@jambonz/sdk
    Preparing search index...

    Interface AgentLlmOptions

    Per-call LLM configuration for the agent verb.

    interface AgentLlmOptions {
        systemPrompt?: string;
        messages?: LlmMessage[];
        initialMessages?: LlmMessage[];
        maxTokens?: number;
        temperature?: number;
        reasoningEffort?: "minimal" | "low" | "medium" | "high";
        vendorParams?: Record<string, unknown>;
        tools?: (LlmTool | LlmToolOpenAIWrapped)[];
    }
    Index

    Properties

    systemPrompt? messages? initialMessages? maxTokens? temperature? reasoningEffort? vendorParams? tools?

    Properties

    systemPrompt?: string

    System prompt. Placed vendor-appropriately by the feature-server wrapper.

    messages?: LlmMessage[]

    Seed conversation history. A role:'system' entry is extracted into systemPrompt.

    initialMessages?: LlmMessage[]

    Alias of messages (historical).

    maxTokens?: number

    Maximum tokens the model may generate per turn.

    temperature?: number

    Sampling temperature.

    reasoningEffort?: "minimal" | "low" | "medium" | "high"

    Has no effect. Set the vendor's own thinking parameter in vendorParams instead, e.g. { reasoning_effort: 'none' } for openai.

    vendorParams?: Record<string, unknown>

    Vendor-specific request parameters, passed to the vendor as-is, e.g. { provider: { sort: 'latency' } } for openrouter. Added to the request body (OpenAI-compatible vendors) or generation config (google, vertex-gemini); never overrides a field jambonz sets. Ignored by anthropic and bedrock.

    tools?: (LlmTool | LlmToolOpenAIWrapped)[]

    Tools / functions the model may call.

    Settings

    Member Visibility

    On This Page

    Properties