@tanstack/ai 0.31.0 → 0.33.0

package/dist/esm/activities/generateSpeech/index.js

@@ -1,6 +1,7 @@
 import { aiEventClient } from "@tanstack/ai-event-client";
 import { streamGenerationResult } from "../stream-generation-result.js";
 import { resolveDebugOption } from "../../logger/resolve.js";
+import { createGenerationContext, runGenerationStart, runGenerationUsage, runGenerationFinish, runGenerationError } from "../middleware/run.js";
 const kind = "tts";
 function createId(prefix) {
   return `${prefix}-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`;
@@ -14,12 +15,27 @@ function generateSpeech(options) {
   return runGenerateSpeech(options);
 }
 async function runGenerateSpeech(options) {
-  const { adapter, stream: _stream, debug: _debug, ...rest } = options;
+  const {
+    adapter,
+    stream: _stream,
+    debug: _debug,
+    middleware,
+    ...rest
+  } = options;
   const model = adapter.model;
   const requestId = createId("speech");
   const startTime = Date.now();
   const logger = resolveDebugOption(options.debug);
   const providerName = adapter.provider ?? adapter.name ?? "unknown";
+  const mwCtx = createGenerationContext({
+    requestId,
+    activity: "tts",
+    provider: adapter.name,
+    model,
+    modelOptions: rest.modelOptions,
+    createId
+  });
+  await runGenerationStart(middleware, mwCtx);
   aiEventClient.emit("speech:request:started", {
     requestId,
     provider: adapter.name,
@@ -63,6 +79,11 @@ async function runGenerateSpeech(options) {
       bytes: result.audio.length,
       contentType: result.contentType
     });
+    if (result.usage) await runGenerationUsage(middleware, mwCtx, result.usage);
+    await runGenerationFinish(middleware, mwCtx, {
+      duration,
+      usage: result.usage
+    });
     return result;
   } catch (error) {
     const duration = Date.now() - startTime;
@@ -76,6 +97,10 @@ async function runGenerateSpeech(options) {
       modelOptions: rest.modelOptions,
       timestamp: Date.now()
     });
+    await runGenerationError(middleware, mwCtx, {
+      error,
+      duration
+    });
     logger.errors("generateSpeech activity failed", {
       error,
       source: "generateSpeech"

package/dist/esm/activities/generateSpeech/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sources":["../../../../src/activities/generateSpeech/index.ts"],"sourcesContent":["/**\n * TTS Activity\n *\n * Generates speech audio from text using text-to-speech models.\n * This is a self-contained module with implementation, types, and JSDoc.\n */\n\nimport { aiEventClient } from '@tanstack/ai-event-client'\nimport { streamGenerationResult } from '../stream-generation-result.js'\nimport { resolveDebugOption } from '../../logger/resolve'\nimport type { InternalLogger } from '../../logger/internal-logger'\nimport type { DebugOption } from '../../logger/types'\nimport type { TTSAdapter } from './adapter'\nimport type { StreamChunk, TTSResult } from '../../types'\n\n// ===========================\n// Activity Kind\n// ===========================\n\n/** The adapter kind this activity handles */\nexport const kind = 'tts' as const\n\n// ===========================\n// Type Extraction Helpers\n// ===========================\n\n/**\n * Extract provider options from a TTSAdapter via ~types.\n */\nexport type TTSProviderOptions<TAdapter> =\n  TAdapter extends TTSAdapter<any, any>\n    ? TAdapter['~types']['providerOptions']\n    : object\n\n// ===========================\n// Activity Options Type\n// ===========================\n\n/**\n * Options for the TTS activity.\n * The model is extracted from the adapter's model property.\n *\n * @template TAdapter - The TTS adapter type\n * @template TStream - Whether to stream the output\n */\nexport interface TTSActivityOptions<\n  TAdapter extends TTSAdapter<string, TTSProviderOptions<TAdapter>>,\n  TStream extends boolean = false,\n> {\n  /** The TTS adapter to use (must be created with a model) */\n  adapter: TAdapter & { kind: typeof kind }\n  /** The text to convert to speech */\n  text: string\n  /** The voice to use for generation */\n  voice?: string\n  /** The output audio format */\n  format?: 'mp3' | 'opus' | 'aac' | 'flac' | 'wav' | 'pcm'\n  /** The speed of the generated audio (0.25 to 4.0) */\n  speed?: number\n  /** Provider-specific options for TTS generation */\n  modelOptions?: TTSProviderOptions<TAdapter>\n  /**\n   * Whether to stream the generation result.\n   * When true, returns an AsyncIterable<StreamChunk> for streaming transport.\n   * When false or not provided, returns a Promise<TTSResult>.\n   *\n   * @default false\n   */\n  stream?: TStream\n  /**\n   * Enable debug logging. Pass `true` to enable all categories, `false` to\n   * silence everything including errors, or a `DebugConfig` object for granular\n   * control and/or a custom `Logger`.\n   */\n  debug?: DebugOption\n}\n\n// ===========================\n// Activity Result Type\n// ===========================\n\n/**\n * Result type for the TTS activity.\n * - If stream is true: AsyncIterable<StreamChunk>\n * - Otherwise: Promise<TTSResult>\n */\nexport type TTSActivityResult<TStream extends boolean = false> =\n  TStream extends true ? AsyncIterable<StreamChunk> : Promise<TTSResult>\n\nfunction createId(prefix: string): string {\n  return `${prefix}-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`\n}\n\n// ===========================\n// Activity Implementation\n// ===========================\n\n/**\n * TTS activity - generates speech from text.\n *\n * Uses AI text-to-speech models to create audio from natural language text.\n *\n * @example Generate speech from text\n * ```ts\n * import { generateSpeech } from '@tanstack/ai'\n * import { openaiSpeech } from '@tanstack/ai-openai'\n *\n * const result = await generateSpeech({\n *   adapter: openaiSpeech('tts-1-hd'),\n *   text: 'Hello, welcome to TanStack AI!',\n *   voice: 'nova'\n * })\n *\n * console.log(result.audio) // base64-encoded audio\n * ```\n *\n * @example With format and speed options\n * ```ts\n * const result = await generateSpeech({\n *   adapter: openaiSpeech('tts-1'),\n *   text: 'This is slower speech.',\n *   voice: 'alloy',\n *   format: 'wav',\n *   speed: 0.8\n * })\n * ```\n */\nexport function generateSpeech<\n  TAdapter extends TTSAdapter<string, TTSProviderOptions<TAdapter>>,\n  TStream extends boolean = false,\n>(options: TTSActivityOptions<TAdapter, TStream>): TTSActivityResult<TStream> {\n  if (options.stream) {\n    return streamGenerationResult(() =>\n      runGenerateSpeech(options),\n    ) as TTSActivityResult<TStream>\n  }\n  return runGenerateSpeech(options) as TTSActivityResult<TStream>\n}\n\n/**\n * Run the core TTS generation logic (non-streaming).\n */\nasync function runGenerateSpeech<\n  TAdapter extends TTSAdapter<string, TTSProviderOptions<TAdapter>>,\n>(options: TTSActivityOptions<TAdapter, boolean>): Promise<TTSResult> {\n  const { adapter, stream: _stream, debug: _debug, ...rest } = options\n  const model = adapter.model\n  const requestId = createId('speech')\n  const startTime = Date.now()\n  const logger: InternalLogger = resolveDebugOption(options.debug)\n  const providerName =\n    (adapter as { name?: string; provider?: string }).provider ??\n    (adapter as { name?: string }).name ??\n    'unknown'\n\n  aiEventClient.emit('speech:request:started', {\n    requestId,\n    provider: adapter.name,\n    model,\n    text: rest.text,\n    voice: rest.voice,\n    format: rest.format,\n    speed: rest.speed,\n    modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n    timestamp: startTime,\n  })\n\n  logger.request(`activity=generateSpeech provider=${providerName}`, {\n    provider: providerName,\n    model,\n  })\n\n  try {\n    const result = await adapter.generateSpeech({ ...rest, model, logger })\n    const duration = Date.now() - startTime\n\n    aiEventClient.emit('speech:request:completed', {\n      requestId,\n      provider: adapter.name,\n      model,\n      audio: result.audio,\n      format: result.format,\n      audioDuration: result.duration,\n      contentType: result.contentType,\n      duration,\n      modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n      timestamp: Date.now(),\n    })\n\n    if (result.usage) {\n      aiEventClient.emit('speech:usage', {\n        requestId,\n        model,\n        usage: result.usage,\n        modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n        timestamp: Date.now(),\n      })\n    }\n\n    logger.output(`activity=generateSpeech bytes=${result.audio.length}`, {\n      bytes: result.audio.length,\n      contentType: result.contentType,\n    })\n\n    return result\n  } catch (error) {\n    const duration = Date.now() - startTime\n    const err = error as Error\n    aiEventClient.emit('speech:request:error', {\n      requestId,\n      provider: adapter.name,\n      model,\n      error: { message: err.message, name: err.name },\n      duration,\n      modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n      timestamp: Date.now(),\n    })\n    logger.errors('generateSpeech activity failed', {\n      error,\n      source: 'generateSpeech',\n    })\n    throw error\n  }\n}\n\n// ===========================\n// Options Factory\n// ===========================\n\n/**\n * Create typed options for the generateSpeech() function without executing.\n */\nexport function createSpeechOptions<\n  TAdapter extends TTSAdapter<string, TTSProviderOptions<TAdapter>>,\n  TStream extends boolean = false,\n>(\n  options: TTSActivityOptions<TAdapter, TStream>,\n): TTSActivityOptions<TAdapter, TStream> {\n  return options\n}\n\n// Re-export adapter types\nexport type { TTSAdapter, TTSAdapterConfig, AnyTTSAdapter } from './adapter'\nexport { BaseTTSAdapter } from './adapter'\n"],"names":[],"mappings":";;;AAoBO,MAAM,OAAO;AAqEpB,SAAS,SAAS,QAAwB;AACxC,SAAO,GAAG,MAAM,IAAI,KAAK,IAAA,CAAK,IAAI,KAAK,OAAA,EAAS,SAAS,EAAE,EAAE,MAAM,GAAG,CAAC,CAAC;AAC1E;AAoCO,SAAS,eAGd,SAA4E;AAC5E,MAAI,QAAQ,QAAQ;AAClB,WAAO;AAAA,MAAuB,MAC5B,kBAAkB,OAAO;AAAA,IAAA;AAAA,EAE7B;AACA,SAAO,kBAAkB,OAAO;AAClC;AAKA,eAAe,kBAEb,SAAoE;AACpE,QAAM,EAAE,SAAS,QAAQ,SAAS,OAAO,QAAQ,GAAG,SAAS;AAC7D,QAAM,QAAQ,QAAQ;AACtB,QAAM,YAAY,SAAS,QAAQ;AACnC,QAAM,YAAY,KAAK,IAAA;AACvB,QAAM,SAAyB,mBAAmB,QAAQ,KAAK;AAC/D,QAAM,eACH,QAAiD,YACjD,QAA8B,QAC/B;AAEF,gBAAc,KAAK,0BAA0B;AAAA,IAC3C;AAAA,IACA,UAAU,QAAQ;AAAA,IAClB;AAAA,IACA,MAAM,KAAK;AAAA,IACX,OAAO,KAAK;AAAA,IACZ,QAAQ,KAAK;AAAA,IACb,OAAO,KAAK;AAAA,IACZ,cAAc,KAAK;AAAA,IACnB,WAAW;AAAA,EAAA,CACZ;AAED,SAAO,QAAQ,oCAAoC,YAAY,IAAI;AAAA,IACjE,UAAU;AAAA,IACV;AAAA,EAAA,CACD;AAED,MAAI;AACF,UAAM,SAAS,MAAM,QAAQ,eAAe,EAAE,GAAG,MAAM,OAAO,QAAQ;AACtE,UAAM,WAAW,KAAK,IAAA,IAAQ;AAE9B,kBAAc,KAAK,4BAA4B;AAAA,MAC7C;AAAA,MACA,UAAU,QAAQ;AAAA,MAClB;AAAA,MACA,OAAO,OAAO;AAAA,MACd,QAAQ,OAAO;AAAA,MACf,eAAe,OAAO;AAAA,MACtB,aAAa,OAAO;AAAA,MACpB;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,WAAW,KAAK,IAAA;AAAA,IAAI,CACrB;AAED,QAAI,OAAO,OAAO;AAChB,oBAAc,KAAK,gBAAgB;AAAA,QACjC;AAAA,QACA;AAAA,QACA,OAAO,OAAO;AAAA,QACd,cAAc,KAAK;AAAA,QACnB,WAAW,KAAK,IAAA;AAAA,MAAI,CACrB;AAAA,IACH;AAEA,WAAO,OAAO,iCAAiC,OAAO,MAAM,MAAM,IAAI;AAAA,MACpE,OAAO,OAAO,MAAM;AAAA,MACpB,aAAa,OAAO;AAAA,IAAA,CACrB;AAED,WAAO;AAAA,EACT,SAAS,OAAO;AACd,UAAM,WAAW,KAAK,IAAA,IAAQ;AAC9B,UAAM,MAAM;AACZ,kBAAc,KAAK,wBAAwB;AAAA,MACzC;AAAA,MACA,UAAU,QAAQ;AAAA,MAClB;AAAA,MACA,OAAO,EAAE,SAAS,IAAI,SAAS,MAAM,IAAI,KAAA;AAAA,MACzC;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,WAAW,KAAK,IAAA;AAAA,IAAI,CACrB;AACD,WAAO,OAAO,kCAAkC;AAAA,MAC9C;AAAA,MACA,QAAQ;AAAA,IAAA,CACT;AACD,UAAM;AAAA,EACR;AACF;AASO,SAAS,oBAId,SACuC;AACvC,SAAO;AACT;"}
+{"version":3,"file":"index.js","sources":["../../../../src/activities/generateSpeech/index.ts"],"sourcesContent":["/**\n * TTS Activity\n *\n * Generates speech audio from text using text-to-speech models.\n * This is a self-contained module with implementation, types, and JSDoc.\n */\n\nimport { aiEventClient } from '@tanstack/ai-event-client'\nimport { streamGenerationResult } from '../stream-generation-result.js'\nimport { resolveDebugOption } from '../../logger/resolve'\nimport {\n  createGenerationContext,\n  runGenerationError,\n  runGenerationFinish,\n  runGenerationStart,\n  runGenerationUsage,\n} from '../middleware'\nimport type { InternalLogger } from '../../logger/internal-logger'\nimport type { DebugOption } from '../../logger/types'\nimport type { GenerationMiddleware } from '../middleware'\nimport type { TTSAdapter } from './adapter'\nimport type { StreamChunk, TTSResult } from '../../types'\n\n// ===========================\n// Activity Kind\n// ===========================\n\n/** The adapter kind this activity handles */\nexport const kind = 'tts' as const\n\n// ===========================\n// Type Extraction Helpers\n// ===========================\n\n/**\n * Extract provider options from a TTSAdapter via ~types.\n */\nexport type TTSProviderOptions<TAdapter> =\n  TAdapter extends TTSAdapter<any, any>\n    ? TAdapter['~types']['providerOptions']\n    : object\n\n// ===========================\n// Activity Options Type\n// ===========================\n\n/**\n * Options for the TTS activity.\n * The model is extracted from the adapter's model property.\n *\n * @template TAdapter - The TTS adapter type\n * @template TStream - Whether to stream the output\n */\nexport interface TTSActivityOptions<\n  TAdapter extends TTSAdapter<string, TTSProviderOptions<TAdapter>>,\n  TStream extends boolean = false,\n> {\n  /** The TTS adapter to use (must be created with a model) */\n  adapter: TAdapter & { kind: typeof kind }\n  /** The text to convert to speech */\n  text: string\n  /** The voice to use for generation */\n  voice?: string\n  /** The output audio format */\n  format?: 'mp3' | 'opus' | 'aac' | 'flac' | 'wav' | 'pcm'\n  /** The speed of the generated audio (0.25 to 4.0) */\n  speed?: number\n  /** Provider-specific options for TTS generation */\n  modelOptions?: TTSProviderOptions<TAdapter>\n  /**\n   * Whether to stream the generation result.\n   * When true, returns an AsyncIterable<StreamChunk> for streaming transport.\n   * When false or not provided, returns a Promise<TTSResult>.\n   *\n   * @default false\n   */\n  stream?: TStream\n  /**\n   * Enable debug logging. Pass `true` to enable all categories, `false` to\n   * silence everything including errors, or a `DebugConfig` object for granular\n   * control and/or a custom `Logger`.\n   */\n  debug?: DebugOption\n  /**\n   * Observe-only middleware notified on start, usage, success, and error. Pass\n   * `otelMiddleware()` to emit OpenTelemetry spans, or implement the\n   * `GenerationMiddleware` contract for a custom backend.\n   */\n  middleware?: Array<GenerationMiddleware>\n}\n\n// ===========================\n// Activity Result Type\n// ===========================\n\n/**\n * Result type for the TTS activity.\n * - If stream is true: AsyncIterable<StreamChunk>\n * - Otherwise: Promise<TTSResult>\n */\nexport type TTSActivityResult<TStream extends boolean = false> =\n  TStream extends true ? AsyncIterable<StreamChunk> : Promise<TTSResult>\n\nfunction createId(prefix: string): string {\n  return `${prefix}-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`\n}\n\n// ===========================\n// Activity Implementation\n// ===========================\n\n/**\n * TTS activity - generates speech from text.\n *\n * Uses AI text-to-speech models to create audio from natural language text.\n *\n * @example Generate speech from text\n * ```ts\n * import { generateSpeech } from '@tanstack/ai'\n * import { openaiSpeech } from '@tanstack/ai-openai'\n *\n * const result = await generateSpeech({\n *   adapter: openaiSpeech('tts-1-hd'),\n *   text: 'Hello, welcome to TanStack AI!',\n *   voice: 'nova'\n * })\n *\n * console.log(result.audio) // base64-encoded audio\n * ```\n *\n * @example With format and speed options\n * ```ts\n * const result = await generateSpeech({\n *   adapter: openaiSpeech('tts-1'),\n *   text: 'This is slower speech.',\n *   voice: 'alloy',\n *   format: 'wav',\n *   speed: 0.8\n * })\n * ```\n */\nexport function generateSpeech<\n  TAdapter extends TTSAdapter<string, TTSProviderOptions<TAdapter>>,\n  TStream extends boolean = false,\n>(options: TTSActivityOptions<TAdapter, TStream>): TTSActivityResult<TStream> {\n  if (options.stream) {\n    return streamGenerationResult(() =>\n      runGenerateSpeech(options),\n    ) as TTSActivityResult<TStream>\n  }\n  return runGenerateSpeech(options) as TTSActivityResult<TStream>\n}\n\n/**\n * Run the core TTS generation logic (non-streaming).\n */\nasync function runGenerateSpeech<\n  TAdapter extends TTSAdapter<string, TTSProviderOptions<TAdapter>>,\n>(options: TTSActivityOptions<TAdapter, boolean>): Promise<TTSResult> {\n  const {\n    adapter,\n    stream: _stream,\n    debug: _debug,\n    middleware,\n    ...rest\n  } = options\n  const model = adapter.model\n  const requestId = createId('speech')\n  const startTime = Date.now()\n  const logger: InternalLogger = resolveDebugOption(options.debug)\n  const providerName =\n    (adapter as { name?: string; provider?: string }).provider ??\n    (adapter as { name?: string }).name ??\n    'unknown'\n\n  const mwCtx = createGenerationContext({\n    requestId,\n    activity: 'tts',\n    provider: adapter.name,\n    model,\n    modelOptions: rest.modelOptions,\n    createId,\n  })\n\n  await runGenerationStart(middleware, mwCtx)\n\n  aiEventClient.emit('speech:request:started', {\n    requestId,\n    provider: adapter.name,\n    model,\n    text: rest.text,\n    voice: rest.voice,\n    format: rest.format,\n    speed: rest.speed,\n    modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n    timestamp: startTime,\n  })\n\n  logger.request(`activity=generateSpeech provider=${providerName}`, {\n    provider: providerName,\n    model,\n  })\n\n  try {\n    const result = await adapter.generateSpeech({ ...rest, model, logger })\n    const duration = Date.now() - startTime\n\n    aiEventClient.emit('speech:request:completed', {\n      requestId,\n      provider: adapter.name,\n      model,\n      audio: result.audio,\n      format: result.format,\n      audioDuration: result.duration,\n      contentType: result.contentType,\n      duration,\n      modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n      timestamp: Date.now(),\n    })\n\n    if (result.usage) {\n      aiEventClient.emit('speech:usage', {\n        requestId,\n        model,\n        usage: result.usage,\n        modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n        timestamp: Date.now(),\n      })\n    }\n\n    logger.output(`activity=generateSpeech bytes=${result.audio.length}`, {\n      bytes: result.audio.length,\n      contentType: result.contentType,\n    })\n\n    if (result.usage) await runGenerationUsage(middleware, mwCtx, result.usage)\n    await runGenerationFinish(middleware, mwCtx, {\n      duration,\n      usage: result.usage,\n    })\n\n    return result\n  } catch (error) {\n    const duration = Date.now() - startTime\n    const err = error as Error\n    aiEventClient.emit('speech:request:error', {\n      requestId,\n      provider: adapter.name,\n      model,\n      error: { message: err.message, name: err.name },\n      duration,\n      modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n      timestamp: Date.now(),\n    })\n    await runGenerationError(middleware, mwCtx, {\n      error,\n      duration,\n    })\n    logger.errors('generateSpeech activity failed', {\n      error,\n      source: 'generateSpeech',\n    })\n    throw error\n  }\n}\n\n// ===========================\n// Options Factory\n// ===========================\n\n/**\n * Create typed options for the generateSpeech() function without executing.\n */\nexport function createSpeechOptions<\n  TAdapter extends TTSAdapter<string, TTSProviderOptions<TAdapter>>,\n  TStream extends boolean = false,\n>(\n  options: TTSActivityOptions<TAdapter, TStream>,\n): TTSActivityOptions<TAdapter, TStream> {\n  return options\n}\n\n// Re-export adapter types\nexport type { TTSAdapter, TTSAdapterConfig, AnyTTSAdapter } from './adapter'\nexport { BaseTTSAdapter } from './adapter'\n"],"names":[],"mappings":";;;;AA4BO,MAAM,OAAO;AA2EpB,SAAS,SAAS,QAAwB;AACxC,SAAO,GAAG,MAAM,IAAI,KAAK,IAAA,CAAK,IAAI,KAAK,OAAA,EAAS,SAAS,EAAE,EAAE,MAAM,GAAG,CAAC,CAAC;AAC1E;AAoCO,SAAS,eAGd,SAA4E;AAC5E,MAAI,QAAQ,QAAQ;AAClB,WAAO;AAAA,MAAuB,MAC5B,kBAAkB,OAAO;AAAA,IAAA;AAAA,EAE7B;AACA,SAAO,kBAAkB,OAAO;AAClC;AAKA,eAAe,kBAEb,SAAoE;AACpE,QAAM;AAAA,IACJ;AAAA,IACA,QAAQ;AAAA,IACR,OAAO;AAAA,IACP;AAAA,IACA,GAAG;AAAA,EAAA,IACD;AACJ,QAAM,QAAQ,QAAQ;AACtB,QAAM,YAAY,SAAS,QAAQ;AACnC,QAAM,YAAY,KAAK,IAAA;AACvB,QAAM,SAAyB,mBAAmB,QAAQ,KAAK;AAC/D,QAAM,eACH,QAAiD,YACjD,QAA8B,QAC/B;AAEF,QAAM,QAAQ,wBAAwB;AAAA,IACpC;AAAA,IACA,UAAU;AAAA,IACV,UAAU,QAAQ;AAAA,IAClB;AAAA,IACA,cAAc,KAAK;AAAA,IACnB;AAAA,EAAA,CACD;AAED,QAAM,mBAAmB,YAAY,KAAK;AAE1C,gBAAc,KAAK,0BAA0B;AAAA,IAC3C;AAAA,IACA,UAAU,QAAQ;AAAA,IAClB;AAAA,IACA,MAAM,KAAK;AAAA,IACX,OAAO,KAAK;AAAA,IACZ,QAAQ,KAAK;AAAA,IACb,OAAO,KAAK;AAAA,IACZ,cAAc,KAAK;AAAA,IACnB,WAAW;AAAA,EAAA,CACZ;AAED,SAAO,QAAQ,oCAAoC,YAAY,IAAI;AAAA,IACjE,UAAU;AAAA,IACV;AAAA,EAAA,CACD;AAED,MAAI;AACF,UAAM,SAAS,MAAM,QAAQ,eAAe,EAAE,GAAG,MAAM,OAAO,QAAQ;AACtE,UAAM,WAAW,KAAK,IAAA,IAAQ;AAE9B,kBAAc,KAAK,4BAA4B;AAAA,MAC7C;AAAA,MACA,UAAU,QAAQ;AAAA,MAClB;AAAA,MACA,OAAO,OAAO;AAAA,MACd,QAAQ,OAAO;AAAA,MACf,eAAe,OAAO;AAAA,MACtB,aAAa,OAAO;AAAA,MACpB;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,WAAW,KAAK,IAAA;AAAA,IAAI,CACrB;AAED,QAAI,OAAO,OAAO;AAChB,oBAAc,KAAK,gBAAgB;AAAA,QACjC;AAAA,QACA;AAAA,QACA,OAAO,OAAO;AAAA,QACd,cAAc,KAAK;AAAA,QACnB,WAAW,KAAK,IAAA;AAAA,MAAI,CACrB;AAAA,IACH;AAEA,WAAO,OAAO,iCAAiC,OAAO,MAAM,MAAM,IAAI;AAAA,MACpE,OAAO,OAAO,MAAM;AAAA,MACpB,aAAa,OAAO;AAAA,IAAA,CACrB;AAED,QAAI,OAAO,MAAO,OAAM,mBAAmB,YAAY,OAAO,OAAO,KAAK;AAC1E,UAAM,oBAAoB,YAAY,OAAO;AAAA,MAC3C;AAAA,MACA,OAAO,OAAO;AAAA,IAAA,CACf;AAED,WAAO;AAAA,EACT,SAAS,OAAO;AACd,UAAM,WAAW,KAAK,IAAA,IAAQ;AAC9B,UAAM,MAAM;AACZ,kBAAc,KAAK,wBAAwB;AAAA,MACzC;AAAA,MACA,UAAU,QAAQ;AAAA,MAClB;AAAA,MACA,OAAO,EAAE,SAAS,IAAI,SAAS,MAAM,IAAI,KAAA;AAAA,MACzC;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,WAAW,KAAK,IAAA;AAAA,IAAI,CACrB;AACD,UAAM,mBAAmB,YAAY,OAAO;AAAA,MAC1C;AAAA,MACA;AAAA,IAAA,CACD;AACD,WAAO,OAAO,kCAAkC;AAAA,MAC9C;AAAA,MACA,QAAQ;AAAA,IAAA,CACT;AACD,UAAM;AAAA,EACR;AACF;AASO,SAAS,oBAId,SACuC;AACvC,SAAO;AACT;"}

package/dist/esm/activities/generateTranscription/index.d.ts

@@ -1,4 +1,5 @@
 import { DebugOption } from '../../logger/types.js';
+import { GenerationMiddleware } from '../middleware.js';
 import { TranscriptionAdapter } from './adapter.js';
 import { StreamChunk, TranscriptionResult } from '../../types.js';
 /** The adapter kind this activity handles */
@@ -43,6 +44,12 @@ export interface TranscriptionActivityOptions<TAdapter extends TranscriptionAdap
      * control and/or a custom `Logger`.
      */
     debug?: DebugOption;
+    /**
+     * Observe-only middleware notified on start, usage, success, and error. Pass
+     * `otelMiddleware()` to emit OpenTelemetry spans, or implement the
+     * `GenerationMiddleware` contract for a custom backend.
+     */
+    middleware?: Array<GenerationMiddleware>;
 }
 /**
  * Result type for the transcription activity.

package/dist/esm/activities/generateTranscription/index.js

@@ -1,6 +1,7 @@
 import { aiEventClient } from "@tanstack/ai-event-client";
 import { streamGenerationResult } from "../stream-generation-result.js";
 import { resolveDebugOption } from "../../logger/resolve.js";
+import { createGenerationContext, runGenerationStart, runGenerationUsage, runGenerationFinish, runGenerationError } from "../middleware/run.js";
 const kind = "transcription";
 function createId(prefix) {
   return `${prefix}-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`;
@@ -16,12 +17,27 @@ function generateTranscription(options) {
   );
 }
 async function runGenerateTranscription(options) {
-  const { adapter, stream: _stream, debug: _debug, ...rest } = options;
+  const {
+    adapter,
+    stream: _stream,
+    debug: _debug,
+    middleware,
+    ...rest
+  } = options;
   const model = adapter.model;
   const requestId = createId("transcription");
   const startTime = Date.now();
   const logger = resolveDebugOption(options.debug);
   const providerName = adapter.provider ?? adapter.name ?? "unknown";
+  const mwCtx = createGenerationContext({
+    requestId,
+    activity: "transcription",
+    provider: adapter.name,
+    model,
+    modelOptions: rest.modelOptions,
+    createId
+  });
+  await runGenerationStart(middleware, mwCtx);
   aiEventClient.emit("transcription:request:started", {
     requestId,
     provider: adapter.name,
@@ -53,6 +69,11 @@ async function runGenerateTranscription(options) {
       `activity=generateTranscription length=${result.text.length}`,
       { hasText: !!result.text }
     );
+    if (result.usage) await runGenerationUsage(middleware, mwCtx, result.usage);
+    await runGenerationFinish(middleware, mwCtx, {
+      duration,
+      usage: result.usage
+    });
     return result;
   } catch (error) {
     const duration = Date.now() - startTime;
@@ -66,6 +87,10 @@ async function runGenerateTranscription(options) {
       modelOptions: rest.modelOptions,
       timestamp: Date.now()
     });
+    await runGenerationError(middleware, mwCtx, {
+      error,
+      duration
+    });
     logger.errors("generateTranscription activity failed", {
       error,
       source: "generateTranscription"

package/dist/esm/activities/generateTranscription/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sources":["../../../../src/activities/generateTranscription/index.ts"],"sourcesContent":["/**\n * Transcription Activity\n *\n * Transcribes audio to text using speech-to-text models.\n * This is a self-contained module with implementation, types, and JSDoc.\n */\n\nimport { aiEventClient } from '@tanstack/ai-event-client'\nimport { streamGenerationResult } from '../stream-generation-result.js'\nimport { resolveDebugOption } from '../../logger/resolve'\nimport type { InternalLogger } from '../../logger/internal-logger'\nimport type { DebugOption } from '../../logger/types'\nimport type { TranscriptionAdapter } from './adapter'\nimport type { StreamChunk, TranscriptionResult } from '../../types'\n\n// ===========================\n// Activity Kind\n// ===========================\n\n/** The adapter kind this activity handles */\nexport const kind = 'transcription' as const\n\n// ===========================\n// Type Extraction Helpers\n// ===========================\n\n/**\n * Extract provider options from a TranscriptionAdapter via ~types.\n */\nexport type TranscriptionProviderOptions<TAdapter> =\n  TAdapter extends TranscriptionAdapter<any, any>\n    ? TAdapter['~types']['providerOptions']\n    : object\n\n// ===========================\n// Activity Options Type\n// ===========================\n\n/**\n * Options for the transcription activity.\n * The model is extracted from the adapter's model property.\n *\n * @template TAdapter - The transcription adapter type\n * @template TStream - Whether to stream the output\n */\nexport interface TranscriptionActivityOptions<\n  TAdapter extends TranscriptionAdapter<\n    string,\n    TranscriptionProviderOptions<TAdapter>\n  >,\n  TStream extends boolean = false,\n> {\n  /** The transcription adapter to use (must be created with a model) */\n  adapter: TAdapter & { kind: typeof kind }\n  /** The audio data to transcribe - can be base64 string, File, Blob, or Buffer */\n  audio: string | File | Blob | ArrayBuffer\n  /** The language of the audio in ISO-639-1 format (e.g., 'en') */\n  language?: string\n  /** An optional prompt to guide the transcription */\n  prompt?: string\n  /** The format of the transcription output */\n  responseFormat?: 'json' | 'text' | 'srt' | 'verbose_json' | 'vtt'\n  /** Provider-specific options for transcription */\n  modelOptions?: TranscriptionProviderOptions<TAdapter>\n  /**\n   * Whether to stream the transcription result.\n   * When true, returns an AsyncIterable<StreamChunk> for streaming transport.\n   * When false or not provided, returns a Promise<TranscriptionResult>.\n   *\n   * @default false\n   */\n  stream?: TStream\n  /**\n   * Enable debug logging. Pass `true` to enable all categories, `false` to\n   * silence everything including errors, or a `DebugConfig` object for granular\n   * control and/or a custom `Logger`.\n   */\n  debug?: DebugOption\n}\n\n// ===========================\n// Activity Result Type\n// ===========================\n\n/**\n * Result type for the transcription activity.\n * - If stream is true: AsyncIterable<StreamChunk>\n * - Otherwise: Promise<TranscriptionResult>\n */\nexport type TranscriptionActivityResult<TStream extends boolean = false> =\n  TStream extends true\n    ? AsyncIterable<StreamChunk>\n    : Promise<TranscriptionResult>\n\nfunction createId(prefix: string): string {\n  return `${prefix}-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`\n}\n\n// ===========================\n// Activity Implementation\n// ===========================\n\n/**\n * Transcription activity - converts audio to text.\n *\n * Uses AI speech-to-text models to transcribe audio content.\n *\n * @example Transcribe an audio file\n * ```ts\n * import { generateTranscription } from '@tanstack/ai'\n * import { openaiTranscription } from '@tanstack/ai-openai'\n *\n * const result = await generateTranscription({\n *   adapter: openaiTranscription('whisper-1'),\n *   audio: audioFile, // File, Blob, or base64 string\n *   language: 'en'\n * })\n *\n * console.log(result.text)\n * ```\n *\n * @example With verbose output for timestamps\n * ```ts\n * const result = await generateTranscription({\n *   adapter: openaiTranscription('whisper-1'),\n *   audio: audioFile,\n *   responseFormat: 'verbose_json'\n * })\n *\n * result.segments?.forEach(segment => {\n *   console.log(`[${segment.start}s - ${segment.end}s]: ${segment.text}`)\n * })\n * ```\n *\n * @example Streaming transcription result\n * ```ts\n * for await (const chunk of generateTranscription({\n *   adapter: openaiTranscription('whisper-1'),\n *   audio: audioFile,\n *   stream: true\n * })) {\n *   console.log(chunk)\n * }\n * ```\n */\nexport function generateTranscription<\n  TAdapter extends TranscriptionAdapter<\n    string,\n    TranscriptionProviderOptions<TAdapter>\n  >,\n  TStream extends boolean = false,\n>(\n  options: TranscriptionActivityOptions<TAdapter, TStream>,\n): TranscriptionActivityResult<TStream> {\n  if (options.stream) {\n    return streamGenerationResult(() =>\n      runGenerateTranscription(options),\n    ) as TranscriptionActivityResult<TStream>\n  }\n\n  return runGenerateTranscription(\n    options,\n  ) as TranscriptionActivityResult<TStream>\n}\n\n/**\n * Run non-streaming transcription\n */\nasync function runGenerateTranscription<\n  TAdapter extends TranscriptionAdapter<\n    string,\n    TranscriptionProviderOptions<TAdapter>\n  >,\n>(\n  options: TranscriptionActivityOptions<TAdapter, boolean>,\n): Promise<TranscriptionResult> {\n  const { adapter, stream: _stream, debug: _debug, ...rest } = options\n  const model = adapter.model\n  const requestId = createId('transcription')\n  const startTime = Date.now()\n  const logger: InternalLogger = resolveDebugOption(options.debug)\n  const providerName =\n    (adapter as { name?: string; provider?: string }).provider ??\n    (adapter as { name?: string }).name ??\n    'unknown'\n\n  aiEventClient.emit('transcription:request:started', {\n    requestId,\n    provider: adapter.name,\n    model,\n    language: rest.language,\n    prompt: rest.prompt,\n    responseFormat: rest.responseFormat,\n    modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n    timestamp: startTime,\n  })\n\n  logger.request(`activity=generateTranscription provider=${providerName}`, {\n    provider: providerName,\n    model,\n  })\n\n  try {\n    const result = await adapter.transcribe({ ...rest, model, logger })\n    const duration = Date.now() - startTime\n\n    aiEventClient.emit('transcription:request:completed', {\n      requestId,\n      provider: adapter.name,\n      model,\n      text: result.text,\n      language: result.language,\n      duration,\n      modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n      timestamp: Date.now(),\n    })\n\n    logger.output(\n      `activity=generateTranscription length=${result.text.length}`,\n      { hasText: !!result.text },\n    )\n\n    return result\n  } catch (error) {\n    const duration = Date.now() - startTime\n    const err = error as Error\n    aiEventClient.emit('transcription:request:error', {\n      requestId,\n      provider: adapter.name,\n      model,\n      error: { message: err.message, name: err.name },\n      duration,\n      modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n      timestamp: Date.now(),\n    })\n    logger.errors('generateTranscription activity failed', {\n      error,\n      source: 'generateTranscription',\n    })\n    throw error\n  }\n}\n\n// ===========================\n// Options Factory\n// ===========================\n\n/**\n * Create typed options for the generateTranscription() function without executing.\n */\nexport function createTranscriptionOptions<\n  TAdapter extends TranscriptionAdapter<\n    string,\n    TranscriptionProviderOptions<TAdapter>\n  >,\n  TStream extends boolean = false,\n>(\n  options: TranscriptionActivityOptions<TAdapter, TStream>,\n): TranscriptionActivityOptions<TAdapter, TStream> {\n  return options\n}\n\n// Re-export adapter types\nexport type {\n  TranscriptionAdapter,\n  TranscriptionAdapterConfig,\n  AnyTranscriptionAdapter,\n} from './adapter'\nexport { BaseTranscriptionAdapter } from './adapter'\n"],"names":[],"mappings":";;;AAoBO,MAAM,OAAO;AA0EpB,SAAS,SAAS,QAAwB;AACxC,SAAO,GAAG,MAAM,IAAI,KAAK,IAAA,CAAK,IAAI,KAAK,OAAA,EAAS,SAAS,EAAE,EAAE,MAAM,GAAG,CAAC,CAAC;AAC1E;AAiDO,SAAS,sBAOd,SACsC;AACtC,MAAI,QAAQ,QAAQ;AAClB,WAAO;AAAA,MAAuB,MAC5B,yBAAyB,OAAO;AAAA,IAAA;AAAA,EAEpC;AAEA,SAAO;AAAA,IACL;AAAA,EAAA;AAEJ;AAKA,eAAe,yBAMb,SAC8B;AAC9B,QAAM,EAAE,SAAS,QAAQ,SAAS,OAAO,QAAQ,GAAG,SAAS;AAC7D,QAAM,QAAQ,QAAQ;AACtB,QAAM,YAAY,SAAS,eAAe;AAC1C,QAAM,YAAY,KAAK,IAAA;AACvB,QAAM,SAAyB,mBAAmB,QAAQ,KAAK;AAC/D,QAAM,eACH,QAAiD,YACjD,QAA8B,QAC/B;AAEF,gBAAc,KAAK,iCAAiC;AAAA,IAClD;AAAA,IACA,UAAU,QAAQ;AAAA,IAClB;AAAA,IACA,UAAU,KAAK;AAAA,IACf,QAAQ,KAAK;AAAA,IACb,gBAAgB,KAAK;AAAA,IACrB,cAAc,KAAK;AAAA,IACnB,WAAW;AAAA,EAAA,CACZ;AAED,SAAO,QAAQ,2CAA2C,YAAY,IAAI;AAAA,IACxE,UAAU;AAAA,IACV;AAAA,EAAA,CACD;AAED,MAAI;AACF,UAAM,SAAS,MAAM,QAAQ,WAAW,EAAE,GAAG,MAAM,OAAO,QAAQ;AAClE,UAAM,WAAW,KAAK,IAAA,IAAQ;AAE9B,kBAAc,KAAK,mCAAmC;AAAA,MACpD;AAAA,MACA,UAAU,QAAQ;AAAA,MAClB;AAAA,MACA,MAAM,OAAO;AAAA,MACb,UAAU,OAAO;AAAA,MACjB;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,WAAW,KAAK,IAAA;AAAA,IAAI,CACrB;AAED,WAAO;AAAA,MACL,yCAAyC,OAAO,KAAK,MAAM;AAAA,MAC3D,EAAE,SAAS,CAAC,CAAC,OAAO,KAAA;AAAA,IAAK;AAG3B,WAAO;AAAA,EACT,SAAS,OAAO;AACd,UAAM,WAAW,KAAK,IAAA,IAAQ;AAC9B,UAAM,MAAM;AACZ,kBAAc,KAAK,+BAA+B;AAAA,MAChD;AAAA,MACA,UAAU,QAAQ;AAAA,MAClB;AAAA,MACA,OAAO,EAAE,SAAS,IAAI,SAAS,MAAM,IAAI,KAAA;AAAA,MACzC;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,WAAW,KAAK,IAAA;AAAA,IAAI,CACrB;AACD,WAAO,OAAO,yCAAyC;AAAA,MACrD;AAAA,MACA,QAAQ;AAAA,IAAA,CACT;AACD,UAAM;AAAA,EACR;AACF;AASO,SAAS,2BAOd,SACiD;AACjD,SAAO;AACT;"}
+{"version":3,"file":"index.js","sources":["../../../../src/activities/generateTranscription/index.ts"],"sourcesContent":["/**\n * Transcription Activity\n *\n * Transcribes audio to text using speech-to-text models.\n * This is a self-contained module with implementation, types, and JSDoc.\n */\n\nimport { aiEventClient } from '@tanstack/ai-event-client'\nimport { streamGenerationResult } from '../stream-generation-result.js'\nimport { resolveDebugOption } from '../../logger/resolve'\nimport {\n  createGenerationContext,\n  runGenerationError,\n  runGenerationFinish,\n  runGenerationStart,\n  runGenerationUsage,\n} from '../middleware'\nimport type { InternalLogger } from '../../logger/internal-logger'\nimport type { DebugOption } from '../../logger/types'\nimport type { GenerationMiddleware } from '../middleware'\nimport type { TranscriptionAdapter } from './adapter'\nimport type { StreamChunk, TranscriptionResult } from '../../types'\n\n// ===========================\n// Activity Kind\n// ===========================\n\n/** The adapter kind this activity handles */\nexport const kind = 'transcription' as const\n\n// ===========================\n// Type Extraction Helpers\n// ===========================\n\n/**\n * Extract provider options from a TranscriptionAdapter via ~types.\n */\nexport type TranscriptionProviderOptions<TAdapter> =\n  TAdapter extends TranscriptionAdapter<any, any>\n    ? TAdapter['~types']['providerOptions']\n    : object\n\n// ===========================\n// Activity Options Type\n// ===========================\n\n/**\n * Options for the transcription activity.\n * The model is extracted from the adapter's model property.\n *\n * @template TAdapter - The transcription adapter type\n * @template TStream - Whether to stream the output\n */\nexport interface TranscriptionActivityOptions<\n  TAdapter extends TranscriptionAdapter<\n    string,\n    TranscriptionProviderOptions<TAdapter>\n  >,\n  TStream extends boolean = false,\n> {\n  /** The transcription adapter to use (must be created with a model) */\n  adapter: TAdapter & { kind: typeof kind }\n  /** The audio data to transcribe - can be base64 string, File, Blob, or Buffer */\n  audio: string | File | Blob | ArrayBuffer\n  /** The language of the audio in ISO-639-1 format (e.g., 'en') */\n  language?: string\n  /** An optional prompt to guide the transcription */\n  prompt?: string\n  /** The format of the transcription output */\n  responseFormat?: 'json' | 'text' | 'srt' | 'verbose_json' | 'vtt'\n  /** Provider-specific options for transcription */\n  modelOptions?: TranscriptionProviderOptions<TAdapter>\n  /**\n   * Whether to stream the transcription result.\n   * When true, returns an AsyncIterable<StreamChunk> for streaming transport.\n   * When false or not provided, returns a Promise<TranscriptionResult>.\n   *\n   * @default false\n   */\n  stream?: TStream\n  /**\n   * Enable debug logging. Pass `true` to enable all categories, `false` to\n   * silence everything including errors, or a `DebugConfig` object for granular\n   * control and/or a custom `Logger`.\n   */\n  debug?: DebugOption\n  /**\n   * Observe-only middleware notified on start, usage, success, and error. Pass\n   * `otelMiddleware()` to emit OpenTelemetry spans, or implement the\n   * `GenerationMiddleware` contract for a custom backend.\n   */\n  middleware?: Array<GenerationMiddleware>\n}\n\n// ===========================\n// Activity Result Type\n// ===========================\n\n/**\n * Result type for the transcription activity.\n * - If stream is true: AsyncIterable<StreamChunk>\n * - Otherwise: Promise<TranscriptionResult>\n */\nexport type TranscriptionActivityResult<TStream extends boolean = false> =\n  TStream extends true\n    ? AsyncIterable<StreamChunk>\n    : Promise<TranscriptionResult>\n\nfunction createId(prefix: string): string {\n  return `${prefix}-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`\n}\n\n// ===========================\n// Activity Implementation\n// ===========================\n\n/**\n * Transcription activity - converts audio to text.\n *\n * Uses AI speech-to-text models to transcribe audio content.\n *\n * @example Transcribe an audio file\n * ```ts\n * import { generateTranscription } from '@tanstack/ai'\n * import { openaiTranscription } from '@tanstack/ai-openai'\n *\n * const result = await generateTranscription({\n *   adapter: openaiTranscription('whisper-1'),\n *   audio: audioFile, // File, Blob, or base64 string\n *   language: 'en'\n * })\n *\n * console.log(result.text)\n * ```\n *\n * @example With verbose output for timestamps\n * ```ts\n * const result = await generateTranscription({\n *   adapter: openaiTranscription('whisper-1'),\n *   audio: audioFile,\n *   responseFormat: 'verbose_json'\n * })\n *\n * result.segments?.forEach(segment => {\n *   console.log(`[${segment.start}s - ${segment.end}s]: ${segment.text}`)\n * })\n * ```\n *\n * @example Streaming transcription result\n * ```ts\n * for await (const chunk of generateTranscription({\n *   adapter: openaiTranscription('whisper-1'),\n *   audio: audioFile,\n *   stream: true\n * })) {\n *   console.log(chunk)\n * }\n * ```\n */\nexport function generateTranscription<\n  TAdapter extends TranscriptionAdapter<\n    string,\n    TranscriptionProviderOptions<TAdapter>\n  >,\n  TStream extends boolean = false,\n>(\n  options: TranscriptionActivityOptions<TAdapter, TStream>,\n): TranscriptionActivityResult<TStream> {\n  if (options.stream) {\n    return streamGenerationResult(() =>\n      runGenerateTranscription(options),\n    ) as TranscriptionActivityResult<TStream>\n  }\n\n  return runGenerateTranscription(\n    options,\n  ) as TranscriptionActivityResult<TStream>\n}\n\n/**\n * Run non-streaming transcription\n */\nasync function runGenerateTranscription<\n  TAdapter extends TranscriptionAdapter<\n    string,\n    TranscriptionProviderOptions<TAdapter>\n  >,\n>(\n  options: TranscriptionActivityOptions<TAdapter, boolean>,\n): Promise<TranscriptionResult> {\n  const {\n    adapter,\n    stream: _stream,\n    debug: _debug,\n    middleware,\n    ...rest\n  } = options\n  const model = adapter.model\n  const requestId = createId('transcription')\n  const startTime = Date.now()\n  const logger: InternalLogger = resolveDebugOption(options.debug)\n  const providerName =\n    (adapter as { name?: string; provider?: string }).provider ??\n    (adapter as { name?: string }).name ??\n    'unknown'\n\n  const mwCtx = createGenerationContext({\n    requestId,\n    activity: 'transcription',\n    provider: adapter.name,\n    model,\n    modelOptions: rest.modelOptions,\n    createId,\n  })\n\n  await runGenerationStart(middleware, mwCtx)\n\n  aiEventClient.emit('transcription:request:started', {\n    requestId,\n    provider: adapter.name,\n    model,\n    language: rest.language,\n    prompt: rest.prompt,\n    responseFormat: rest.responseFormat,\n    modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n    timestamp: startTime,\n  })\n\n  logger.request(`activity=generateTranscription provider=${providerName}`, {\n    provider: providerName,\n    model,\n  })\n\n  try {\n    const result = await adapter.transcribe({ ...rest, model, logger })\n    const duration = Date.now() - startTime\n\n    aiEventClient.emit('transcription:request:completed', {\n      requestId,\n      provider: adapter.name,\n      model,\n      text: result.text,\n      language: result.language,\n      duration,\n      modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n      timestamp: Date.now(),\n    })\n\n    logger.output(\n      `activity=generateTranscription length=${result.text.length}`,\n      { hasText: !!result.text },\n    )\n\n    if (result.usage) await runGenerationUsage(middleware, mwCtx, result.usage)\n    await runGenerationFinish(middleware, mwCtx, {\n      duration,\n      usage: result.usage,\n    })\n\n    return result\n  } catch (error) {\n    const duration = Date.now() - startTime\n    const err = error as Error\n    aiEventClient.emit('transcription:request:error', {\n      requestId,\n      provider: adapter.name,\n      model,\n      error: { message: err.message, name: err.name },\n      duration,\n      modelOptions: rest.modelOptions as Record<string, unknown> | undefined,\n      timestamp: Date.now(),\n    })\n    await runGenerationError(middleware, mwCtx, {\n      error,\n      duration,\n    })\n    logger.errors('generateTranscription activity failed', {\n      error,\n      source: 'generateTranscription',\n    })\n    throw error\n  }\n}\n\n// ===========================\n// Options Factory\n// ===========================\n\n/**\n * Create typed options for the generateTranscription() function without executing.\n */\nexport function createTranscriptionOptions<\n  TAdapter extends TranscriptionAdapter<\n    string,\n    TranscriptionProviderOptions<TAdapter>\n  >,\n  TStream extends boolean = false,\n>(\n  options: TranscriptionActivityOptions<TAdapter, TStream>,\n): TranscriptionActivityOptions<TAdapter, TStream> {\n  return options\n}\n\n// Re-export adapter types\nexport type {\n  TranscriptionAdapter,\n  TranscriptionAdapterConfig,\n  AnyTranscriptionAdapter,\n} from './adapter'\nexport { BaseTranscriptionAdapter } from './adapter'\n"],"names":[],"mappings":";;;;AA4BO,MAAM,OAAO;AAgFpB,SAAS,SAAS,QAAwB;AACxC,SAAO,GAAG,MAAM,IAAI,KAAK,IAAA,CAAK,IAAI,KAAK,OAAA,EAAS,SAAS,EAAE,EAAE,MAAM,GAAG,CAAC,CAAC;AAC1E;AAiDO,SAAS,sBAOd,SACsC;AACtC,MAAI,QAAQ,QAAQ;AAClB,WAAO;AAAA,MAAuB,MAC5B,yBAAyB,OAAO;AAAA,IAAA;AAAA,EAEpC;AAEA,SAAO;AAAA,IACL;AAAA,EAAA;AAEJ;AAKA,eAAe,yBAMb,SAC8B;AAC9B,QAAM;AAAA,IACJ;AAAA,IACA,QAAQ;AAAA,IACR,OAAO;AAAA,IACP;AAAA,IACA,GAAG;AAAA,EAAA,IACD;AACJ,QAAM,QAAQ,QAAQ;AACtB,QAAM,YAAY,SAAS,eAAe;AAC1C,QAAM,YAAY,KAAK,IAAA;AACvB,QAAM,SAAyB,mBAAmB,QAAQ,KAAK;AAC/D,QAAM,eACH,QAAiD,YACjD,QAA8B,QAC/B;AAEF,QAAM,QAAQ,wBAAwB;AAAA,IACpC;AAAA,IACA,UAAU;AAAA,IACV,UAAU,QAAQ;AAAA,IAClB;AAAA,IACA,cAAc,KAAK;AAAA,IACnB;AAAA,EAAA,CACD;AAED,QAAM,mBAAmB,YAAY,KAAK;AAE1C,gBAAc,KAAK,iCAAiC;AAAA,IAClD;AAAA,IACA,UAAU,QAAQ;AAAA,IAClB;AAAA,IACA,UAAU,KAAK;AAAA,IACf,QAAQ,KAAK;AAAA,IACb,gBAAgB,KAAK;AAAA,IACrB,cAAc,KAAK;AAAA,IACnB,WAAW;AAAA,EAAA,CACZ;AAED,SAAO,QAAQ,2CAA2C,YAAY,IAAI;AAAA,IACxE,UAAU;AAAA,IACV;AAAA,EAAA,CACD;AAED,MAAI;AACF,UAAM,SAAS,MAAM,QAAQ,WAAW,EAAE,GAAG,MAAM,OAAO,QAAQ;AAClE,UAAM,WAAW,KAAK,IAAA,IAAQ;AAE9B,kBAAc,KAAK,mCAAmC;AAAA,MACpD;AAAA,MACA,UAAU,QAAQ;AAAA,MAClB;AAAA,MACA,MAAM,OAAO;AAAA,MACb,UAAU,OAAO;AAAA,MACjB;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,WAAW,KAAK,IAAA;AAAA,IAAI,CACrB;AAED,WAAO;AAAA,MACL,yCAAyC,OAAO,KAAK,MAAM;AAAA,MAC3D,EAAE,SAAS,CAAC,CAAC,OAAO,KAAA;AAAA,IAAK;AAG3B,QAAI,OAAO,MAAO,OAAM,mBAAmB,YAAY,OAAO,OAAO,KAAK;AAC1E,UAAM,oBAAoB,YAAY,OAAO;AAAA,MAC3C;AAAA,MACA,OAAO,OAAO;AAAA,IAAA,CACf;AAED,WAAO;AAAA,EACT,SAAS,OAAO;AACd,UAAM,WAAW,KAAK,IAAA,IAAQ;AAC9B,UAAM,MAAM;AACZ,kBAAc,KAAK,+BAA+B;AAAA,MAChD;AAAA,MACA,UAAU,QAAQ;AAAA,MAClB;AAAA,MACA,OAAO,EAAE,SAAS,IAAI,SAAS,MAAM,IAAI,KAAA;AAAA,MACzC;AAAA,MACA,cAAc,KAAK;AAAA,MACnB,WAAW,KAAK,IAAA;AAAA,IAAI,CACrB;AACD,UAAM,mBAAmB,YAAY,OAAO;AAAA,MAC1C;AAAA,MACA;AAAA,IAAA,CACD;AACD,WAAO,OAAO,yCAAyC;AAAA,MACrD;AAAA,MACA,QAAQ;AAAA,IAAA,CACT;AACD,UAAM;AAAA,EACR;AACF;AASO,SAAS,2BAOd,SACiD;AACjD,SAAO;AACT;"}

package/dist/esm/activities/generateVideo/adapter.d.ts

@@ -1,4 +1,33 @@
-import { VideoGenerationOptions, VideoJobResult, VideoStatusResult, VideoUrlResult } from '../../types.js';
+import { ModelInputModalitiesByName, VideoGenerationOptions, VideoJobResult, VideoStatusResult, VideoUrlResult } from '../../types.js';
+/**
+ * Structured description of the durations a video model accepts.
+ *
+ * Tagged union so the same shape can express discrete enums (OpenAI Sora,
+ * Veo), continuous ranges, mixed shapes, and models with no duration field.
+ * Consumed by `VideoAdapter.availableDurations()`.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type DurationOptions<T extends string | number | undefined> = {
+    kind: 'discrete';
+    values: ReadonlyArray<NonNullable<T>>;
+} | {
+    kind: 'range';
+    min: number;
+    max: number;
+    step?: number;
+    unit: 'seconds';
+} | {
+    kind: 'mixed';
+    values: ReadonlyArray<NonNullable<T>>;
+    range?: {
+        min: number;
+        max: number;
+        step?: number;
+    };
+} | {
+    kind: 'none';
+};
 /**
  * Configuration for video adapter instances
  *
@@ -24,8 +53,13 @@ export interface VideoAdapterConfig {
  * - TProviderOptions: Provider-specific options (already resolved)
  * - TModelProviderOptionsByName: Map from model name to its specific provider options
  * - TModelSizeByName: Map from model name to its supported sizes
+ * - TModelInputModalitiesByName: Map from model name to the non-text prompt
+ *   modalities it accepts (constrains the `prompt` part types at compile time)
+ * - TModelDurationByName: Map from model name to its supported duration
+ *   union. Defaults to `Record<string, number>` so adapters that haven't
+ *   declared a map keep today's `duration?: number` typing.
  */
-export interface VideoAdapter<TModel extends string = string, TProviderOptions extends object = Record<string, unknown>, TModelProviderOptionsByName extends Record<string, any> = Record<string, any>, TModelSizeByName extends Record<string, string | undefined> = Record<string, string>> {
+export interface VideoAdapter<TModel extends string = string, TProviderOptions extends object = Record<string, unknown>, TModelProviderOptionsByName extends Record<string, any> = Record<string, any>, TModelSizeByName extends Record<string, string | undefined> = Record<string, string>, TModelInputModalitiesByName extends ModelInputModalitiesByName = ModelInputModalitiesByName, TModelDurationByName extends Record<string, string | number | undefined> = Record<string, number>> {
     /** Discriminator for adapter kind - used to determine API shape */
     readonly kind: 'video';
     /** Adapter name identifier */
@@ -39,12 +73,14 @@ export interface VideoAdapter<TModel extends string = string, TProviderOptions e
         providerOptions: TProviderOptions;
         modelProviderOptionsByName: TModelProviderOptionsByName;
         modelSizeByName: TModelSizeByName;
+        modelInputModalitiesByName: TModelInputModalitiesByName;
+        modelDurationByName: TModelDurationByName;
     };
     /**
      * Create a new video generation job.
      * Returns a job ID that can be used to poll for status and retrieve the video.
      */
-    createVideoJob: (options: VideoGenerationOptions<TProviderOptions, TModelSizeByName[TModel]>) => Promise<VideoJobResult>;
+    createVideoJob: (options: VideoGenerationOptions<TProviderOptions, TModelSizeByName[TModel], TModelDurationByName[TModel]>) => Promise<VideoJobResult>;
     /**
      * Get the current status of a video generation job.
      */
@@ -54,12 +90,23 @@ export interface VideoAdapter<TModel extends string = string, TProviderOptions e
      * Should only be called after status is 'completed'.
      */
     getVideoUrl: (jobId: string) => Promise<VideoUrlResult>;
+    /**
+     * Describe the durations this adapter's model accepts. Returns a tagged
+     * union so consumers can render UI / coerce input without provider-specific
+     * knowledge.
+     */
+    availableDurations: () => DurationOptions<TModelDurationByName[TModel]>;
+    /**
+     * Coerce a raw seconds value to the closest valid duration for this model.
+     * Returns `undefined` for models with no duration field.
+     */
+    snapDuration: (seconds: number) => TModelDurationByName[TModel] | undefined;
 }
 /**
  * A VideoAdapter with any/unknown type parameters.
  * Useful as a constraint in generic functions and interfaces.
  */
-export type AnyVideoAdapter = VideoAdapter<any, any, any, any>;
+export type AnyVideoAdapter = VideoAdapter<any, any, any, any, any, any>;
 /**
  * Abstract base class for video generation adapters.
  * Extend this class to implement a video adapter for a specific provider.
@@ -68,7 +115,7 @@ export type AnyVideoAdapter = VideoAdapter<any, any, any, any>;
  *
  * Generic parameters match VideoAdapter - all pre-resolved by the provider function.
  */
-export declare abstract class BaseVideoAdapter<TModel extends string = string, TProviderOptions extends object = Record<string, unknown>, TModelProviderOptionsByName extends Record<string, any> = Record<string, any>, TModelSizeByName extends Record<string, string | undefined> = Record<string, string>> implements VideoAdapter<TModel, TProviderOptions, TModelProviderOptionsByName, TModelSizeByName> {
+export declare abstract class BaseVideoAdapter<TModel extends string = string, TProviderOptions extends object = Record<string, unknown>, TModelProviderOptionsByName extends Record<string, any> = Record<string, any>, TModelSizeByName extends Record<string, string | undefined> = Record<string, string>, TModelInputModalitiesByName extends ModelInputModalitiesByName = ModelInputModalitiesByName, TModelDurationByName extends Record<string, string | number | undefined> = Record<string, number>> implements VideoAdapter<TModel, TProviderOptions, TModelProviderOptionsByName, TModelSizeByName, TModelInputModalitiesByName, TModelDurationByName> {
     readonly kind: "video";
     abstract readonly name: string;
     readonly model: TModel;
@@ -76,11 +123,23 @@ export declare abstract class BaseVideoAdapter<TModel extends string = string, T
         providerOptions: TProviderOptions;
         modelProviderOptionsByName: TModelProviderOptionsByName;
         modelSizeByName: TModelSizeByName;
+        modelInputModalitiesByName: TModelInputModalitiesByName;
+        modelDurationByName: TModelDurationByName;
     };
     protected config: VideoAdapterConfig;
     constructor(config: VideoAdapterConfig | undefined, model: TModel);
-    abstract createVideoJob(options: VideoGenerationOptions<TProviderOptions, TModelSizeByName[TModel]>): Promise<VideoJobResult>;
+    abstract createVideoJob(options: VideoGenerationOptions<TProviderOptions, TModelSizeByName[TModel], TModelDurationByName[TModel]>): Promise<VideoJobResult>;
     abstract getVideoStatus(jobId: string): Promise<VideoStatusResult>;
     abstract getVideoUrl(jobId: string): Promise<VideoUrlResult>;
+    /**
+     * Default implementation returns `{ kind: 'none' }`. Adapters that have
+     * declared their per-model duration map should override this.
+     */
+    availableDurations(): DurationOptions<TModelDurationByName[TModel]>;
+    /**
+     * Default implementation returns `undefined`. Adapters that have declared
+     * their per-model duration map should override.
+     */
+    snapDuration(_seconds: number): TModelDurationByName[TModel] | undefined;
     protected generateId(): string;
 }

package/dist/esm/activities/generateVideo/adapter.js

@@ -6,6 +6,20 @@ class BaseVideoAdapter {
     this.config = config;
     this.model = model;
   }
+  /**
+   * Default implementation returns `{ kind: 'none' }`. Adapters that have
+   * declared their per-model duration map should override this.
+   */
+  availableDurations() {
+    return { kind: "none" };
+  }
+  /**
+   * Default implementation returns `undefined`. Adapters that have declared
+   * their per-model duration map should override.
+   */
+  snapDuration(_seconds) {
+    return void 0;
+  }
   generateId() {
     return `${this.name}-${Date.now()}-${Math.random().toString(36).substring(7)}`;
   }

package/dist/esm/activities/generateVideo/adapter.js.map

@@ -1 +1 @@
-{"version":3,"file":"adapter.js","sources":["../../../../src/activities/generateVideo/adapter.ts"],"sourcesContent":["import type {\n  VideoGenerationOptions,\n  VideoJobResult,\n  VideoStatusResult,\n  VideoUrlResult,\n} from '../../types'\n\n/**\n * Configuration for video adapter instances\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport interface VideoAdapterConfig {\n  apiKey?: string\n  baseUrl?: string\n  timeout?: number\n  maxRetries?: number\n  headers?: Record<string, string>\n}\n\n/**\n * Video adapter interface with pre-resolved generics.\n *\n * An adapter is created by a provider function: `provider('model')` → `adapter`\n * All type resolution happens at the provider call site, not in this interface.\n *\n * @experimental Video generation is an experimental feature and may change.\n *\n * Generic parameters:\n * - TModel: The specific model name (e.g., 'sora-2')\n * - TProviderOptions: Provider-specific options (already resolved)\n * - TModelProviderOptionsByName: Map from model name to its specific provider options\n * - TModelSizeByName: Map from model name to its supported sizes\n */\nexport interface VideoAdapter<\n  TModel extends string = string,\n  TProviderOptions extends object = Record<string, unknown>,\n  TModelProviderOptionsByName extends Record<string, any> = Record<string, any>,\n  TModelSizeByName extends Record<string, string | undefined> = Record<\n    string,\n    string\n  >,\n> {\n  /** Discriminator for adapter kind - used to determine API shape */\n  readonly kind: 'video'\n  /** Adapter name identifier */\n  readonly name: string\n  /** The model this adapter is configured for */\n  readonly model: TModel\n\n  /**\n   * @internal Type-only properties for inference. Not assigned at runtime.\n   */\n  '~types': {\n    providerOptions: TProviderOptions\n    modelProviderOptionsByName: TModelProviderOptionsByName\n    modelSizeByName: TModelSizeByName\n  }\n\n  /**\n   * Create a new video generation job.\n   * Returns a job ID that can be used to poll for status and retrieve the video.\n   */\n  createVideoJob: (\n    options: VideoGenerationOptions<TProviderOptions, TModelSizeByName[TModel]>,\n  ) => Promise<VideoJobResult>\n\n  /**\n   * Get the current status of a video generation job.\n   */\n  getVideoStatus: (jobId: string) => Promise<VideoStatusResult>\n\n  /**\n   * Get the URL to download/view the generated video.\n   * Should only be called after status is 'completed'.\n   */\n  getVideoUrl: (jobId: string) => Promise<VideoUrlResult>\n}\n\n/**\n * A VideoAdapter with any/unknown type parameters.\n * Useful as a constraint in generic functions and interfaces.\n */\nexport type AnyVideoAdapter = VideoAdapter<any, any, any, any>\n\n/**\n * Abstract base class for video generation adapters.\n * Extend this class to implement a video adapter for a specific provider.\n *\n * @experimental Video generation is an experimental feature and may change.\n *\n * Generic parameters match VideoAdapter - all pre-resolved by the provider function.\n */\nexport abstract class BaseVideoAdapter<\n  TModel extends string = string,\n  TProviderOptions extends object = Record<string, unknown>,\n  TModelProviderOptionsByName extends Record<string, any> = Record<string, any>,\n  TModelSizeByName extends Record<string, string | undefined> = Record<\n    string,\n    string\n  >,\n> implements VideoAdapter<\n  TModel,\n  TProviderOptions,\n  TModelProviderOptionsByName,\n  TModelSizeByName\n> {\n  readonly kind = 'video' as const\n  abstract readonly name: string\n  readonly model: TModel\n\n  // Type-only property - never assigned at runtime\n  declare '~types': {\n    providerOptions: TProviderOptions\n    modelProviderOptionsByName: TModelProviderOptionsByName\n    modelSizeByName: TModelSizeByName\n  }\n\n  protected config: VideoAdapterConfig\n\n  constructor(config: VideoAdapterConfig = {}, model: TModel) {\n    this.config = config\n    this.model = model\n  }\n\n  abstract createVideoJob(\n    options: VideoGenerationOptions<TProviderOptions, TModelSizeByName[TModel]>,\n  ): Promise<VideoJobResult>\n\n  abstract getVideoStatus(jobId: string): Promise<VideoStatusResult>\n\n  abstract getVideoUrl(jobId: string): Promise<VideoUrlResult>\n\n  protected generateId(): string {\n    return `${this.name}-${Date.now()}-${Math.random().toString(36).substring(7)}`\n  }\n}\n"],"names":[],"mappings":"AA6FO,MAAe,iBAapB;AAAA,EACS,OAAO;AAAA,EAEP;AAAA,EASC;AAAA,EAEV,YAAY,SAA6B,CAAA,GAAI,OAAe;AAC1D,SAAK,SAAS;AACd,SAAK,QAAQ;AAAA,EACf;AAAA,EAUU,aAAqB;AAC7B,WAAO,GAAG,KAAK,IAAI,IAAI,KAAK,KAAK,IAAI,KAAK,OAAA,EAAS,SAAS,EAAE,EAAE,UAAU,CAAC,CAAC;AAAA,EAC9E;AACF;"}
+{"version":3,"file":"adapter.js","sources":["../../../../src/activities/generateVideo/adapter.ts"],"sourcesContent":["import type {\n  ModelInputModalitiesByName,\n  VideoGenerationOptions,\n  VideoJobResult,\n  VideoStatusResult,\n  VideoUrlResult,\n} from '../../types'\n\n/**\n * Structured description of the durations a video model accepts.\n *\n * Tagged union so the same shape can express discrete enums (OpenAI Sora,\n * Veo), continuous ranges, mixed shapes, and models with no duration field.\n * Consumed by `VideoAdapter.availableDurations()`.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type DurationOptions<T extends string | number | undefined> =\n  | { kind: 'discrete'; values: ReadonlyArray<NonNullable<T>> }\n  | { kind: 'range'; min: number; max: number; step?: number; unit: 'seconds' }\n  | {\n      kind: 'mixed'\n      values: ReadonlyArray<NonNullable<T>>\n      range?: { min: number; max: number; step?: number }\n    }\n  | { kind: 'none' }\n\n/**\n * Configuration for video adapter instances\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport interface VideoAdapterConfig {\n  apiKey?: string\n  baseUrl?: string\n  timeout?: number\n  maxRetries?: number\n  headers?: Record<string, string>\n}\n\n/**\n * Video adapter interface with pre-resolved generics.\n *\n * An adapter is created by a provider function: `provider('model')` → `adapter`\n * All type resolution happens at the provider call site, not in this interface.\n *\n * @experimental Video generation is an experimental feature and may change.\n *\n * Generic parameters:\n * - TModel: The specific model name (e.g., 'sora-2')\n * - TProviderOptions: Provider-specific options (already resolved)\n * - TModelProviderOptionsByName: Map from model name to its specific provider options\n * - TModelSizeByName: Map from model name to its supported sizes\n * - TModelInputModalitiesByName: Map from model name to the non-text prompt\n *   modalities it accepts (constrains the `prompt` part types at compile time)\n * - TModelDurationByName: Map from model name to its supported duration\n *   union. Defaults to `Record<string, number>` so adapters that haven't\n *   declared a map keep today's `duration?: number` typing.\n */\nexport interface VideoAdapter<\n  TModel extends string = string,\n  TProviderOptions extends object = Record<string, unknown>,\n  TModelProviderOptionsByName extends Record<string, any> = Record<string, any>,\n  TModelSizeByName extends Record<string, string | undefined> = Record<\n    string,\n    string\n  >,\n  TModelInputModalitiesByName extends ModelInputModalitiesByName =\n    ModelInputModalitiesByName,\n  TModelDurationByName extends Record<string, string | number | undefined> =\n    Record<string, number>,\n> {\n  /** Discriminator for adapter kind - used to determine API shape */\n  readonly kind: 'video'\n  /** Adapter name identifier */\n  readonly name: string\n  /** The model this adapter is configured for */\n  readonly model: TModel\n\n  /**\n   * @internal Type-only properties for inference. Not assigned at runtime.\n   */\n  '~types': {\n    providerOptions: TProviderOptions\n    modelProviderOptionsByName: TModelProviderOptionsByName\n    modelSizeByName: TModelSizeByName\n    modelInputModalitiesByName: TModelInputModalitiesByName\n    modelDurationByName: TModelDurationByName\n  }\n\n  /**\n   * Create a new video generation job.\n   * Returns a job ID that can be used to poll for status and retrieve the video.\n   */\n  createVideoJob: (\n    options: VideoGenerationOptions<\n      TProviderOptions,\n      TModelSizeByName[TModel],\n      TModelDurationByName[TModel]\n    >,\n  ) => Promise<VideoJobResult>\n\n  /**\n   * Get the current status of a video generation job.\n   */\n  getVideoStatus: (jobId: string) => Promise<VideoStatusResult>\n\n  /**\n   * Get the URL to download/view the generated video.\n   * Should only be called after status is 'completed'.\n   */\n  getVideoUrl: (jobId: string) => Promise<VideoUrlResult>\n\n  /**\n   * Describe the durations this adapter's model accepts. Returns a tagged\n   * union so consumers can render UI / coerce input without provider-specific\n   * knowledge.\n   */\n  availableDurations: () => DurationOptions<TModelDurationByName[TModel]>\n\n  /**\n   * Coerce a raw seconds value to the closest valid duration for this model.\n   * Returns `undefined` for models with no duration field.\n   */\n  snapDuration: (seconds: number) => TModelDurationByName[TModel] | undefined\n}\n\n/**\n * A VideoAdapter with any/unknown type parameters.\n * Useful as a constraint in generic functions and interfaces.\n */\nexport type AnyVideoAdapter = VideoAdapter<any, any, any, any, any, any>\n\n/**\n * Abstract base class for video generation adapters.\n * Extend this class to implement a video adapter for a specific provider.\n *\n * @experimental Video generation is an experimental feature and may change.\n *\n * Generic parameters match VideoAdapter - all pre-resolved by the provider function.\n */\nexport abstract class BaseVideoAdapter<\n  TModel extends string = string,\n  TProviderOptions extends object = Record<string, unknown>,\n  TModelProviderOptionsByName extends Record<string, any> = Record<string, any>,\n  TModelSizeByName extends Record<string, string | undefined> = Record<\n    string,\n    string\n  >,\n  TModelInputModalitiesByName extends ModelInputModalitiesByName =\n    ModelInputModalitiesByName,\n  TModelDurationByName extends Record<string, string | number | undefined> =\n    Record<string, number>,\n> implements VideoAdapter<\n  TModel,\n  TProviderOptions,\n  TModelProviderOptionsByName,\n  TModelSizeByName,\n  TModelInputModalitiesByName,\n  TModelDurationByName\n> {\n  readonly kind = 'video' as const\n  abstract readonly name: string\n  readonly model: TModel\n\n  // Type-only property - never assigned at runtime\n  declare '~types': {\n    providerOptions: TProviderOptions\n    modelProviderOptionsByName: TModelProviderOptionsByName\n    modelSizeByName: TModelSizeByName\n    modelInputModalitiesByName: TModelInputModalitiesByName\n    modelDurationByName: TModelDurationByName\n  }\n\n  protected config: VideoAdapterConfig\n\n  constructor(config: VideoAdapterConfig = {}, model: TModel) {\n    this.config = config\n    this.model = model\n  }\n\n  abstract createVideoJob(\n    options: VideoGenerationOptions<\n      TProviderOptions,\n      TModelSizeByName[TModel],\n      TModelDurationByName[TModel]\n    >,\n  ): Promise<VideoJobResult>\n\n  abstract getVideoStatus(jobId: string): Promise<VideoStatusResult>\n\n  abstract getVideoUrl(jobId: string): Promise<VideoUrlResult>\n\n  /**\n   * Default implementation returns `{ kind: 'none' }`. Adapters that have\n   * declared their per-model duration map should override this.\n   */\n  availableDurations(): DurationOptions<TModelDurationByName[TModel]> {\n    return { kind: 'none' }\n  }\n\n  /**\n   * Default implementation returns `undefined`. Adapters that have declared\n   * their per-model duration map should override.\n   */\n  snapDuration(_seconds: number): TModelDurationByName[TModel] | undefined {\n    return undefined\n  }\n\n  protected generateId(): string {\n    return `${this.name}-${Date.now()}-${Math.random().toString(36).substring(7)}`\n  }\n}\n"],"names":[],"mappings":"AA6IO,MAAe,iBAmBpB;AAAA,EACS,OAAO;AAAA,EAEP;AAAA,EAWC;AAAA,EAEV,YAAY,SAA6B,CAAA,GAAI,OAAe;AAC1D,SAAK,SAAS;AACd,SAAK,QAAQ;AAAA,EACf;AAAA;AAAA;AAAA;AAAA;AAAA,EAkBA,qBAAoE;AAClE,WAAO,EAAE,MAAM,OAAA;AAAA,EACjB;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,aAAa,UAA4D;AACvE,WAAO;AAAA,EACT;AAAA,EAEU,aAAqB;AAC7B,WAAO,GAAG,KAAK,IAAI,IAAI,KAAK,KAAK,IAAI,KAAK,OAAA,EAAS,SAAS,EAAE,EAAE,UAAU,CAAC,CAAC;AAAA,EAC9E;AACF;"}

package/dist/esm/activities/generateVideo/index.d.ts

@@ -1,6 +1,7 @@
 import { DebugOption } from '../../logger/types.js';
+import { GenerationMiddleware } from '../middleware.js';
 import { VideoAdapter } from './adapter.js';
-import { StreamChunk, TokenUsage, VideoJobResult, VideoStatusResult, VideoUrlResult } from '../../types.js';
+import { MediaPrompt, MediaPromptFor, StreamChunk, TokenUsage, VideoJobResult, VideoStatusResult, VideoUrlResult } from '../../types.js';
 /** The adapter kind this activity handles */
 export declare const kind: "video";
 /**
@@ -11,6 +12,19 @@ export type VideoProviderOptions<TAdapter> = TAdapter extends VideoAdapter<any,
  * Extract the size type for a VideoAdapter's model via ~types.
  */
 export type VideoSizeForAdapter<TAdapter> = TAdapter extends VideoAdapter<infer TModel, any, any, infer TSizeMap> ? TModel extends keyof TSizeMap ? TSizeMap[TModel] : string : string;
+/**
+ * Extract the prompt type a model accepts from a VideoAdapter via ~types.
+ * Mirrors `ImagePromptForModel`: models in the adapter's input-modality map
+ * get a `prompt` narrowed to text + their supported part types; adapters
+ * without a map fall back to the full MediaPrompt.
+ */
+export type VideoPromptForAdapter<TAdapter> = TAdapter extends VideoAdapter<infer TModel, any, any, any, infer ModsByName> ? string extends keyof ModsByName ? MediaPrompt : TModel extends keyof ModsByName ? MediaPromptFor<ModsByName[TModel][number]> : MediaPrompt : MediaPrompt;
+/**
+ * Extract the duration type for a VideoAdapter's model via ~types.
+ * Mirrors `VideoSizeForAdapter`. Falls back to `number` for adapters that
+ * haven't declared per-model duration constraints.
+ */
+export type VideoDurationForAdapter<TAdapter> = TAdapter extends VideoAdapter<infer TModel, any, any, any, any, infer TDurationMap> ? TModel extends keyof TDurationMap ? TDurationMap[TModel] : number : number;
 /**
  * Base options shared by all video activity operations.
  * The model is extracted from the adapter's model property.
@@ -33,12 +47,25 @@ interface VideoActivityBaseOptions<TAdapter extends VideoAdapter<string, any, an
 export type VideoCreateOptions<TAdapter extends VideoAdapter<string, any, any, any>, TStream extends boolean = false> = VideoActivityBaseOptions<TAdapter> & {
     /** Request type - create a new job (default if not specified) */
     request?: 'create';
-    /** Text description of the desired video */
-    prompt: string;
+    /**
+     * Description of the desired video. Either a plain string, or — for models
+     * that support image-conditioned generation — an ordered array of content
+     * parts interleaving text with image inputs. Image parts may carry
+     * `metadata.role` (`'start_frame' | 'end_frame' | 'reference' |
+     * 'character'`) to disambiguate intent; positional fallback otherwise. The
+     * accepted part types are narrowed per model via the adapter's
+     * input-modality map.
+     */
+    prompt: VideoPromptForAdapter<TAdapter>;
     /** Video size — format depends on the provider (e.g., "16:9", "1280x720") */
     size?: VideoSizeForAdapter<TAdapter>;
-    /** Video duration in seconds */
-    duration?: number;
+    /**
+     * Video duration in seconds. Adapters that declare a per-model duration
+     * map narrow this to the model's valid union (e.g. `4 | 6 | 8` for Veo 3).
+     * Pass `adapter.snapDuration(seconds)` to coerce raw seconds to a valid
+     * value.
+     */
+    duration?: VideoDurationForAdapter<TAdapter>;
     /**
      * Whether to stream the video generation lifecycle.
      * When true, returns an AsyncIterable<StreamChunk> that handles the full
@@ -60,6 +87,14 @@ export type VideoCreateOptions<TAdapter extends VideoAdapter<string, any, any, a
      * control and/or a custom `Logger`.
      */
     debug?: DebugOption;
+    /**
+     * Observe-only middleware notified on start, usage, success, and error. Pass
+     * `otelMiddleware()` to emit OpenTelemetry spans, or implement the
+     * `GenerationMiddleware` contract for a custom backend. In streaming mode the
+     * span covers the full create→poll→complete lifecycle; in non-streaming mode
+     * it covers job submission. An abandoned stream fires `onAbort`.
+     */
+    middleware?: Array<GenerationMiddleware>;
 } & ({} extends VideoProviderOptions<TAdapter> ? {
     /** Provider-specific options for video generation */ modelOptions?: VideoProviderOptions<TAdapter>;
 } : {

package/dist/esm/activities/generateVideo/index.js

@@ -1,6 +1,7 @@
 import { aiEventClient } from "@tanstack/ai-event-client";
 import { toRunErrorPayload } from "../error-payload.js";
 import { resolveDebugOption } from "../../logger/resolve.js";
+import { createGenerationContext, runGenerationStart, runGenerationUsage, runGenerationFinish, runGenerationError, runGenerationAbort } from "../middleware/run.js";
 const kind = "video";
 function createId(prefix) {
   return `${prefix}-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`;
@@ -14,10 +15,21 @@ function generateVideo(options) {
   return runCreateVideoJob(options);
 }
 async function runCreateVideoJob(options) {
-  const { adapter, prompt, size, duration, modelOptions } = options;
+  const { adapter, prompt, size, duration, modelOptions, middleware } = options;
   const model = adapter.model;
+  const requestId = createId("video");
+  const startTime = Date.now();
   const logger = resolveDebugOption(options.debug);
   const providerName = adapter.provider ?? adapter.name ?? "unknown";
+  const mwCtx = createGenerationContext({
+    requestId,
+    activity: "video",
+    provider: adapter.name,
+    model,
+    modelOptions,
+    createId
+  });
+  await runGenerationStart(middleware, mwCtx);
   logger.request(`activity=generateVideo provider=${providerName}`, {
     provider: providerName,
     model
@@ -35,8 +47,15 @@ async function runCreateVideoJob(options) {
       jobId: result.jobId,
       model: result.model
     });
+    await runGenerationFinish(middleware, mwCtx, {
+      duration: Date.now() - startTime
+    });
     return result;
   } catch (error) {
+    await runGenerationError(middleware, mwCtx, {
+      error,
+      duration: Date.now() - startTime
+    });
     logger.errors("generateVideo activity failed", {
       error,
       source: "generateVideo"
@@ -48,9 +67,11 @@ function sleep(ms) {
   return new Promise((resolve) => setTimeout(resolve, ms));
 }
 async function* runStreamingVideoGeneration(options) {
-  const { adapter, prompt, size, duration, modelOptions } = options;
+  const { adapter, prompt, size, duration, modelOptions, middleware } = options;
   const model = adapter.model;
   const runId = options.runId ?? createId("run");
+  const requestId = createId("video");
+  const obsStartTime = Date.now();
   const pollingInterval = options.pollingInterval ?? 2e3;
   const maxDuration = options.maxDuration ?? 6e5;
   const logger = resolveDebugOption(options.debug);
@@ -62,6 +83,15 @@ async function* runStreamingVideoGeneration(options) {
     threadId,
     timestamp: Date.now()
   };
+  const mwCtx = createGenerationContext({
+    requestId,
+    activity: "video",
+    provider: adapter.name,
+    model,
+    modelOptions,
+    createId
+  });
+  await runGenerationStart(middleware, mwCtx);
   logger.request(
     `activity=generateVideo provider=${providerName} stream=true`,
     {
@@ -69,6 +99,7 @@ async function* runStreamingVideoGeneration(options) {
       model
     }
   );
+  let settled = false;
   try {
     const jobResult = await adapter.createVideoJob({
       model,
@@ -108,6 +139,13 @@ async function* runStreamingVideoGeneration(options) {
             url: urlResult.url
           }
         );
+        if (urlResult.usage)
+          await runGenerationUsage(middleware, mwCtx, urlResult.usage);
+        await runGenerationFinish(middleware, mwCtx, {
+          duration: Date.now() - obsStartTime,
+          usage: urlResult.usage
+        });
+        settled = true;
         yield {
           type: "CUSTOM",
           name: "generation:result",
@@ -136,6 +174,11 @@ async function* runStreamingVideoGeneration(options) {
     throw new Error("Video generation timed out");
   } catch (error) {
     const payload = toRunErrorPayload(error, "Video generation failed");
+    settled = true;
+    await runGenerationError(middleware, mwCtx, {
+      error,
+      duration: Date.now() - obsStartTime
+    });
     logger.errors("generateVideo activity failed", {
       message: payload.message,
       code: payload.code,
@@ -150,6 +193,13 @@ async function* runStreamingVideoGeneration(options) {
       error: payload,
       timestamp: Date.now()
     };
+  } finally {
+    if (!settled) {
+      await runGenerationAbort(middleware, mwCtx, {
+        reason: "Video generation stream abandoned before completion",
+        duration: Date.now() - obsStartTime
+      });
+    }
   }
 }
 async function getVideoJobStatus(options) {