FlowSet — multiple flows in one Pod

A FlowSet serves several flows from a single runner Pod. Each flow gets its own per-flow ConfigMaps under /genkit/flows/<name>/ and its own credentials mount, so two flows can use the same provider with different credentials without collisions.

When to reach for FlowSet

• You have many small flows, and one Pod per flow would be wasteful.
• Several flows share most of their dependencies and you want a single rolling-update unit.
• You want per-flow credentials isolated as files instead of env vars.

If a flow is high-traffic or needs its own scaling profile, keep it as a standalone Flow.

Example

apiVersion: genkit.dev/v1alpha1
kind: FlowSet
metadata:
  name: assistants
spec:
  image: ghcr.io/xavidop/genkit-runner:v0.4.1
  port: 8080
  flows:
    - name: greeter
      modelRef: { name: claude-opus }
      promptRefs:
        - { name: greeting }
    - name: summarizer
      modelRef: { name: claude-opus }
      promptRefs:
        - { name: summarize }
    - name: translator
      modelRef: { name: claude-haiku }
      promptRefs:
        - { name: translate }

Routes exposed by the single Pod:

• POST /greeter
• POST /summarizer
• POST /translator

How credentials are mounted

Each flow’s PluginConfig.credentialsRef Secret is mounted at:

/genkit/flows/<flow-name>/credentials/

One file per key. So with the manifest above, the Pod will have:

/genkit/
├── manifest.json
└── flows/
    ├── greeter/
    │   ├── prompts/greeting.prompt
    │   ├── config.json
    │   └── credentials/
    │       └── ANTHROPIC_API_KEY
    ├── summarizer/
    │   └── ...
    └── translator/
        └── ...

A custom runner only needs to read the right credentials per request (by looking at the route). The reference runner does this automatically.

Content-hash rollout

Just like Flow, the operator computes a SHA-256 over all per-flow ConfigMaps plus the manifest and writes it to the Pod template annotation genkit.dev/content-hash. Editing any referenced Prompt, Tool, Model, or PluginConfig triggers a rolling update of the single shared Deployment.

Inline model spec and prompts (no Model or Prompt CRs)

Each flow entry in a FlowSet supports modelSpec and prompts[].prompt for the same inline approach available in a standalone Flow. Only the PluginConfig CR is still required.

The example below sets up two flows without any Model or Prompt CRs:

apiVersion: genkit.dev/v1alpha1
kind: FlowSet
metadata:
  name: assistants
spec:
  image: ghcr.io/xavidop/genkit-runner:v0.4.1
  port: 8080
  flows:
    - name: greeter
      modelSpec:
        provider: anthropic
        model: claude-opus-4-7
        pluginConfigRef:
          name: anthropic-config
        defaultConfig:
          temperature: 0.3
      prompts:
        - prompt:
            name: greeting
            content: |
              ---
              model: anthropic/claude-opus-4-7
              ---
              Greet the user named {{name}} in a single sentence.
    - name: summarizer
      modelSpec:
        provider: anthropic
        model: claude-opus-4-7
        pluginConfigRef:
          name: anthropic-config
        defaultConfig:
          temperature: 0.5
      prompts:
        - prompt:
            name: summarize
            content: |
              ---
              model: anthropic/claude-opus-4-7
              ---
              Summarize the following text in one paragraph: {{text}}

You can also mix approaches within the same FlowSet — one flow may use modelRef / promptRefs while another uses modelSpec / prompts[].prompt inline.

The same constraints apply as for standalone flows:

• modelRef and modelSpec are mutually exclusive per flow entry.
• Within each prompts list item, promptRef and prompt are mutually exclusive.