Kubernetes & OpenShift

Prefer a cluster over a single VM? Kodus ships Helm charts for Kubernetes and OpenShift in the kodus-installer repository. This page is the overview — the charts’ README is the complete reference (air-gapped installs, ExternalSecrets, operators, observability, and SOC 2-oriented hardening).

Requirements

Helm 3.8+
Kubernetes 1.28+ or OpenShift 4.14+
For mode: operator data stores, the matching operator installed on the cluster

Quick start (bundled data stores)

The default bundled mode runs PostgreSQL (pgvector), MongoDB, and RabbitMQ as StatefulSets the chart manages — the equivalent of Docker Compose USE_LOCAL_DB=true. One command, no operators:

git clone https://github.com/kodustech/kodus-installer
cd kodus-installer/charts/kodus
helm dependency build
helm install kodus . -n kodus --create-namespace \
  --set imageTag=<version> \
  --set global.config.WEB_HOSTNAME_API=api.kodus.example.com \
  --set global.config.NEXTAUTH_URL=https://kodus.example.com \
  --set ingress.hosts.web.host=kodus.example.com \
  --set ingress.hosts.api.host=api.kodus.example.com

imageTag pins the Kodus release for the whole stack (like Docker Compose IMAGE_TAG); use a real tag, never latest, in production. Auth/crypto secrets are generated automatically with the correct format and stay stable across upgrades.

The webhooks service needs its own public host. Point ingress.hosts.webhooks.host (or route.hosts.webhooks.host on OpenShift) at a dedicated hostname and the chart auto-derives API_<provider>_CODE_MANAGEMENT_WEBHOOK for you.

Data store modes

Each store (postgres, mongodb, rabbitmq) has an independent mode:

mode	what it does	for
`bundled` (default)	a StatefulSet managed by the chart	dev / trials
`external`	your managed/self-managed service via an `existingSecret`	production
`operator`	a CR reconciled by a cluster operator (HA) — CloudNativePG, RabbitMQ Cluster Operator, MongoDB Community Operator	production

For production, prefer external or operator — they hand database operation to something purpose-built.

OpenShift

Add the OpenShift overlay — Routes replace Ingress, and pod UIDs are left to the namespace SCC (restricted-v2), so no UIDs are hardcoded:

helm install kodus . -f values.yaml -f values-openshift.yaml \
  -n kodus --create-namespace --set imageTag=<version>

Leave route.hosts.*.host empty to let OpenShift auto-assign the <name>-<namespace>.apps… hostname. Note that some clusters mirror or block Docker Hub library/* images, so the bundled mongo image may need an override (--set mongodb.bundled.image=mirror.gcr.io/library/mongo:8) or an external / operator data store.

Optional services & verification

MCP manager — --set services.mcp-manager.enabled=true
Analytics worker — --set services.worker-analytics.enabled=true
Health check — ../../scripts/doctor-k8s.sh -n kodus inspects workloads, config, and the real health endpoints.

Full Helm reference

Data-store operator setup, reusing existing databases, air-gapped installs, secrets via ExternalSecrets, observability, and security defaults.

Concepts

Local Quickstart

Deploy with Docker

Deploy with Kubernetes

Git providers configuration

Kubernetes & OpenShift

Requirements

Quick start (bundled data stores)

Data store modes

OpenShift

Optional services & verification

Full Helm reference

​Requirements

​Quick start (bundled data stores)

​Data store modes

​OpenShift

​Optional services & verification

Full Helm reference

Requirements

Quick start (bundled data stores)

Data store modes

OpenShift

Optional services & verification