@tanstack/ai-gemini 0.19.1 → 0.20.0

package/dist/esm/model-meta.js.map

@@ -1 +1 @@
-{"version":3,"file":"model-meta.js","sources":["../../src/model-meta.ts"],"sourcesContent":["import type {\n  GeminiCachedContentOptions,\n  GeminiCommonConfigOptions,\n  GeminiSafetyOptions,\n  GeminiStructuredOutputOptions,\n  GeminiThinkingOptions,\n  GeminiToolConfigOptions,\n} from './text/text-provider-options'\n\ninterface ModelMeta<TProviderOptions = unknown> {\n  name: string\n  supports: {\n    input: Array<'text' | 'image' | 'audio' | 'video' | 'document'>\n    output: Array<'text' | 'image' | 'audio' | 'video'>\n    capabilities?: Array<\n      | 'audio_generation'\n      | 'batch_api'\n      | 'caching'\n      | 'function_calling'\n      | 'live_api'\n      | 'structured_output'\n      | 'thinking'\n    >\n    tools?: Array<\n      | 'code_execution'\n      | 'file_search'\n      | 'google_search'\n      | 'google_search_retrieval'\n      | 'google_maps'\n      | 'url_context'\n      | 'computer_use'\n    >\n  }\n  max_input_tokens?: number\n  max_output_tokens?: number\n  knowledge_cutoff?: string\n  pricing?: {\n    input: {\n      normal: number\n      cached?: number\n    }\n    output: {\n      normal: number\n    }\n  }\n  /**\n   * Type-level description of which provider options this model supports.\n   */\n  providerOptions?: TProviderOptions\n}\n\nconst GEMINI_3_1_PRO = {\n  name: 'gemini-3.1-pro-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_FLASH = {\n  name: 'gemini-3-flash-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.5,\n    },\n    output: {\n      normal: 3,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_PRO_IMAGE = {\n  name: 'gemini-3-pro-image-preview',\n  max_input_tokens: 65_536,\n  max_output_tokens: 32_768,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 2,\n    },\n    output: {\n      normal: 0.134,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_IMAGE = {\n  name: 'gemini-3.1-flash-image-preview',\n  max_input_tokens: 65_536,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_LITE_IMAGE = {\n  name: 'gemini-3.1-flash-lite-image',\n  max_input_tokens: 65_536,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_LITE = {\n  name: 'gemini-3.1-flash-lite',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_LITE_PREVIEW = {\n  name: 'gemini-3.1-flash-lite-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_PRO = {\n  name: 'gemini-2.5-pro',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: [\n      'code_execution',\n      'file_search',\n      'google_maps',\n      'google_search',\n      'url_context',\n    ],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_PRO_TTS = {\n  name: 'gemini-2.5-pro-preview-tts',\n  max_input_tokens: 8_192,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst GEMINI_2_5_FLASH = {\n  name: 'gemini-2.5-flash',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: [\n      'code_execution',\n      'file_search',\n      'google_maps',\n      'google_search',\n      'url_context',\n    ],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_FLASH_IMAGE = {\n  name: 'gemini-2.5-flash-image',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-06-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'caching', 'structured_output'],\n    tools: ['file_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.3,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n/**\nconst GEMINI_2_5_FLASH_LIVE = {\n  name: 'gemini-2.5-flash-native-audio-preview-09-2025',\n  max_input_tokens: 141_072,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'audio', 'video'],\n    output: ['text', 'audio'],\n    capabilities: [\n      'audio_generation',\n      'file_search',\n      'function_calling',\n      'live_api',\n      'search_grounding',\n      'thinking',\n    ],\n  },\n  pricing: {\n    // todo find this info\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions &\n  GeminiThinkingOptions\n>\n*/\nconst GEMINI_2_5_FLASH_TTS = {\n  name: 'gemini-2.5-flash-preview-tts',\n  max_input_tokens: 8_192,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation', 'batch_api'],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Gemini 3.1 Flash TTS Preview - latest expressive TTS model with\n * 200+ audio tags, 70+ languages, and multi-speaker dialogue support.\n * @see https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-tts-preview\n */\nconst GEMINI_3_1_FLASH_TTS = {\n  name: 'gemini-3.1-flash-tts-preview',\n  max_input_tokens: 32_768,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation', 'batch_api'],\n  },\n  pricing: {\n    input: {\n      normal: 0.5,\n    },\n    output: {\n      normal: 10,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Lyria 3 Pro Preview — Google's flagship music generation model.\n * Generates full-length songs with multiple verses, choruses, and bridges.\n * Outputs MP3 or WAV at 48 kHz stereo.\n * @see https://ai.google.dev/gemini-api/docs/models/lyria-3-pro-preview\n */\nconst LYRIA_3_PRO = {\n  name: 'lyria-3-pro-preview',\n  max_input_tokens: 131_072,\n  supports: {\n    input: ['text', 'image'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta\n\n/**\n * Lyria 3 Clip Preview — 30-second music clips in MP3 format.\n * @see https://ai.google.dev/gemini-api/docs/music-generation\n */\nconst LYRIA_3_CLIP = {\n  name: 'lyria-3-clip-preview',\n  max_input_tokens: 131_072,\n  supports: {\n    input: ['text', 'image'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GEMINI_2_5_FLASH_LITE = {\n  name: 'gemini-2.5-flash-lite',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'google_maps', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst IMAGEN_4_GENERATE = {\n  name: 'imagen-4.0-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_4_GENERATE_ULTRA = {\n  name: 'imagen-4.0-ultra-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.6,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_4_GENERATE_FAST = {\n  name: 'imagen-4.0-fast-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.2,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Veo video generation models. Pricing is per second of generated video\n * (audio+video rate where the model supports audio).\n * @experimental Veo video generation is an experimental feature and may change.\n */\nconst VEO_3_1_PREVIEW = {\n  name: 'veo-3.1-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst VEO_3_1_FAST_PREVIEW = {\n  name: 'veo-3.1-fast-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst VEO_3_1_LITE_PREVIEW = {\n  name: 'veo-3.1-lite-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.05,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst GEMINI_3_5_FLASH = {\n  name: 'gemini-3.5-flash',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  supports: {\n    input: ['text', 'image', 'video', 'document', 'audio'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 1.5,\n      cached: 0.15,\n    },\n    output: {\n      normal: 9,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nexport const GEMINI_MODELS = [\n  GEMINI_3_5_FLASH.name,\n  GEMINI_3_1_PRO.name,\n  GEMINI_3_FLASH.name,\n  GEMINI_3_1_FLASH_LITE.name,\n  GEMINI_3_1_FLASH_LITE_PREVIEW.name,\n  GEMINI_2_5_PRO.name,\n  GEMINI_2_5_FLASH.name,\n  GEMINI_2_5_FLASH_LITE.name,\n] as const\n\n/**\n * Gemini models that support combining `tools` + `responseSchema` in a\n * single streaming `generateContent` call (per issue #605). Per the\n * provider matrix, Gemini 3.x natively interleaves the schema-constrained\n * answer with function-calling on one pass; Gemini 2.x is unsupported /\n * brittle and keeps the engine's legacy finalization fallback.\n */\nexport const GEMINI_COMBINED_TOOLS_AND_SCHEMA_MODELS = new Set<string>([\n  GEMINI_3_1_PRO.name,\n  GEMINI_3_FLASH.name,\n  GEMINI_3_1_FLASH_LITE.name,\n  GEMINI_3_1_FLASH_LITE_PREVIEW.name,\n  GEMINI_3_5_FLASH.name,\n])\n\nexport type GeminiModels = (typeof GEMINI_MODELS)[number]\n\nexport type GeminiImageModels = (typeof GEMINI_IMAGE_MODELS)[number]\n\nexport const GEMINI_IMAGE_MODELS = [\n  GEMINI_3_1_FLASH_IMAGE.name,\n  GEMINI_3_1_FLASH_LITE_IMAGE.name,\n  GEMINI_3_PRO_IMAGE.name,\n  GEMINI_2_5_FLASH_IMAGE.name,\n  IMAGEN_4_GENERATE.name,\n  IMAGEN_4_GENERATE_FAST.name,\n  IMAGEN_4_GENERATE_ULTRA.name,\n] as const\n\n/**\n * Text-to-speech models\n * @experimental Gemini TTS is an experimental feature and may change.\n */\nexport const GEMINI_TTS_MODELS = [\n  GEMINI_3_1_FLASH_TTS.name,\n  GEMINI_2_5_FLASH_TTS.name,\n  GEMINI_2_5_PRO_TTS.name,\n] as const\n\n/**\n * Audio generation models (Lyria music generation).\n * @experimental Lyria music generation is an experimental feature and may change.\n */\nexport const GEMINI_AUDIO_MODELS = [\n  LYRIA_3_PRO.name,\n  LYRIA_3_CLIP.name,\n] as const\n\n/**\n * Available voice names for Gemini TTS\n * @see https://ai.google.dev/gemini-api/docs/speech-generation\n */\nexport const GEMINI_TTS_VOICES = [\n  'Zephyr',\n  'Puck',\n  'Charon',\n  'Kore',\n  'Fenrir',\n  'Leda',\n  'Orus',\n  'Aoede',\n  'Callirrhoe',\n  'Autonoe',\n  'Enceladus',\n  'Iapetus',\n  'Umbriel',\n  'Algieba',\n  'Despina',\n  'Erinome',\n  'Algenib',\n  'Rasalgethi',\n  'Laomedeia',\n  'Achernar',\n  'Alnilam',\n  'Schedar',\n  'Gacrux',\n  'Pulcherrima',\n  'Achird',\n  'Zubenelgenubi',\n  'Vindemiatrix',\n  'Sadachbia',\n  'Sadaltager',\n  'Sulafat',\n] as const\n\nexport type GeminiTTSVoice = (typeof GEMINI_TTS_VOICES)[number]\n\n/**\n * Veo video generation models.\n * @experimental Veo video generation is an experimental feature and may change.\n */\nexport const GEMINI_VIDEO_MODELS = [\n  VEO_3_1_PREVIEW.name,\n  VEO_3_1_FAST_PREVIEW.name,\n  VEO_3_1_LITE_PREVIEW.name,\n] as const\n\n// Manual type map for per-model provider options\nexport type GeminiChatModelProviderOptionsByName = {\n  // Models with thinking and structured output support\n  [GEMINI_3_1_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_3_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_3_1_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_3_1_FLASH_LITE_PREVIEW.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_3_5_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n}\n\n/**\n * Type-only map from chat model name to its supported tool capabilities.\n * Based on the 'supports.tools' arrays defined for each model.\n */\nexport type GeminiChatModelToolCapabilitiesByName = {\n  [GEMINI_3_1_PRO.name]: typeof GEMINI_3_1_PRO.supports.tools\n  [GEMINI_3_FLASH.name]: typeof GEMINI_3_FLASH.supports.tools\n  [GEMINI_3_1_FLASH_LITE.name]: typeof GEMINI_3_1_FLASH_LITE.supports.tools\n  [GEMINI_3_1_FLASH_LITE_PREVIEW.name]: typeof GEMINI_3_1_FLASH_LITE_PREVIEW.supports.tools\n  [GEMINI_2_5_PRO.name]: typeof GEMINI_2_5_PRO.supports.tools\n  [GEMINI_2_5_FLASH.name]: typeof GEMINI_2_5_FLASH.supports.tools\n  [GEMINI_2_5_FLASH_LITE.name]: typeof GEMINI_2_5_FLASH_LITE.supports.tools\n  [GEMINI_3_5_FLASH.name]: typeof GEMINI_3_5_FLASH.supports.tools\n}\n\n/**\n * Type-only map from chat model name to its supported input modalities.\n * Based on the 'supports.input' arrays defined for each model.\n * Note: 'document' in the model meta is mapped to 'document' modality.\n * Used by the core AI types to constrain ContentPart types based on the selected model.\n * Note: These must be inlined as readonly arrays (not typeof) because the model\n * constants are not exported and typeof references don't work in .d.ts files\n * when consumed by external packages.\n *\n * @see https://ai.google.dev/gemini-api/docs/vision\n * @see https://ai.google.dev/gemini-api/docs/audio\n * @see https://ai.google.dev/gemini-api/docs/document-processing\n */\nexport type GeminiModelInputModalitiesByName = {\n  // Models with full multimodal support (text, image, audio, video, document)\n  [GEMINI_3_1_PRO.name]: typeof GEMINI_3_1_PRO.supports.input\n  [GEMINI_3_FLASH.name]: typeof GEMINI_3_FLASH.supports.input\n  [GEMINI_3_1_FLASH_LITE.name]: typeof GEMINI_3_1_FLASH_LITE.supports.input\n  [GEMINI_3_1_FLASH_LITE_PREVIEW.name]: typeof GEMINI_3_1_FLASH_LITE_PREVIEW.supports.input\n  [GEMINI_2_5_PRO.name]: typeof GEMINI_2_5_PRO.supports.input\n  [GEMINI_2_5_FLASH_LITE.name]: typeof GEMINI_2_5_FLASH_LITE.supports.input\n  [GEMINI_3_5_FLASH.name]: typeof GEMINI_3_5_FLASH.supports.input\n\n  // Models with text, image, audio, video (no document)\n  [GEMINI_2_5_FLASH.name]: typeof GEMINI_2_5_FLASH.supports.input\n}\n"],"names":[],"mappings":"AAmDA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AASA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AASA,MAAM,qBAAqB;AAAA,EACzB,MAAM;AAkBR;AASA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAkBR;AASA,MAAM,8BAA8B;AAAA,EAClC,MAAM;AAkBR;AASA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAwBR;AASA,MAAM,gCAAgC;AAAA,EACpC,MAAM;AAwBR;AASA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AA8BR;AASA,MAAM,qBAAqB;AAAA,EACzB,MAAM;AAiBR;AAOA,MAAM,mBAAmB;AAAA,EACvB,MAAM;AA8BR;AASA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAkBR;AAyCA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAiBR;AAYA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAiBR;AAaA,MAAM,cAAc;AAAA,EAClB,MAAM;AAeR;AAMA,MAAM,eAAe;AAAA,EACnB,MAAM;AAeR;AAEA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAwBR;AASA,MAAM,oBAAoB;AAAA,EACxB,MAAM;AAeR;AAOA,MAAM,0BAA0B;AAAA,EAC9B,MAAM;AAeR;AAOA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAeR;AAYA,MAAM,kBAAkB;AAAA,EACtB,MAAM;AAeR;AAOA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAeR;AAOA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAeR;AAOA,MAAM,mBAAmB;AAAA,EACvB,MAAM;AAwBR;AASO,MAAM,gBAAgB;AAAA,EAC3B,iBAAiB;AAAA,EACjB,eAAe;AAAA,EACf,eAAe;AAAA,EACf,sBAAsB;AAAA,EACtB,8BAA8B;AAAA,EAC9B,eAAe;AAAA,EACf,iBAAiB;AAAA,EACjB,sBAAsB;AACxB;AASO,MAAM,8DAA8C,IAAY;AAAA,EACrE,eAAe;AAAA,EACf,eAAe;AAAA,EACf,sBAAsB;AAAA,EACtB,8BAA8B;AAAA,EAC9B,iBAAiB;AACnB,CAAC;AAMM,MAAM,sBAAsB;AAAA,EACjC,uBAAuB;AAAA,EACvB,4BAA4B;AAAA,EAC5B,mBAAmB;AAAA,EACnB,uBAAuB;AAAA,EACvB,kBAAkB;AAAA,EAClB,uBAAuB;AAAA,EACvB,wBAAwB;AAC1B;AAMO,MAAM,oBAAoB;AAAA,EAC/B,qBAAqB;AAAA,EACrB,qBAAqB;AAAA,EACrB,mBAAmB;AACrB;AAMO,MAAM,sBAAsB;AAAA,EACjC,YAAY;AAAA,EACZ,aAAa;AACf;AAMO,MAAM,oBAAoB;AAAA,EAC/B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAQO,MAAM,sBAAsB;AAAA,EACjC,gBAAgB;AAAA,EAChB,qBAAqB;AAAA,EACrB,qBAAqB;AACvB;"}
+{"version":3,"file":"model-meta.js","sources":["../../src/model-meta.ts"],"sourcesContent":["import type {\n  GeminiCachedContentOptions,\n  GeminiCommonConfigOptions,\n  GeminiSafetyOptions,\n  GeminiStructuredOutputOptions,\n  GeminiThinkingOptions,\n  GeminiToolConfigOptions,\n} from './text/text-provider-options'\n\ninterface ModelMeta<TProviderOptions = unknown> {\n  name: string\n  supports: {\n    input: Array<'text' | 'image' | 'audio' | 'video' | 'document'>\n    output: Array<'text' | 'image' | 'audio' | 'video'>\n    capabilities?: Array<\n      | 'audio_generation'\n      | 'batch_api'\n      | 'caching'\n      | 'function_calling'\n      | 'live_api'\n      | 'structured_output'\n      | 'thinking'\n    >\n    tools?: Array<\n      | 'code_execution'\n      | 'file_search'\n      | 'google_search'\n      | 'google_search_retrieval'\n      | 'google_maps'\n      | 'url_context'\n      | 'computer_use'\n    >\n  }\n  max_input_tokens?: number\n  max_output_tokens?: number\n  knowledge_cutoff?: string\n  pricing?: {\n    input: {\n      normal: number\n      cached?: number\n    }\n    output: {\n      normal: number\n    }\n  }\n  /**\n   * Type-level description of which provider options this model supports.\n   */\n  providerOptions?: TProviderOptions\n}\n\nconst GEMINI_3_1_PRO = {\n  name: 'gemini-3.1-pro-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_FLASH = {\n  name: 'gemini-3-flash-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.5,\n    },\n    output: {\n      normal: 3,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_PRO_IMAGE = {\n  name: 'gemini-3-pro-image-preview',\n  max_input_tokens: 65_536,\n  max_output_tokens: 32_768,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 2,\n    },\n    output: {\n      normal: 0.134,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_IMAGE = {\n  name: 'gemini-3.1-flash-image-preview',\n  max_input_tokens: 65_536,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_LITE_IMAGE = {\n  name: 'gemini-3.1-flash-lite-image',\n  max_input_tokens: 65_536,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_LITE = {\n  name: 'gemini-3.1-flash-lite',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_LITE_PREVIEW = {\n  name: 'gemini-3.1-flash-lite-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_PRO = {\n  name: 'gemini-2.5-pro',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: [\n      'code_execution',\n      'file_search',\n      'google_maps',\n      'google_search',\n      'url_context',\n    ],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_PRO_TTS = {\n  name: 'gemini-2.5-pro-preview-tts',\n  max_input_tokens: 8_192,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst GEMINI_2_5_FLASH = {\n  name: 'gemini-2.5-flash',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: [\n      'code_execution',\n      'file_search',\n      'google_maps',\n      'google_search',\n      'url_context',\n    ],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_FLASH_IMAGE = {\n  name: 'gemini-2.5-flash-image',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-06-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'caching', 'structured_output'],\n    tools: ['file_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.3,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n/**\nconst GEMINI_2_5_FLASH_LIVE = {\n  name: 'gemini-2.5-flash-native-audio-preview-09-2025',\n  max_input_tokens: 141_072,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'audio', 'video'],\n    output: ['text', 'audio'],\n    capabilities: [\n      'audio_generation',\n      'file_search',\n      'function_calling',\n      'live_api',\n      'search_grounding',\n      'thinking',\n    ],\n  },\n  pricing: {\n    // todo find this info\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions &\n  GeminiThinkingOptions\n>\n*/\nconst GEMINI_2_5_FLASH_TTS = {\n  name: 'gemini-2.5-flash-preview-tts',\n  max_input_tokens: 8_192,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation', 'batch_api'],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Gemini 3.1 Flash TTS Preview - latest expressive TTS model with\n * 200+ audio tags, 70+ languages, and multi-speaker dialogue support.\n * @see https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-tts-preview\n */\nconst GEMINI_3_1_FLASH_TTS = {\n  name: 'gemini-3.1-flash-tts-preview',\n  max_input_tokens: 32_768,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation', 'batch_api'],\n  },\n  pricing: {\n    input: {\n      normal: 0.5,\n    },\n    output: {\n      normal: 10,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Lyria 3 Pro Preview — Google's flagship music generation model.\n * Generates full-length songs with multiple verses, choruses, and bridges.\n * Outputs MP3 or WAV at 48 kHz stereo.\n * @see https://ai.google.dev/gemini-api/docs/models/lyria-3-pro-preview\n */\nconst LYRIA_3_PRO = {\n  name: 'lyria-3-pro-preview',\n  max_input_tokens: 131_072,\n  supports: {\n    input: ['text', 'image'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta\n\n/**\n * Lyria 3 Clip Preview — 30-second music clips in MP3 format.\n * @see https://ai.google.dev/gemini-api/docs/music-generation\n */\nconst LYRIA_3_CLIP = {\n  name: 'lyria-3-clip-preview',\n  max_input_tokens: 131_072,\n  supports: {\n    input: ['text', 'image'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GEMINI_2_5_FLASH_LITE = {\n  name: 'gemini-2.5-flash-lite',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'google_maps', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst IMAGEN_4_GENERATE = {\n  name: 'imagen-4.0-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_4_GENERATE_ULTRA = {\n  name: 'imagen-4.0-ultra-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.6,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_4_GENERATE_FAST = {\n  name: 'imagen-4.0-fast-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.2,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Veo video generation models. Pricing is per second of generated video\n * (audio+video rate where the model supports audio).\n * @experimental Veo video generation is an experimental feature and may change.\n */\nconst VEO_3_1_PREVIEW = {\n  name: 'veo-3.1-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst VEO_3_1_FAST_PREVIEW = {\n  name: 'veo-3.1-fast-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst VEO_3_1_LITE_PREVIEW = {\n  name: 'veo-3.1-lite-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.05,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Gemini Omni Flash — multimodal video generation with conversational\n * editing. Serves only the Interactions API (`generateContent` rejects it),\n * so it routes through the interactions-based path of the video adapter,\n * not Veo's `:predictLongRunning` flow. Pricing is per second of generated\n * video ($0.10/sec). 720p / 24 FPS, 3–10 second clips (default 10s).\n * @experimental Omni video generation is an experimental feature and may change.\n */\nconst GEMINI_OMNI_FLASH_PREVIEW = {\n  name: 'gemini-omni-flash-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image', 'video'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.1,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst GEMINI_3_5_FLASH = {\n  name: 'gemini-3.5-flash',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  supports: {\n    input: ['text', 'image', 'video', 'document', 'audio'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 1.5,\n      cached: 0.15,\n    },\n    output: {\n      normal: 9,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nexport const GEMINI_MODELS = [\n  GEMINI_3_5_FLASH.name,\n  GEMINI_3_1_PRO.name,\n  GEMINI_3_FLASH.name,\n  GEMINI_3_1_FLASH_LITE.name,\n  GEMINI_3_1_FLASH_LITE_PREVIEW.name,\n  GEMINI_2_5_PRO.name,\n  GEMINI_2_5_FLASH.name,\n  GEMINI_2_5_FLASH_LITE.name,\n] as const\n\n/**\n * Gemini models that support combining `tools` + `responseSchema` in a\n * single streaming `generateContent` call (per issue #605). Per the\n * provider matrix, Gemini 3.x natively interleaves the schema-constrained\n * answer with function-calling on one pass; Gemini 2.x is unsupported /\n * brittle and keeps the engine's legacy finalization fallback.\n */\nexport const GEMINI_COMBINED_TOOLS_AND_SCHEMA_MODELS = new Set<string>([\n  GEMINI_3_1_PRO.name,\n  GEMINI_3_FLASH.name,\n  GEMINI_3_1_FLASH_LITE.name,\n  GEMINI_3_1_FLASH_LITE_PREVIEW.name,\n  GEMINI_3_5_FLASH.name,\n])\n\nexport type GeminiModels = (typeof GEMINI_MODELS)[number]\n\nexport type GeminiImageModels = (typeof GEMINI_IMAGE_MODELS)[number]\n\nexport const GEMINI_IMAGE_MODELS = [\n  GEMINI_3_1_FLASH_IMAGE.name,\n  GEMINI_3_1_FLASH_LITE_IMAGE.name,\n  GEMINI_3_PRO_IMAGE.name,\n  GEMINI_2_5_FLASH_IMAGE.name,\n  IMAGEN_4_GENERATE.name,\n  IMAGEN_4_GENERATE_FAST.name,\n  IMAGEN_4_GENERATE_ULTRA.name,\n] as const\n\n/**\n * Text-to-speech models\n * @experimental Gemini TTS is an experimental feature and may change.\n */\nexport const GEMINI_TTS_MODELS = [\n  GEMINI_3_1_FLASH_TTS.name,\n  GEMINI_2_5_FLASH_TTS.name,\n  GEMINI_2_5_PRO_TTS.name,\n] as const\n\n/**\n * Audio generation models (Lyria music generation).\n * @experimental Lyria music generation is an experimental feature and may change.\n */\nexport const GEMINI_AUDIO_MODELS = [\n  LYRIA_3_PRO.name,\n  LYRIA_3_CLIP.name,\n] as const\n\n/**\n * Available voice names for Gemini TTS\n * @see https://ai.google.dev/gemini-api/docs/speech-generation\n */\nexport const GEMINI_TTS_VOICES = [\n  'Zephyr',\n  'Puck',\n  'Charon',\n  'Kore',\n  'Fenrir',\n  'Leda',\n  'Orus',\n  'Aoede',\n  'Callirrhoe',\n  'Autonoe',\n  'Enceladus',\n  'Iapetus',\n  'Umbriel',\n  'Algieba',\n  'Despina',\n  'Erinome',\n  'Algenib',\n  'Rasalgethi',\n  'Laomedeia',\n  'Achernar',\n  'Alnilam',\n  'Schedar',\n  'Gacrux',\n  'Pulcherrima',\n  'Achird',\n  'Zubenelgenubi',\n  'Vindemiatrix',\n  'Sadachbia',\n  'Sadaltager',\n  'Sulafat',\n] as const\n\nexport type GeminiTTSVoice = (typeof GEMINI_TTS_VOICES)[number]\n\n/**\n * Video generation models. Veo models run on the long-running\n * `:predictLongRunning` flow; Gemini Omni Flash runs on the Interactions\n * API — the video adapter routes by model.\n * @experimental Video generation is an experimental feature and may change.\n */\nexport const GEMINI_VIDEO_MODELS = [\n  VEO_3_1_PREVIEW.name,\n  VEO_3_1_FAST_PREVIEW.name,\n  VEO_3_1_LITE_PREVIEW.name,\n  GEMINI_OMNI_FLASH_PREVIEW.name,\n] as const\n\n/**\n * Video models served by the Interactions API rather than Veo's\n * `:predictLongRunning` operations flow.\n * @experimental Omni video generation is an experimental feature and may change.\n */\nexport const GEMINI_INTERACTIONS_VIDEO_MODELS = [\n  GEMINI_OMNI_FLASH_PREVIEW.name,\n] as const\n\n// Manual type map for per-model provider options\nexport type GeminiChatModelProviderOptionsByName = {\n  // Models with thinking and structured output support\n  [GEMINI_3_1_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_3_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_3_1_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_3_1_FLASH_LITE_PREVIEW.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_3_5_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n}\n\n/**\n * Type-only map from chat model name to its supported tool capabilities.\n * Based on the 'supports.tools' arrays defined for each model.\n */\nexport type GeminiChatModelToolCapabilitiesByName = {\n  [GEMINI_3_1_PRO.name]: typeof GEMINI_3_1_PRO.supports.tools\n  [GEMINI_3_FLASH.name]: typeof GEMINI_3_FLASH.supports.tools\n  [GEMINI_3_1_FLASH_LITE.name]: typeof GEMINI_3_1_FLASH_LITE.supports.tools\n  [GEMINI_3_1_FLASH_LITE_PREVIEW.name]: typeof GEMINI_3_1_FLASH_LITE_PREVIEW.supports.tools\n  [GEMINI_2_5_PRO.name]: typeof GEMINI_2_5_PRO.supports.tools\n  [GEMINI_2_5_FLASH.name]: typeof GEMINI_2_5_FLASH.supports.tools\n  [GEMINI_2_5_FLASH_LITE.name]: typeof GEMINI_2_5_FLASH_LITE.supports.tools\n  [GEMINI_3_5_FLASH.name]: typeof GEMINI_3_5_FLASH.supports.tools\n}\n\n/**\n * Type-only map from chat model name to its supported input modalities.\n * Based on the 'supports.input' arrays defined for each model.\n * Note: 'document' in the model meta is mapped to 'document' modality.\n * Used by the core AI types to constrain ContentPart types based on the selected model.\n * Note: These must be inlined as readonly arrays (not typeof) because the model\n * constants are not exported and typeof references don't work in .d.ts files\n * when consumed by external packages.\n *\n * @see https://ai.google.dev/gemini-api/docs/vision\n * @see https://ai.google.dev/gemini-api/docs/audio\n * @see https://ai.google.dev/gemini-api/docs/document-processing\n */\nexport type GeminiModelInputModalitiesByName = {\n  // Models with full multimodal support (text, image, audio, video, document)\n  [GEMINI_3_1_PRO.name]: typeof GEMINI_3_1_PRO.supports.input\n  [GEMINI_3_FLASH.name]: typeof GEMINI_3_FLASH.supports.input\n  [GEMINI_3_1_FLASH_LITE.name]: typeof GEMINI_3_1_FLASH_LITE.supports.input\n  [GEMINI_3_1_FLASH_LITE_PREVIEW.name]: typeof GEMINI_3_1_FLASH_LITE_PREVIEW.supports.input\n  [GEMINI_2_5_PRO.name]: typeof GEMINI_2_5_PRO.supports.input\n  [GEMINI_2_5_FLASH_LITE.name]: typeof GEMINI_2_5_FLASH_LITE.supports.input\n  [GEMINI_3_5_FLASH.name]: typeof GEMINI_3_5_FLASH.supports.input\n\n  // Models with text, image, audio, video (no document)\n  [GEMINI_2_5_FLASH.name]: typeof GEMINI_2_5_FLASH.supports.input\n}\n"],"names":[],"mappings":"AAmDA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AASA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AASA,MAAM,qBAAqB;AAAA,EACzB,MAAM;AAkBR;AASA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAkBR;AASA,MAAM,8BAA8B;AAAA,EAClC,MAAM;AAkBR;AASA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAwBR;AASA,MAAM,gCAAgC;AAAA,EACpC,MAAM;AAwBR;AASA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AA8BR;AASA,MAAM,qBAAqB;AAAA,EACzB,MAAM;AAiBR;AAOA,MAAM,mBAAmB;AAAA,EACvB,MAAM;AA8BR;AASA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAkBR;AAyCA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAiBR;AAYA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAiBR;AAaA,MAAM,cAAc;AAAA,EAClB,MAAM;AAeR;AAMA,MAAM,eAAe;AAAA,EACnB,MAAM;AAeR;AAEA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAwBR;AASA,MAAM,oBAAoB;AAAA,EACxB,MAAM;AAeR;AAOA,MAAM,0BAA0B;AAAA,EAC9B,MAAM;AAeR;AAOA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAeR;AAYA,MAAM,kBAAkB;AAAA,EACtB,MAAM;AAeR;AAOA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAeR;AAOA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAeR;AAeA,MAAM,4BAA4B;AAAA,EAChC,MAAM;AAeR;AAOA,MAAM,mBAAmB;AAAA,EACvB,MAAM;AAwBR;AASO,MAAM,gBAAgB;AAAA,EAC3B,iBAAiB;AAAA,EACjB,eAAe;AAAA,EACf,eAAe;AAAA,EACf,sBAAsB;AAAA,EACtB,8BAA8B;AAAA,EAC9B,eAAe;AAAA,EACf,iBAAiB;AAAA,EACjB,sBAAsB;AACxB;AASO,MAAM,8DAA8C,IAAY;AAAA,EACrE,eAAe;AAAA,EACf,eAAe;AAAA,EACf,sBAAsB;AAAA,EACtB,8BAA8B;AAAA,EAC9B,iBAAiB;AACnB,CAAC;AAMM,MAAM,sBAAsB;AAAA,EACjC,uBAAuB;AAAA,EACvB,4BAA4B;AAAA,EAC5B,mBAAmB;AAAA,EACnB,uBAAuB;AAAA,EACvB,kBAAkB;AAAA,EAClB,uBAAuB;AAAA,EACvB,wBAAwB;AAC1B;AAMO,MAAM,oBAAoB;AAAA,EAC/B,qBAAqB;AAAA,EACrB,qBAAqB;AAAA,EACrB,mBAAmB;AACrB;AAMO,MAAM,sBAAsB;AAAA,EACjC,YAAY;AAAA,EACZ,aAAa;AACf;AAMO,MAAM,oBAAoB;AAAA,EAC/B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAUO,MAAM,sBAAsB;AAAA,EACjC,gBAAgB;AAAA,EAChB,qBAAqB;AAAA,EACrB,qBAAqB;AAAA,EACrB,0BAA0B;AAC5B;AAOO,MAAM,mCAAmC;AAAA,EAC9C,0BAA0B;AAC5B;"}

package/dist/esm/realtime/adapter.d.ts

@@ -0,0 +1,22 @@
+import { RealtimeAdapter } from '@tanstack/ai';
+import { GeminiRealtimeOptions } from './types.js';
+/**
+ * Creates a Gemini realtime adapter for client-side use.
+ *
+ * @param options - Optional configuration
+ * @returns A RealtimeAdapter for use with RealtimeClient
+ *
+ * @example
+ * ```typescript
+ * import { RealtimeClient } from '@tanstack/ai-client'
+ * import { geminiRealtime } from '@tanstack/ai-gemini'
+ *
+ * const client = new RealtimeClient({
+ *   getToken: () => fetch('/api/realtime-token').then(r => r.json()),
+ *   adapter: geminiRealtime(),
+ *   onGoAway: () => client.updateSession({ ... }) // Resume session with new config (available only for Gemini Live adapter)
+ * })
+ *
+ * ```
+ */
+export declare function geminiRealtime(options?: GeminiRealtimeOptions): RealtimeAdapter;

package/dist/esm/realtime/adapter.js

@@ -0,0 +1,233 @@
+import { createRealtimeEventEmitter } from "@tanstack/ai";
+import { AudioPlayer, base64ToArrayBuffer, AudioStreamer } from "./utils.js";
+import { GeminiLiveClient } from "./client.js";
+function geminiRealtime(options = {}) {
+  return {
+    provider: "gemini",
+    connect(token, clientTools) {
+      return createWebSocketConnection(token, options.model, clientTools);
+    }
+  };
+}
+async function createWebSocketConnection(token, model = "gemini-3.1-flash-live-preview", tools) {
+  const { emit, on: realtimeEventEmitterOn } = createRealtimeEventEmitter();
+  let currentMode = "idle";
+  let currentMessageId = null;
+  let messageIdCounter = 0;
+  function generateMessageId() {
+    return `gemini-msg-${Date.now()}-${++messageIdCounter}`;
+  }
+  const client = new GeminiLiveClient(token.token, model, tools);
+  let message = {
+    id: "",
+    role: "assistant",
+    timestamp: 0,
+    parts: []
+  };
+  let pendingAssistantResponse = "";
+  client.onClose = () => {
+    emit("status_change", { status: "idle" });
+    emit("mode_change", { mode: "idle" });
+  };
+  client.onError = (error) => {
+    emit("error", { error });
+    emit("status_change", { status: "error" });
+    emit("mode_change", { mode: "idle" });
+  };
+  client.onReceiveResponse = (response) => {
+    switch (response.type) {
+      case "text":
+        message.parts.push({
+          type: "text",
+          content: response.data
+        });
+        break;
+      case "audio": {
+        const pcm = base64ToArrayBuffer(response.data.audioData);
+        message.parts.push({
+          type: "audio",
+          transcript: response.data.transcript,
+          audioData: pcm
+        });
+        if (currentMode !== "speaking") {
+          currentMode = "speaking";
+          emit("mode_change", { mode: "speaking" });
+        }
+        audioPlayer.play(pcm).catch((error) => emit("error", { error }));
+        break;
+      }
+      case "go_away":
+        emit("go_away", { timeLeft: response.data.timeLeft });
+        break;
+      case "usage_metadata":
+        emit("usage", {
+          completionTokens: response.data.responseTokenCount ?? 0,
+          promptTokens: response.data.promptTokenCount ?? 0,
+          totalTokens: response.data.totalTokenCount ?? 0
+        });
+        break;
+      case "input_transcription":
+        if (response.data.finished && currentMode !== "thinking") {
+          currentMode = "thinking";
+          emit("mode_change", { mode: "thinking" });
+        }
+        emit("transcript", {
+          isFinal: response.data.finished,
+          transcript: response.data.text,
+          role: "user"
+        });
+        break;
+      case "output_transcription":
+        pendingAssistantResponse += response.data.text;
+        emit("transcript", {
+          isFinal: response.data.finished,
+          transcript: pendingAssistantResponse,
+          role: "assistant"
+        });
+        break;
+      case "interrupted":
+        audioPlayer.interrupt();
+        currentMode = "listening";
+        emit("mode_change", { mode: "listening" });
+        emit("interrupted", { messageId: currentMessageId ?? void 0 });
+        break;
+      case "tool_call":
+        for (const tool of response.data.functionCalls || []) {
+          if (tool.id && tool.name) {
+            emit("tool_call", {
+              toolCallId: tool.id,
+              input: tool.args,
+              toolName: tool.name
+            });
+          }
+        }
+        break;
+      case "turn_complete":
+        currentMessageId = generateMessageId();
+        message.id = currentMessageId;
+        message.timestamp = Date.now();
+        message.parts.push({
+          type: "text",
+          content: pendingAssistantResponse
+        });
+        emit("message_complete", { message });
+        pendingAssistantResponse = "";
+        message = {
+          id: "",
+          role: "assistant",
+          timestamp: 0,
+          parts: []
+        };
+        currentMode = "listening";
+        emit("mode_change", { mode: "listening" });
+        break;
+      case "setup_complete":
+        emit("status_change", { status: "connected" });
+        break;
+      case "error":
+        emit("error", {
+          error: new Error(response.data)
+        });
+        break;
+    }
+  };
+  await client.connect();
+  const audioStreamer = new AudioStreamer(client);
+  const audioPlayer = new AudioPlayer();
+  try {
+    await audioPlayer.init();
+    await audioStreamer.start();
+  } catch (error) {
+    audioStreamer.stop();
+    audioPlayer.destroy();
+    client.disconnect();
+    throw error;
+  }
+  const connection = {
+    async disconnect() {
+      audioStreamer.stop();
+      audioPlayer.destroy();
+      client.disconnect();
+      currentMode = "idle";
+      emit("status_change", { status: "idle" });
+    },
+    async startAudioCapture() {
+      audioStreamer.startAudioCapture();
+      currentMode = "listening";
+      emit("mode_change", { mode: "listening" });
+    },
+    stopAudioCapture() {
+      audioStreamer.stopAudioCapture();
+      currentMode = "idle";
+      emit("mode_change", { mode: "idle" });
+    },
+    sendText(text) {
+      client.sendTextMessage(text);
+      currentMode = "thinking";
+      emit("mode_change", { mode: "thinking" });
+    },
+    sendImage(imageData, mimeType) {
+      client.sendImageMessage(imageData, mimeType);
+      currentMode = "thinking";
+      emit("mode_change", { mode: "thinking" });
+    },
+    sendToolResult(callId, result) {
+      client.sendToolResponse([
+        {
+          id: callId,
+          response: {
+            result
+          }
+        }
+      ]);
+    },
+    updateSession(config) {
+      client.updateSession(config).catch((error) => emit("error", { error }));
+      emit("status_change", { status: "reconnecting" });
+    },
+    updateToken(newToken) {
+      client.updateToken(newToken);
+      emit("status_change", { status: "reconnecting" });
+    },
+    interrupt() {
+      audioPlayer.interrupt();
+      currentMode = "listening";
+      emit("mode_change", { mode: "listening" });
+      emit("interrupted", { messageId: currentMessageId ?? void 0 });
+    },
+    on: realtimeEventEmitterOn,
+    getAudioVisualization() {
+      return {
+        get inputLevel() {
+          return audioStreamer.inputLevel;
+        },
+        get outputLevel() {
+          return audioPlayer.outputLevel;
+        },
+        getInputFrequencyData() {
+          return audioStreamer.inputFrequencyData;
+        },
+        getOutputFrequencyData() {
+          return audioPlayer.outputFrequencyData;
+        },
+        getInputTimeDomainData() {
+          return audioStreamer.inputTimeDomainData;
+        },
+        getOutputTimeDomainData() {
+          return audioPlayer.outputTimeDomainData;
+        },
+        get inputSampleRate() {
+          return audioStreamer.inputSampleRate;
+        },
+        get outputSampleRate() {
+          return audioPlayer.outputSampleRate;
+        }
+      };
+    }
+  };
+  return connection;
+}
+export {
+  geminiRealtime
+};
+//# sourceMappingURL=adapter.js.map

package/dist/esm/realtime/adapter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"adapter.js","sources":["../../../src/realtime/adapter.ts"],"sourcesContent":["import { createRealtimeEventEmitter } from '@tanstack/ai'\nimport { AudioPlayer, AudioStreamer, base64ToArrayBuffer } from './utils'\nimport { GeminiLiveClient } from './client'\nimport type { LiveResponse } from './client'\nimport type {\n  AnyClientTool,\n  AudioVisualization,\n  RealtimeAdapter,\n  RealtimeConnection,\n  RealtimeMessage,\n  RealtimeMode,\n  RealtimeToken,\n} from '@tanstack/ai'\nimport type { GeminiRealtimeModel, GeminiRealtimeOptions } from './types'\n\n/**\n * Creates a Gemini realtime adapter for client-side use.\n *\n * @param options - Optional configuration\n * @returns A RealtimeAdapter for use with RealtimeClient\n *\n * @example\n * ```typescript\n * import { RealtimeClient } from '@tanstack/ai-client'\n * import { geminiRealtime } from '@tanstack/ai-gemini'\n *\n * const client = new RealtimeClient({\n *   getToken: () => fetch('/api/realtime-token').then(r => r.json()),\n *   adapter: geminiRealtime(),\n *   onGoAway: () => client.updateSession({ ... }) // Resume session with new config (available only for Gemini Live adapter)\n * })\n *\n * ```\n */\nexport function geminiRealtime(\n  options: GeminiRealtimeOptions = {},\n): RealtimeAdapter {\n  return {\n    provider: 'gemini',\n\n    connect(\n      token: RealtimeToken,\n      clientTools?: ReadonlyArray<AnyClientTool>,\n    ): Promise<RealtimeConnection> {\n      return createWebSocketConnection(token, options.model, clientTools)\n    },\n  }\n}\n\n/**\n * Creates a WebSocket connection to Gemini's realtime API\n */\nasync function createWebSocketConnection(\n  token: RealtimeToken,\n  model: GeminiRealtimeModel = 'gemini-3.1-flash-live-preview',\n  tools?: ReadonlyArray<AnyClientTool>,\n): Promise<RealtimeConnection> {\n  const { emit, on: realtimeEventEmitterOn } = createRealtimeEventEmitter()\n\n  // Current state\n  let currentMode: RealtimeMode = 'idle'\n  let currentMessageId: string | null = null\n  let messageIdCounter = 0\n\n  function generateMessageId(): string {\n    return `gemini-msg-${Date.now()}-${++messageIdCounter}`\n  }\n\n  const client = new GeminiLiveClient(token.token, model, tools)\n\n  let message: RealtimeMessage = {\n    id: '',\n    role: 'assistant',\n    timestamp: 0,\n    parts: [],\n  }\n\n  let pendingAssistantResponse = ''\n\n  client.onClose = () => {\n    emit('status_change', { status: 'idle' })\n    emit('mode_change', { mode: 'idle' })\n  }\n\n  client.onError = (error) => {\n    emit('error', { error })\n    emit('status_change', { status: 'error' })\n    emit('mode_change', { mode: 'idle' })\n  }\n\n  client.onReceiveResponse = (response: LiveResponse) => {\n    switch (response.type) {\n      case 'text':\n        message.parts.push({\n          type: 'text',\n          content: response.data,\n        })\n        break\n      case 'audio': {\n        // Decode once, then reuse for both the stored message part and playback.\n        const pcm = base64ToArrayBuffer(response.data.audioData)\n        message.parts.push({\n          type: 'audio',\n          transcript: response.data.transcript,\n          audioData: pcm,\n        })\n        if (currentMode !== 'speaking') {\n          currentMode = 'speaking'\n          emit('mode_change', { mode: 'speaking' })\n        }\n        audioPlayer.play(pcm).catch((error) => emit('error', { error }))\n        break\n      }\n      case 'go_away':\n        emit('go_away', { timeLeft: response.data.timeLeft })\n        break\n      case 'usage_metadata':\n        emit('usage', {\n          completionTokens: response.data.responseTokenCount ?? 0,\n          promptTokens: response.data.promptTokenCount ?? 0,\n          totalTokens: response.data.totalTokenCount ?? 0,\n        })\n        break\n      case 'input_transcription':\n        if (response.data.finished && currentMode !== 'thinking') {\n          currentMode = 'thinking'\n          emit('mode_change', { mode: 'thinking' })\n        }\n        emit('transcript', {\n          isFinal: response.data.finished,\n          transcript: response.data.text,\n          role: 'user',\n        })\n        break\n      case 'output_transcription':\n        pendingAssistantResponse += response.data.text\n        emit('transcript', {\n          isFinal: response.data.finished,\n          transcript: pendingAssistantResponse,\n          role: 'assistant',\n        })\n        break\n      case 'interrupted':\n        audioPlayer.interrupt()\n        currentMode = 'listening'\n        emit('mode_change', { mode: 'listening' })\n        emit('interrupted', { messageId: currentMessageId ?? undefined })\n        break\n      case 'tool_call':\n        for (const tool of response.data.functionCalls || []) {\n          if (tool.id && tool.name) {\n            emit('tool_call', {\n              toolCallId: tool.id,\n              input: tool.args,\n              toolName: tool.name,\n            })\n          }\n        }\n        break\n      case 'turn_complete':\n        currentMessageId = generateMessageId()\n        message.id = currentMessageId\n        message.timestamp = Date.now()\n        message.parts.push({\n          type: 'text',\n          content: pendingAssistantResponse,\n        })\n        emit('message_complete', { message })\n\n        pendingAssistantResponse = ''\n        message = {\n          id: '',\n          role: 'assistant',\n          timestamp: 0,\n          parts: [],\n        }\n        currentMode = 'listening'\n        emit('mode_change', { mode: 'listening' })\n        break\n      case 'setup_complete':\n        emit('status_change', { status: 'connected' })\n        break\n      case 'error':\n        emit('error', {\n          error: new Error(response.data),\n        })\n        break\n      case 'thought':\n      case 'session_resumption_update':\n        // Handled internally by GeminiLiveClient; not surfaced to the client.\n        break\n    }\n  }\n\n  await client.connect()\n\n  const audioStreamer = new AudioStreamer(client)\n  const audioPlayer = new AudioPlayer()\n  try {\n    await audioPlayer.init()\n    await audioStreamer.start()\n  } catch (error) {\n    // Tear down the socket + audio graph if mic/worklet setup fails (e.g. the\n    // user denied microphone access) rather than leaking an open connection.\n    audioStreamer.stop()\n    audioPlayer.destroy()\n    client.disconnect()\n    throw error\n  }\n\n  const connection: RealtimeConnection = {\n    async disconnect() {\n      audioStreamer.stop()\n      audioPlayer.destroy()\n      client.disconnect()\n      currentMode = 'idle'\n      emit('status_change', { status: 'idle' })\n    },\n\n    async startAudioCapture() {\n      // Audio capture is established during connection setup\n      audioStreamer.startAudioCapture()\n      currentMode = 'listening'\n      emit('mode_change', { mode: 'listening' })\n    },\n\n    stopAudioCapture() {\n      audioStreamer.stopAudioCapture()\n      currentMode = 'idle'\n      emit('mode_change', { mode: 'idle' })\n    },\n\n    sendText(text: string) {\n      client.sendTextMessage(text)\n      currentMode = 'thinking'\n      emit('mode_change', { mode: 'thinking' })\n    },\n\n    sendImage(imageData: string, mimeType: string) {\n      client.sendImageMessage(imageData, mimeType)\n      currentMode = 'thinking'\n      emit('mode_change', { mode: 'thinking' })\n    },\n\n    sendToolResult(callId: string, result: string) {\n      client.sendToolResponse([\n        {\n          id: callId,\n          response: {\n            result,\n          },\n        },\n      ])\n    },\n\n    updateSession(config) {\n      client.updateSession(config).catch((error) => emit('error', { error }))\n      emit('status_change', { status: 'reconnecting' })\n    },\n\n    updateToken(newToken) {\n      client.updateToken(newToken)\n      emit('status_change', { status: 'reconnecting' })\n    },\n\n    interrupt() {\n      audioPlayer.interrupt()\n      currentMode = 'listening'\n      emit('mode_change', { mode: 'listening' })\n      emit('interrupted', { messageId: currentMessageId ?? undefined })\n    },\n    on: realtimeEventEmitterOn,\n    getAudioVisualization(): AudioVisualization {\n      return {\n        get inputLevel() {\n          return audioStreamer.inputLevel\n        },\n\n        get outputLevel() {\n          return audioPlayer.outputLevel\n        },\n\n        getInputFrequencyData() {\n          return audioStreamer.inputFrequencyData\n        },\n\n        getOutputFrequencyData() {\n          return audioPlayer.outputFrequencyData\n        },\n\n        getInputTimeDomainData() {\n          return audioStreamer.inputTimeDomainData\n        },\n\n        getOutputTimeDomainData() {\n          return audioPlayer.outputTimeDomainData\n        },\n\n        get inputSampleRate() {\n          return audioStreamer.inputSampleRate\n        },\n\n        get outputSampleRate() {\n          return audioPlayer.outputSampleRate\n        },\n      }\n    },\n  }\n\n  return connection\n}\n"],"names":[],"mappings":";;;AAkCO,SAAS,eACd,UAAiC,IAChB;AACjB,SAAO;AAAA,IACL,UAAU;AAAA,IAEV,QACE,OACA,aAC6B;AAC7B,aAAO,0BAA0B,OAAO,QAAQ,OAAO,WAAW;AAAA,IACpE;AAAA,EAAA;AAEJ;AAKA,eAAe,0BACb,OACA,QAA6B,iCAC7B,OAC6B;AAC7B,QAAM,EAAE,MAAM,IAAI,uBAAA,IAA2B,2BAAA;AAG7C,MAAI,cAA4B;AAChC,MAAI,mBAAkC;AACtC,MAAI,mBAAmB;AAEvB,WAAS,oBAA4B;AACnC,WAAO,cAAc,KAAK,IAAA,CAAK,IAAI,EAAE,gBAAgB;AAAA,EACvD;AAEA,QAAM,SAAS,IAAI,iBAAiB,MAAM,OAAO,OAAO,KAAK;AAE7D,MAAI,UAA2B;AAAA,IAC7B,IAAI;AAAA,IACJ,MAAM;AAAA,IACN,WAAW;AAAA,IACX,OAAO,CAAA;AAAA,EAAC;AAGV,MAAI,2BAA2B;AAE/B,SAAO,UAAU,MAAM;AACrB,SAAK,iBAAiB,EAAE,QAAQ,OAAA,CAAQ;AACxC,SAAK,eAAe,EAAE,MAAM,OAAA,CAAQ;AAAA,EACtC;AAEA,SAAO,UAAU,CAAC,UAAU;AAC1B,SAAK,SAAS,EAAE,OAAO;AACvB,SAAK,iBAAiB,EAAE,QAAQ,QAAA,CAAS;AACzC,SAAK,eAAe,EAAE,MAAM,OAAA,CAAQ;AAAA,EACtC;AAEA,SAAO,oBAAoB,CAAC,aAA2B;AACrD,YAAQ,SAAS,MAAA;AAAA,MACf,KAAK;AACH,gBAAQ,MAAM,KAAK;AAAA,UACjB,MAAM;AAAA,UACN,SAAS,SAAS;AAAA,QAAA,CACnB;AACD;AAAA,MACF,KAAK,SAAS;AAEZ,cAAM,MAAM,oBAAoB,SAAS,KAAK,SAAS;AACvD,gBAAQ,MAAM,KAAK;AAAA,UACjB,MAAM;AAAA,UACN,YAAY,SAAS,KAAK;AAAA,UAC1B,WAAW;AAAA,QAAA,CACZ;AACD,YAAI,gBAAgB,YAAY;AAC9B,wBAAc;AACd,eAAK,eAAe,EAAE,MAAM,WAAA,CAAY;AAAA,QAC1C;AACA,oBAAY,KAAK,GAAG,EAAE,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,MAAA,CAAO,CAAC;AAC/D;AAAA,MACF;AAAA,MACA,KAAK;AACH,aAAK,WAAW,EAAE,UAAU,SAAS,KAAK,UAAU;AACpD;AAAA,MACF,KAAK;AACH,aAAK,SAAS;AAAA,UACZ,kBAAkB,SAAS,KAAK,sBAAsB;AAAA,UACtD,cAAc,SAAS,KAAK,oBAAoB;AAAA,UAChD,aAAa,SAAS,KAAK,mBAAmB;AAAA,QAAA,CAC/C;AACD;AAAA,MACF,KAAK;AACH,YAAI,SAAS,KAAK,YAAY,gBAAgB,YAAY;AACxD,wBAAc;AACd,eAAK,eAAe,EAAE,MAAM,WAAA,CAAY;AAAA,QAC1C;AACA,aAAK,cAAc;AAAA,UACjB,SAAS,SAAS,KAAK;AAAA,UACvB,YAAY,SAAS,KAAK;AAAA,UAC1B,MAAM;AAAA,QAAA,CACP;AACD;AAAA,MACF,KAAK;AACH,oCAA4B,SAAS,KAAK;AAC1C,aAAK,cAAc;AAAA,UACjB,SAAS,SAAS,KAAK;AAAA,UACvB,YAAY;AAAA,UACZ,MAAM;AAAA,QAAA,CACP;AACD;AAAA,MACF,KAAK;AACH,oBAAY,UAAA;AACZ,sBAAc;AACd,aAAK,eAAe,EAAE,MAAM,YAAA,CAAa;AACzC,aAAK,eAAe,EAAE,WAAW,oBAAoB,QAAW;AAChE;AAAA,MACF,KAAK;AACH,mBAAW,QAAQ,SAAS,KAAK,iBAAiB,CAAA,GAAI;AACpD,cAAI,KAAK,MAAM,KAAK,MAAM;AACxB,iBAAK,aAAa;AAAA,cAChB,YAAY,KAAK;AAAA,cACjB,OAAO,KAAK;AAAA,cACZ,UAAU,KAAK;AAAA,YAAA,CAChB;AAAA,UACH;AAAA,QACF;AACA;AAAA,MACF,KAAK;AACH,2BAAmB,kBAAA;AACnB,gBAAQ,KAAK;AACb,gBAAQ,YAAY,KAAK,IAAA;AACzB,gBAAQ,MAAM,KAAK;AAAA,UACjB,MAAM;AAAA,UACN,SAAS;AAAA,QAAA,CACV;AACD,aAAK,oBAAoB,EAAE,SAAS;AAEpC,mCAA2B;AAC3B,kBAAU;AAAA,UACR,IAAI;AAAA,UACJ,MAAM;AAAA,UACN,WAAW;AAAA,UACX,OAAO,CAAA;AAAA,QAAC;AAEV,sBAAc;AACd,aAAK,eAAe,EAAE,MAAM,YAAA,CAAa;AACzC;AAAA,MACF,KAAK;AACH,aAAK,iBAAiB,EAAE,QAAQ,YAAA,CAAa;AAC7C;AAAA,MACF,KAAK;AACH,aAAK,SAAS;AAAA,UACZ,OAAO,IAAI,MAAM,SAAS,IAAI;AAAA,QAAA,CAC/B;AACD;AAAA,IAIA;AAAA,EAEN;AAEA,QAAM,OAAO,QAAA;AAEb,QAAM,gBAAgB,IAAI,cAAc,MAAM;AAC9C,QAAM,cAAc,IAAI,YAAA;AACxB,MAAI;AACF,UAAM,YAAY,KAAA;AAClB,UAAM,cAAc,MAAA;AAAA,EACtB,SAAS,OAAO;AAGd,kBAAc,KAAA;AACd,gBAAY,QAAA;AACZ,WAAO,WAAA;AACP,UAAM;AAAA,EACR;AAEA,QAAM,aAAiC;AAAA,IACrC,MAAM,aAAa;AACjB,oBAAc,KAAA;AACd,kBAAY,QAAA;AACZ,aAAO,WAAA;AACP,oBAAc;AACd,WAAK,iBAAiB,EAAE,QAAQ,OAAA,CAAQ;AAAA,IAC1C;AAAA,IAEA,MAAM,oBAAoB;AAExB,oBAAc,kBAAA;AACd,oBAAc;AACd,WAAK,eAAe,EAAE,MAAM,YAAA,CAAa;AAAA,IAC3C;AAAA,IAEA,mBAAmB;AACjB,oBAAc,iBAAA;AACd,oBAAc;AACd,WAAK,eAAe,EAAE,MAAM,OAAA,CAAQ;AAAA,IACtC;AAAA,IAEA,SAAS,MAAc;AACrB,aAAO,gBAAgB,IAAI;AAC3B,oBAAc;AACd,WAAK,eAAe,EAAE,MAAM,WAAA,CAAY;AAAA,IAC1C;AAAA,IAEA,UAAU,WAAmB,UAAkB;AAC7C,aAAO,iBAAiB,WAAW,QAAQ;AAC3C,oBAAc;AACd,WAAK,eAAe,EAAE,MAAM,WAAA,CAAY;AAAA,IAC1C;AAAA,IAEA,eAAe,QAAgB,QAAgB;AAC7C,aAAO,iBAAiB;AAAA,QACtB;AAAA,UACE,IAAI;AAAA,UACJ,UAAU;AAAA,YACR;AAAA,UAAA;AAAA,QACF;AAAA,MACF,CACD;AAAA,IACH;AAAA,IAEA,cAAc,QAAQ;AACpB,aAAO,cAAc,MAAM,EAAE,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,MAAA,CAAO,CAAC;AACtE,WAAK,iBAAiB,EAAE,QAAQ,eAAA,CAAgB;AAAA,IAClD;AAAA,IAEA,YAAY,UAAU;AACpB,aAAO,YAAY,QAAQ;AAC3B,WAAK,iBAAiB,EAAE,QAAQ,eAAA,CAAgB;AAAA,IAClD;AAAA,IAEA,YAAY;AACV,kBAAY,UAAA;AACZ,oBAAc;AACd,WAAK,eAAe,EAAE,MAAM,YAAA,CAAa;AACzC,WAAK,eAAe,EAAE,WAAW,oBAAoB,QAAW;AAAA,IAClE;AAAA,IACA,IAAI;AAAA,IACJ,wBAA4C;AAC1C,aAAO;AAAA,QACL,IAAI,aAAa;AACf,iBAAO,cAAc;AAAA,QACvB;AAAA,QAEA,IAAI,cAAc;AAChB,iBAAO,YAAY;AAAA,QACrB;AAAA,QAEA,wBAAwB;AACtB,iBAAO,cAAc;AAAA,QACvB;AAAA,QAEA,yBAAyB;AACvB,iBAAO,YAAY;AAAA,QACrB;AAAA,QAEA,yBAAyB;AACvB,iBAAO,cAAc;AAAA,QACvB;AAAA,QAEA,0BAA0B;AACxB,iBAAO,YAAY;AAAA,QACrB;AAAA,QAEA,IAAI,kBAAkB;AACpB,iBAAO,cAAc;AAAA,QACvB;AAAA,QAEA,IAAI,mBAAmB;AACrB,iBAAO,YAAY;AAAA,QACrB;AAAA,MAAA;AAAA,IAEJ;AAAA,EAAA;AAGF,SAAO;AACT;"}

package/dist/esm/realtime/client.d.ts

@@ -0,0 +1,98 @@
+import { FunctionResponse, LiveClientMessage, LiveServerGoAway, LiveServerMessage, LiveServerSessionResumptionUpdate, LiveServerToolCall, UsageMetadata } from '@google/genai';
+import { AnyClientTool, RealtimeSessionConfig, RealtimeToken } from '@tanstack/ai';
+import { GeminiRealtimeModel } from './types.js';
+interface LiveResponsePayloads {
+    text: string;
+    thought: string;
+    audio: {
+        audioData: string;
+        transcript: string;
+    };
+    setup_complete: string;
+    interrupted: string;
+    turn_complete: string;
+    tool_call: LiveServerToolCall;
+    session_resumption_update: LiveServerSessionResumptionUpdate;
+    go_away: LiveServerGoAway;
+    usage_metadata: UsageMetadata;
+    error: string;
+    input_transcription: {
+        text: string;
+        finished: boolean;
+    };
+    output_transcription: {
+        text: string;
+        finished: boolean;
+    };
+}
+export type MultimodalLiveResponseType = keyof LiveResponsePayloads;
+export type LiveResponse = {
+    [K in MultimodalLiveResponseType]: {
+        type: K;
+        data: LiveResponsePayloads[K];
+        endOfTurn: boolean;
+    };
+}[MultimodalLiveResponseType];
+/**
+ * Parses response messages from the Gemini Live API
+ */
+/**
+ * Parses ALL response types from a single server message.
+ * The server can now bundle multiple fields (e.g. audio + transcription)
+ * in the same message. Returns an array of response objects.
+ */
+export declare function parseResponseMessages(data: LiveServerMessage): Array<LiveResponse>;
+export declare class GeminiLiveClient {
+    private token;
+    private model;
+    private readonly responseModalities;
+    private systemInstructions;
+    private googleGrounding;
+    private voiceName;
+    private temperature;
+    private inputAudioTranscription;
+    private outputAudioTranscription;
+    private contextWindowCompression;
+    private proactiveAudio;
+    private enableAffectiveDialog;
+    private thinkingConfig;
+    private speechLanguageCode;
+    private maxOutputTokens;
+    private functionDeclarations;
+    private readonly automaticActivityDetection;
+    private readonly activityHandling;
+    private webSocket;
+    private lastResumptionUpdate;
+    private setupComplete;
+    private connected;
+    onReceiveResponse: (response: LiveResponse) => void;
+    onOpen: () => void;
+    onClose: () => void;
+    onError: (error: Error) => void;
+    constructor(token: string, model: GeminiRealtimeModel, tools?: ReadonlyArray<AnyClientTool>);
+    get isConnected(): boolean;
+    get isSetupComplete(): boolean;
+    /**
+     * Connection management
+     */
+    connect(): Promise<void>;
+    disconnect(): void;
+    /**
+     * Session management
+     */
+    sendInitialSetupMessage(resume?: boolean): void;
+    restartSession(resume?: boolean): Promise<void>;
+    updateToken(token: RealtimeToken): void;
+    updateSession(config: Partial<RealtimeSessionConfig>): Promise<void>;
+    /**
+     * Message transmission & receiving
+     */
+    sendMessage(message: LiveClientMessage): void;
+    onReceiveMessage(messageEvent: MessageEvent): Promise<void>;
+    sendRealtimeInputMessage(data: string, mimeType: string): void;
+    sendAudioMessage(base64PCM: string): void;
+    sendImageMessage(base64: string, mimeType?: string): void;
+    sendTextMessage(text: string): void;
+    sendToolResponse(functionResponses: Array<FunctionResponse>): void;
+}
+export {};