Skip to main content

1. Choose the chart source

The Helm chart is published as a public OCI artifact. If you are installing from a local checkout of the chart instead, skip these variables and use the local path in the helm install commands below.

2. Create the namespace

3. Create required Secrets

Docker Hub pull secret

Required if global.imagePullSecrets is set (needed for hellocobi/* private images).

Backend environment secret

The backend reads its configuration from a Kubernetes Secret. Create it with your environment-specific values:
See Backend Service for the full variable reference.

Frontend environment secret

VITE_API_BASE_URL must match the backend Gateway route, Ingress, or Route hostname.

Connect service secret

Use the same CONNECT_SERVICE_TOKEN value in both the backend and connect secrets.

4. Build chart dependencies

If installing from a local chart directory:
This downloads the dependency charts (PostgreSQL, Qdrant, vllm-stack, MinIO) into charts/.

5. Install

Choose a values file that matches your environment, then install:
Or from a local chart:

6. Verify the rollout

All pods should reach Running / 1/1 Ready. The vLLM pod takes longer (~5 min) while it downloads model weights.

Upgrade

Uninstall

PersistentVolumeClaims are not deleted automatically. Remove them manually if you want to purge all data:

Minimal values file

The following enables all application components with sensible defaults. Substitute your image tags and domain names:

OpenShift values override

For OpenShift, disable Kubernetes Ingress and Gateway, then enable the chart’s frontend and backend API Routes:
OpenShift Routes use edge TLS termination by default — TLS is terminated at the router, plain HTTP to the pods. Update the frontend and backend secrets to use the frontend Route and API Route:
See Ingress & Routes for the full platform-specific configuration.