Kubernetes & Helm

agentevals now supports containerized deployment and Kubernetes installation through the project Helm chart.

What this unlocks

Running agentevals in containers or Kubernetes makes it easier to:

  • standardize environments across teams
  • run eval workloads in CI or shared infrastructure
  • expose a shared UI or evaluation service
  • manage configuration in a repeatable way

Docker image support

The project includes a Dockerfile, which means you can build and run agentevals in a containerized environment.

Typical uses include:

  • local reproducible testing
  • CI jobs
  • scheduled evaluation workloads
  • base images for Kubernetes deployment

Helm chart support

The repository includes a Helm chart at charts/agentevals with chart metadata, templates, and a values.yaml file for configuration.

Typical Helm workflow:

helm install agentevals ./charts/agentevals

In practice, most teams will customize values before installing.

What to configure

The exact settings depend on your environment, but common areas include:

  • image repository and tag
  • service configuration
  • ingress settings
  • environment variables
  • secrets for model or backend providers
  • persistence and resource sizing
  1. validate the container locally
  2. review the Helm chart defaults
  3. create an environment-specific values file
  4. deploy to a non-production namespace first
  5. verify UI and evaluation paths
  6. promote after validation

Rollback strategy

For Kubernetes deployments, prefer staged rollouts and keep a known-good values file so you can quickly roll back to the prior release.