Chatoyant - v0.2.1
    Preparing search index...

    Interface GenerateOptions

    Generation options for text/structured output.

    interface GenerateOptions {
        provider?: ProviderId;
        timeout?: number;
        retries?: number;
        temperature?: number;
        creativity?: CreativityLevel;
        maxTokens?: number;
        topP?: number;
        stop?: string | string[];
        frequencyPenalty?: number;
        presencePenalty?: number;
        reasoning?: ReasoningLevel;
        webSearch?: boolean;
        cache?: boolean;
        extra?: Record<string, unknown>;
    }

    Hierarchy (View Summary)

    Index

    Properties

    provider? timeout? retries? temperature? creativity? maxTokens? topP? stop? frequencyPenalty? presencePenalty? reasoning? webSearch? cache? extra?

    Properties

    provider?: ProviderId

    Override the provider (auto-detected from model by default).

    timeout?: number

    Request timeout in milliseconds.

    120000
    retries?: number

    Number of retries on failure.

    3
    temperature?: number

    Sampling temperature (0-2). Lower = more deterministic, higher = more creative. undefined = provider default.

    Can also use semantic presets via creativity option.

    creativity?: CreativityLevel

    Semantic creativity level. Alternative to raw temperature values.

    • 'precise': Temperature 0 (deterministic)
    • 'balanced': Temperature 0.7 (default)
    • 'creative': Temperature 1.0
    • 'wild': Temperature 1.5

    If both temperature and creativity are set, temperature takes precedence.

    maxTokens?: number

    Maximum tokens to generate. undefined = provider default.

    topP?: number

    Top-p (nucleus) sampling. undefined = provider default.

    stop?: string | string[]

    Stop sequences.

    frequencyPenalty?: number

    Frequency penalty (-2 to 2).

    presencePenalty?: number

    Presence penalty (-2 to 2).

    reasoning?: ReasoningLevel

    Unified reasoning level across providers. Maps automatically to provider-specific implementations:

    • 'off': No reasoning (OpenAI: none, Anthropic: no thinking, xAI: *-non-reasoning)
    • 'low': Light reasoning (OpenAI: low, Anthropic: 2048 tokens)
    • 'medium': Moderate reasoning (OpenAI: medium, Anthropic: 8192 tokens)
    • 'high': Deep reasoning (OpenAI: high, Anthropic: 32768 tokens, xAI: *-reasoning)

    Note: Not all models support reasoning. For unsupported models, this is ignored.

    webSearch?: boolean

    Enable web search (xAI only). Ignored for other providers.

    cache?: boolean

    Enable prompt caching (Anthropic only). Marks the system prompt for caching.

    extra?: Record<string, unknown>

    Arbitrary additional options passed to the provider. Use for bleeding-edge features not yet in the typed interface.

    Settings

    Member Visibility

    On This Page

    Properties