Skip to main content

​
Prerequisites — Self-Managed on Your Kubernetes

Complete the following before installing the Serval Helm chart under the Self-Managed on Your Kubernetes delivery model. Your team operates the cluster and the platform; Serval provides the chart, release artifacts, and product support.
Self-Managed on Your Kubernetes is the highest-effort delivery model and is intended for strictly air-gapped, non-AWS, on-prem, or platform-team-mandated environments. It is more work than Serval-Managed in Your AWS, not less — in addition to running Serval, your team is responsible for:
  • Provisioning and maintaining the Kubernetes cluster itself
  • Ingress controllers and TLS
  • PostgreSQL, Redis, and Temporal (provisioned, backed up, upgraded by you)
  • Monitoring, alerting, on-call, and incident response
  • Backups and restore testing for stateful components
  • Helm chart installs, upgrades, and rollbacks
If you do not have a hard constraint that rules out Serval-Managed in Your AWS, choose that model instead.

​
Kubernetes platform

  • A running Kubernetes cluster (v1.27+) — any cloud provider or on-prem
  • kubectl configured to access the cluster
  • Helm 3.12+
  • An ingress controller deployed (for example, NGINX Ingress, Traefik, or a cloud-native controller)
  • Sufficient node capacity for the chart’s default resource requests

​
Domain and DNS

  • A domain you own with the ability to create DNS records
  • DNS records (configured before or after install) for:
    • app.<your-domain> — the Serval web application
    • *.api.<your-domain> — backend API services (subdomain-based ingress)

​
Datastores

The Serval chart can bundle PostgreSQL, Redis, and Temporal for evaluation, but production deployments typically use managed or externally operated datastores. Plan for:
  • PostgreSQL — primary application database
  • Redis — caching and ephemeral state
  • Temporal — workflow orchestration

​
Private network access (optional)

If Serval needs to reach resources on your internal network (on-prem AD, internal GitLab, etc.), most of it is free: Serval runs on your cluster and uses whatever networking that cluster already has. See Private Network Access for the full picture.

​
LLM / agent API keys

  • Anthropic API key — primary provider for AI agents and help desk functionality
  • OpenAI API key — used for embeddings, image description, and specific model tasks

​
Image pull credentials

Serval provides a Docker Hub pull token during onboarding. You supply this token to the chart via global.imagePullToken.

​
Operational readiness

Because your team owns day-2 operations, confirm ownership for:
  • Monitoring, alerting, and on-call
  • Backups and restore testing for stateful components
  • Upgrade cadence and rollback procedures
See the Production Checklist before going live.

​
Next steps

Deployment instructions

Get access to install steps, configuration references, and operational runbooks.

Deployment

Install the Serval Helm chart from the public OCI registry.