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
Recommended rollout approach
- validate the container locally
- review the Helm chart defaults
- create an environment-specific values file
- deploy to a non-production namespace first
- verify UI and evaluation paths
- 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.