ArkSettings

ArkSettings is analogous to a Kubernetes ConfigMap. It stores reusable prompt fragments and model settings that can be referenced from multiple ArkAgent resources. Define a persona or output format once and share it across many agents.

Example 

apiVersion: arkonis.dev/v1alpha1
kind: ArkSettings
metadata:
  name: analyst-base
  namespace: default
spec:
  temperature: "0.3"
  outputFormat: structured-json
  promptFragments:
    persona: "You are an expert analyst."
    outputRules: "Always cite sources. Flag uncertainty explicitly."

Spec fields

Field Type Required Description
temperature string no Sampling temperature passed to the LLM provider. String-encoded float (e.g. "0.7").
outputFormat string no structured-json instructs the agent to return JSON. markdown and plain are also accepted.
memoryBackend string no in-context (default), redis, or vector-store. For durable backends, use ArkMemory instead.
promptFragments map[string]string no Named prompt fragments merged into the effective system prompt.

promptFragments merge order

Reserved keys have defined merge positions:

Key Merge position
persona Prepended to the system prompt
outputRules Appended after the system prompt

Other keys are appended in alphabetical order.

Referencing from ArkAgent 

apiVersion: arkonis.dev/v1alpha1
kind: ArkAgent
metadata:
  name: analyst-agent
spec:
  model: llama3.2
  systemPrompt: |
    Analyze the provided dataset and identify key trends.
  configRef: analyst-base

The operator resolves configRef during reconciliation, merges promptFragments into the effective system prompt, and injects temperature and outputFormat as environment variables into agent pods.

ArkSettings and the referencing ArkAgent must be in the same namespace.

See also

Apache 2.0 · ARKONIS