PalenaModel

A PalenaModel publishes one logical model to one or more PalenaGateways. It translates into one LiteLLMModel CR per referenced gateway.

Short name: pmAPI group: operator.palena.ai/v1alpha1Scope: Namespaced

spec

Field	Type	Required	Description
gatewayRefs	[]LocalObjectReference	Yes (≥1)	PalenaGateway names (same namespace).
modelName	string	Yes	Public model name exposed by the gateway.
litellmParams	LiteLLMModelParams	Yes	Upstream LiteLLM model parameters.
modelInfo	ModelInfo	No	Optional token limits and cost metadata.
dataClassificationMaxLevel	int (0–4)	No	Upper bound for routing policies.
allowedTeams	[]string	No	Restrict to named LiteLLM teams.

LiteLLMModelParams

Field	Type	Required	Description
model	string	Yes	Upstream model identifier (e.g. openai/gpt-4o-mini).
apiBase	string	No	Custom API endpoint for self-hosted models.
apiKeySecretRef	SecretKeyRef	No	Provider API key secret.
rpm	int	No	Requests/minute rate limit.
tpm	int	No	Tokens/minute rate limit.
timeout	int	No	Per-request timeout (seconds).
streamTimeout	int	No	Streaming timeout (seconds).
maxRetries	int	No	Upstream retry count.

ModelInfo

Field	Type	Description
maxTokens	int	Total context window.
inputCostPerToken	float64	Input token cost (dollars per token).
outputCostPerToken	float64	Output token cost (dollars per token).

status

Field	Type	Description
syncedGateways[].gatewayName	string	Referenced gateway.
syncedGateways[].synced	bool	Successful sync to LiteLLM.
syncedGateways[].litellmModelRef	string	Name of the generated LiteLLMModel.
syncedGateways[].lastSyncTime	string	RFC3339 timestamp.
syncedGateways[].error	string	Most recent sync error, if any.
conditions	[]metav1.Condition	Standard conditions (Synced, Ready).

Print columns

Model, Age.

Example

apiVersion: operator.palena.ai/v1alpha1
kind: PalenaModel
metadata:
  name: gpt-4o-mini
  namespace: palena
spec:
  gatewayRefs:
    - name: production
    - name: staging
  modelName: gpt-4o-mini
  litellmParams:
    model: openai/gpt-4o-mini
    apiKeySecretRef:
      name: openai-keys
      key: apiKey
    rpm: 10000
    tpm: 2000000
  modelInfo:
    maxTokens: 128000
    inputCostPerToken: 0.00000015
    outputCostPerToken: 0.0000006
  dataClassificationMaxLevel: 2

Multi-gateway fan-out

When gatewayRefs lists more than one gateway, the controller creates one LiteLLMModel CR per gateway, each named <modelName>-<gatewayName>. Removing a gateway from the list cascades into deletion of the orphaned LiteLLMModel.