AxAgentOptimizeOptions

TypeScript

type AxAgentOptimizeOptions<_IN, _OUT> = object & Pick<AxOptimizerArgs, 
  | "numTrials"
  | "minibatch"
  | "minibatchSize"
  | "earlyStoppingTrials"
  | "minImprovementThreshold"
  | "sampleCount"
| "seed">;

Defined in: https://github.com/ax-llm/ax/blob/e4219a44b7581a55a40d9dd62d2c281062ff078d/src/ax/agent/agentInternal/agentOptimizeTypes.ts#L126

Type declaration

Name	Type	Description
`apply?`	`boolean`	-
`bootstrap?`	\| `boolean` \| `AxGEPABootstrapOptions`	-
`debugOptimizer?`	`boolean`	-
`judgeAI?`	`Readonly`<`AxAIService`>	Optional separate judge model. Defaults to the agent’s `judgeAI`, then `teacherAI`, then the student model.
`judgeOptions?`	`AxAgentJudgeOptions`	-
`maxMetricCalls?`	`number`	-
`mcpEvaluation?`	\| `Readonly`<{ `mcp?`: \| `AxMCPClient` \| readonly `AxMCPClient`[]; `mode`: `"replay"`; `ucp?`: \| `AxUCPClient` \| readonly `AxUCPClient`[]; }> \| `Readonly`<{ `mcp?`: \| `AxMCPClient` \| readonly `AxMCPClient`[]; `mode`: `"live"`; `ucp?`: \| `AxUCPClient` \| readonly `AxUCPClient`[]; }>	Evaluation defaults to replay/sandbox clients. Live mode must be explicit because optimizers evaluate the same task repeatedly.
`metric?`	`AxMetricFn`	Optional deterministic scorer. If omitted, optimize() uses the built-in LLM judge.
`onEarlyStop()?`	(`reason`: `string`, `stats`: `Readonly`<`AxOptimizationStats`>) => `void`	-
`onProgress()?`	(`progress`: `Readonly`<`AxOptimizationProgress`>) => `void`	-
`optimizerLogger?`	`AxOptimizerLoggerFunction`	-
`studentAI?`	`Readonly`<`AxAIService`>	-
`target?`	`AxAgentOptimizeTarget`	Optional optimization scope. Defaults to `'actor'`.
`teacherAI?`	`Readonly`<`AxAIService`>	-
`verbose?`	`boolean`	-

Type Parameters

Type Parameter	Default type
`_IN` extends `AxGenIn`	`AxGenIn`
`_OUT` extends `AxGenOut`	`AxGenOut`