Short answer: your Helm chart version and the Prefect application version are independent. The chart you installed (2025.3.13181020) defines Kubernetes resources, but the actual Prefect version that runs comes from the Docker image tag you set. Because your

values.yaml

uses a floating tag

prefectTag: 3-latest

, your deployment pulled the newest 3.x image (currently 3.4.7). The UI shows the server/API version from that image, not the Helm chart version. What’s happening - Helm chart version: version of the chart templates (how resources are rendered), not the Prefect app itself. The chart may list an

appVersion

(e.g., 3.2.12) but that’s informational unless you pin the image tag. - Image tag (

prefectTag

): controls which Prefect server/API+UI image is deployed.

3-latest

is a moving tag and will resolve to whatever the latest 3.x release is at the time your pods pull the image. - UI version: reads the Prefect server/API version from the running container image, so it reflects 3.4.7 because that’s the image you’re running. Why it changed - Any redeploy/rollout or pod restart with

imagePullPolicy: Always

will re-pull

3-latest

, which now points to 3.4.7. Best practices to avoid surprises - Pin a specific version tag instead of a floating one:

prefectTag: "3.4.7"

Optionally pin by digest for immutability. - Ensure your pull policy won’t re-pull unnecessarily:

imagePullPolicy: IfNotPresent

- If your chart has separate tags for sub-components (e.g., agent/worker), pin those too. If you want to align to a specific Prefect version, update

prefectTag

to that version (e.g.,

3.2.12

) in your

values.yaml

and run a Helm upgrade. If you’d like, share your

values.yaml

(redact any secrets) and I can point out every place you should pin. References - Self-hosted Prefect server - Infrastructure as code guidance