cloudflare/ai workers-ai-provider@3.1.0

on GitHub

Minor Changes

• #389 8538cd5 Thanks @vaibhavshn! - Add transcription, text-to-speech, and reranking support to the Workers AI provider.

  New capabilities

  • Transcription (provider.transcription(model)) — implements TranscriptionModelV3. Supports Whisper models (@cf/openai/whisper, whisper-tiny-en, whisper-large-v3-turbo) and Deepgram Nova-3 (@cf/deepgram/nova-3). Handles model-specific input formats: number arrays for basic Whisper, base64 for v3-turbo via REST, and { body, contentType } for Nova-3 via binding or raw binary upload for Nova-3 via REST.

  • Speech / TTS (provider.speech(model)) — implements SpeechModelV3. Supports Workers AI TTS models including Deepgram Aura-1 (@cf/deepgram/aura-1). Accepts text, voice, and speed options. Returns audio as Uint8Array. Uses returnRawResponse to handle binary audio from the REST path without JSON parsing.

  • Reranking (provider.reranking(model)) — implements RerankingModelV3. Supports BGE reranker models (@cf/baai/bge-reranker-base, bge-reranker-v2-m3). Converts AI SDK's document format to Workers AI's { query, contexts, top_k } input. Handles both text and JSON object documents.

  Bug fixes

  • AbortSignal passthrough — createRun REST shim now passes the abort signal to fetch, enabling request cancellation and timeout handling. Previously the signal was silently dropped.
  • Nova-3 REST support — Added createRunBinary utility for models that require raw binary upload instead of JSON (used by Nova-3 transcription via REST).

  Usage

  import { createWorkersAI } from "workers-ai-provider";
  import {
    experimental_transcribe,
    experimental_generateSpeech,
    rerank,
  } from "ai";
  
  const workersai = createWorkersAI({ binding: env.AI });
  
  // Transcription
  const transcript = await experimental_transcribe({
    model: workersai.transcription("@cf/openai/whisper-large-v3-turbo"),
    audio: audioData,
    mediaType: "audio/wav",
  });
  
  // Speech
  const speech = await experimental_generateSpeech({
    model: workersai.speech("@cf/deepgram/aura-1"),
    text: "Hello world",
    voice: "asteria",
  });
  
  // Reranking
  const ranked = await rerank({
    model: workersai.reranking("@cf/baai/bge-reranker-base"),
    query: "What is machine learning?",
    documents: ["ML is a branch of AI.", "The weather is sunny."],
  });