Create the SeekrFlow values file

Every setting for your deployment lives in a single file, my-values.yaml. Environment differences are settings within that file rather than separate procedures. An Amazon Web Services (AWS) deployment and an on-premises deployment run the same install with different values. Either path uses the values from Collect credentials and values.

Generate the starter file

With the sfe CLI
Without the sfe CLI

Before you start, complete Set up the sfe CLI. Go to the folder where you want to keep the values file, such as your home folder, then run:

sfe bootstrap -r <version> -o my-values.yaml --provider aws

Replace <version> with the SeekrFlow release you’re installing. For an on-premises deployment, replace aws with on-prem.sfe asks a series of questions and writes the answers to my-values.yaml. Most answers come from Collect credentials and values. Type them rather than pasting from a formatted page. A pasted value can carry an invisible character that breaks image pulls without any error.These questions need more than a value from the checklist, or have a default that’s wrong for most deployments:

Question	What to enter
Container image registry	Seekr’s registry, `seekr.repo.sonatype.app/seekr-docker-release/seekr`, or your own registry if you mirrored the images.
File store path	Your bucket name, without `s3://`.
IRSA role ARN	Leave it blank if you use Amazon Elastic Kubernetes Service (Amazon EKS) Pod Identity.
OpenSearch port and scheme	Accept the defaults, `443` and `https://`.
Enable the Valkey cache	`y` for any deployment that untrusted users can reach. See Size your deployment.
Configure ingress	`y` if SeekrFlow creates its own ingress. Enter your hostname and `alb`.
Accelerators fine-tunes may use	The GPU types in your GPU nodes: `A10`, `A100`, `H100`, or `MI300X`. For any other NVIDIA GPU, such as the L4 or L40S, enter `A10`.
Create storage class	`y`. Enter your Amazon Elastic File System (Amazon EFS) file system ID when asked.
Add a Context Attribution model	`y`. Choose a model by number.
Enable shared tools	`y`. Leave `pythonexecutor` off unless you trust all your users. It runs every team’s Python code in one shared pod, outside the sandbox.
Deploy the bundled Keycloak, and Enable Keycloak SSO	`n` to both for local accounts. `y` to both for the bundled Keycloak, or `n` then `y` for your own Keycloak.
Organization name	Your organization’s name. Users see it in the UI and in invitation emails.
Let SeekrFlow create the secret store resource	`y`. The default is `n`.
Secret store provider	`aws-sm` for AWS Secrets Manager. The default is `aws-ssm`, Parameter Store.

After it saves the file, sfe prints a command to label each GPU node with its GPU type, such as:

kubectl label node <node-name> seekr.io/gpu_type=A10

Run it for each GPU node before your first fine-tuning job. Fine-tuning jobs run only on nodes with this label. To check the labels, run kubectl get nodes -L seekr.io/gpu_type.

Start from the default values file that ships with the chart. Sign in to Seekr’s registry with the credentials from your Seekr account team:

echo "$REGISTRY_PASSWORD" | helm registry login seekr.repo.sonatype.app \
  --username "$REGISTRY_USERNAME" \
  --password-stdin

Pull the chart, unpack it, and copy its default values file:

helm pull oci://seekr.repo.sonatype.app/seekr-docker-release/seekr/flow-helm-chart --version <version>
tar -xzf flow-helm-chart-<version>.tgz
cp flow-helm-chart/values.yaml my-values.yaml

Fill in my-values.yaml with your provisioning values. Keep the unpacked chart. You install from it.

Finish the values file

sfe doesn’t write these settings. Add them before you install.

Image pull secret. Under global, add:

imagePullSecrets:
  - name: seekrflow-artifacts-pull-access

Load balancer annotations. If SeekrFlow creates its own ingress on AWS, add the annotations from Load balancer annotations under global.ingress.annotations.
Kubernetes secrets only. Set global.Secrets.enabled to false.

YAML indentation is significant. Match the surrounding indentation exactly and never use tabs.

The file doesn’t define the models SeekrFlow uses for ingestion, embeddings, and agent planning yet. You add them in Configure models.

Validate

Check that the file matches the chart schema.

With the sfe CLI
Without the sfe CLI

sfe values lint -r <version> -f my-values.yaml

From the folder that contains the unpacked chart, run:

helm lint flow-helm-chart -f my-values.yaml

Validation confirms that each setting in the file is valid. It doesn’t check whether any are missing, such as the ones in Finish the values file.

SeekrFlow Self-Managed

Plan a SeekrFlow deployment

Provision SeekrFlow dependencies

Prepare to install SeekrFlow

Deploy SeekrFlow

Operate SeekrFlow

Reference

Generate the starter file

Finish the values file

Validate

​Generate the starter file

​Finish the values file

​Validate

Generate the starter file

Finish the values file

Validate