Sign-in | An OpenID Connect provider. The chart is set up for Keycloak; any OIDC provider works. |
| A route in | A Gateway API implementation (HTTPRoutes) or an Ingress controller, with DNS and TLS for your host. |
| Postgres | The CloudNativePG operator, or a Postgres of your own with the vector extension (pgvector 0.8 or later), created by a superuser. |
| Keys | One Secret with the platform’s keys, created once, for example from OpenBao through External Secrets. |
| Object storage | Nothing extra by default: the chart runs Garage. Any S3 works instead. |
AI OS runs no image registry and no secrets store of its own. Its component images come from the
registry you name in global.imageRegistry; third-party images are pulled where they are published,
or through a mirror (global.mirror).
The chapters
Section titled “The chapters”Architecture and componentsWhat runs where, who talks to whom, and what each component does.
SizingThe chart's requests, limits, replicas and quotas, and what grows with what.
Installing with HelmThe values you must set, the keys, Keycloak, and the first start.
App hostThe console on a host of its own (ADR 0097).
Network and egressDefault deny, the paths the chart opens, and the one way out.
Backups and restorePostgres, the files, and the keys you must never lose.
Upgrades and migrationsHow a new release rolls out and how the database follows.