Class GeminiAdapter

constructor

• new GeminiAdapter(options: GeminiAdapterOptions): GeminiAdapter

  Creates an instance of GeminiAdapter.

  Parameters

  • options: GeminiAdapterOptions

    Configuration options for the adapter.

  Returns GeminiAdapter

  Throws

  If apiKey is missing in the options.

  See

  https://ai.google.dev/api/rest

ReadonlyproviderName

call

• call(
      prompt: ArtStandardPrompt,
      options: CallOptions,
  ): Promise<AsyncIterable<StreamEvent, any, any>>

  Makes a call to the configured Gemini model. Translates the ArtStandardPrompt into the Gemini API format, sends the request using the @google/genai SDK, and yields StreamEvent objects representing the response (tokens, metadata, errors, end signal).

  Handles both streaming and non-streaming requests based on options.stream.

  Thinking tokens (Gemini):

  • On supported Gemini models (e.g., gemini-2.5-*), you can enable thought output via config.thinkingConfig.
  • This adapter reads provider-specific flags from the call options:
    • options.gemini.thinking.includeThoughts: boolean — when true, requests thought (reasoning) output.
    • options.gemini.thinking.thinkingBudget?: number — optional token budget for thinking.
  • When enabled and supported, the adapter will attempt to differentiate thought vs response parts and set StreamEvent.tokenType accordingly:
    • For planning calls (callContext === 'AGENT_THOUGHT'): AGENT_THOUGHT_LLM_THINKING or AGENT_THOUGHT_LLM_RESPONSE.
    • For synthesis calls (callContext === 'FINAL_SYNTHESIS'): FINAL_SYNTHESIS_LLM_THINKING or FINAL_SYNTHESIS_LLM_RESPONSE.
  • LLMMetadata.thinkingTokens will be populated if the provider reports separate thinking token usage.
  • If the SDK/model does not expose thought parts, the adapter falls back to labeling tokens as ...LLM_RESPONSE.

  Parameters

  • prompt: ArtStandardPrompt

    The standardized prompt messages.
  • options: CallOptions

    Options for the LLM call, including streaming preference, model override, and execution context.

  Returns Promise<AsyncIterable<StreamEvent, any, any>>

  An async iterable that yields StreamEvent objects.

  • TOKEN: Contains a chunk of the response text. tokenType indicates if it's part of agent thought or final synthesis. When Gemini thinking is enabled and available, tokenType may be one of the ...LLM_THINKING or ...LLM_RESPONSE variants to separate thought vs response tokens.
  • METADATA: Contains information like stop reason, token counts, and timing, yielded once at the end.
  • ERROR: Contains any error encountered during translation, SDK call, or response processing.
  • END: Signals the completion of the stream.

  See

  • https://ai.google.dev/api/rest/v1beta/models/generateContent

  Example

  // Enable Gemini thinking (if supported by the selected model)
  const stream = await geminiAdapter.call(prompt, {
    threadId,
    stream: true,
    callContext: 'FINAL_SYNTHESIS',
    providerConfig, // your RuntimeProviderConfig
    gemini: {
      thinking: { includeThoughts: true, thinkingBudget: 8096 }
    }
  });
  for await (const evt of stream) {
    if (evt.type === 'TOKEN') {
      // evt.tokenType may be FINAL_SYNTHESIS_LLM_THINKING or FINAL_SYNTHESIS_LLM_RESPONSE
    }
  }
  Copy