Interface ProcessingOptions

Defines fine-grained control options for how AgentOS processes an individual turn. These options can override system defaults or persona-specific settings for the duration of the current request, allowing for dynamic adjustments to GMI behavior and output.

Interface

ProcessingOptions

interface ProcessingOptions {
    streamUICommands?: boolean;
    maxToolCallIterations?: number;
    preferredModelId?: string;
    preferredProviderId?: string;
    temperature?: number;
    topP?: number;
    maxTokens?: number;
    responseFormat?: {
        type: string;
    };
    disableAdaptation?: boolean;
    debugMode?: boolean;
    forceNewConversation?: boolean;
    customFlags?: Record<string, any>;
    structuredReply?: StructuredReplySpec;
}

Properties

streamUICommands? maxToolCallIterations? preferredModelId? preferredProviderId? temperature? topP? maxTokens? responseFormat? disableAdaptation? debugMode? forceNewConversation? customFlags? structuredReply?

Properties

Optional streamUICommands

streamUICommands?: boolean

Optional maxToolCallIterations

maxToolCallIterations?: number

Optional preferredModelId

preferredModelId?: string

The model for this turn's model calls, over the persona's defaultModelId and the config's default model.

Optional preferredProviderId

preferredProviderId?: string

The provider for this turn's model calls, over the persona's defaultProviderId and the config's default provider. Name it with a preferredModelId of another provider than the persona's. The tool results that continue the turn (handleToolResult(), handleToolResults(), resumeExternalToolRequest()) keep both.

Optional temperature

temperature?: number

Optional topP

topP?: number

Optional maxTokens

maxTokens?: number

Optional responseFormat

responseFormat?: {
    type: string;
}

Type declaration

  • type: string

Optional disableAdaptation

disableAdaptation?: boolean

Optional debugMode

debugMode?: boolean

Optional forceNewConversation

forceNewConversation?: boolean

Optional customFlags

customFlags?: Record<string, any>

Optional structuredReply

structuredReply?: StructuredReplySpec

A reply that must match a schema: the model is instructed, its reply parsed and checked, a reply that does not match asked for again inside the turn, and the final chunk carries the parsed value (structured). See src/api/runtime/structuredReply.ts.

Settings

Member Visibility

Theme

On This Page

streamUICommandsmaxToolCallIterationspreferredModelIdpreferredProviderIdtemperaturetopPmaxTokensresponseFormatdisableAdaptationdebugModeforceNewConversationcustomFlagsstructuredReply