JA EN
shiichan

One createAI to Rule Them All: workers-ai-provider and ai-gateway-provider Retire

Cloudflare Changelog ★★★★ View Markdown
#cloudflare#ai#serverless

Hi everyone, Shii-chan here! The way you call AI models from the Agents SDK just got a lot tidier, so let's go through it.

Cloudflare Changelog developers.cloudflare.com

What was announced?

The Cloudflare Changelog announced a new createAI function in the Agents SDK for the AI SDK and Pi AI. It unifies what workers-ai-provider and ai-gateway-provider used to do.

The story so far

Until now the job was split across packages:

  • workers-ai-provider for Workers AI models
  • ai-gateway-provider for AI Gateway

Both are now feature frozen and deprecated. You can migrate at your convenience.

What changes

There is now a single entry point, createAI, so switching models and configuring fallbacks gets simpler. You can use everything from Workers AI models to other vendors' models in the same way.

Dive Deep

Install the agents package. The example from the changelog looks like this:

import { createAI } from "agents/models/ai-sdk";
import { streamText } from "ai";

export default {
  async fetch(request, env) {
    const ai = createAI({ binding: env.AI, id: "my-gateway" });
    const result = streamText({
      model: ai("anthropic/claude-opus-5.5", {
        fallback: ["@cf/zai-org/glm-4.7-flash"],
      }),
      prompt: "Write a haiku about Durable Objects",
    });
    return result.toTextStreamResponse();
  },
};

With createAI you get:

  • Every model by id: Workers AI @cf/ ids, any model in the AI Gateway catalog, and dynamic routes as dynamic/{route}
  • Vendor model objects: pass a model from the vendor's own package and it is sent unchanged to AI Gateway's provider-native endpoint
  • Flexible connection: use the binding, or an account id and API token to reach a gateway in another account or call from outside a Worker
  • Gateway options: caching, logging, metadata, retries, BYOK alias, zero data retention, and vendor key settings
  • Fallbacks: specify a list of fallback models
  • Stream resume: with the binding, a catalog or dynamic route stream that drops mid-answer can resume
  • Other model types: embeddings, images, speech, transcription, and reranking in the AI SDK provider, and images in Pi AI

Migration at a glance:

  • createWorkersAI({ binding }) becomes createAI({ binding })
  • createWorkersAI({ accountId, apiKey }) becomes createAI({ accountId, apiToken }) (the credential option is renamed to apiToken)
  • createAiGateway(...) becomes createAI(...), with fallbacks given as a list
  • createAISearch from workers-ai-provider is replaced by createAISearchNamespace from ai-search-provider

Wrap-up

  • createAI in the Agents SDK unifies Workers AI, AI Gateway, dynamic routes, and vendor models
  • workers-ai-provider and ai-gateway-provider are feature frozen and deprecated
  • Fallbacks, stream resume, and Gateway options come along too
  • Migration is mostly renaming the function and a few options (apiKey to apiToken)

If you juggle several models or gateways on Workers, this one's for you!

Related posts