@economic/agents-providers

Provider helpers for creating AI SDK language models behind Cloudflare AI Gateway.

The package is meant for server-side code, especially Cloudflare Workers using @economic/agents. It keeps gateway configuration in one place while still giving each provider its own typed model names.

Install

npm install @economic/agents-providers

Basic Usage

Create the providers object once from your Worker bindings, then use it anywhere you need an AI SDK language model.

import { createAiGatewayVertexProviders } from "@economic/agents-providers";
import { env } from "cloudflare:workers";

export const providers = createAiGatewayVertexProviders({
  cloudflareAccountId: env.CLOUDFLARE_ACCOUNT_ID,
  cloudflareAiGatewayId: env.CLOUDFLARE_AI_GATEWAY_ID,
  cloudflareApiToken: env.CLOUDFLARE_API_TOKEN,
  googleCloudProjectId: env.GOOGLE_CLOUD_PROJECT_ID,
  googleApiKey: env.GOOGLE_GENERATIVE_AI_API_KEY,
});

Use the provider in an agent:

import { ChatAgent } from "@economic/agents";
import type { LanguageModel } from "ai";
import { providers } from "./providers";

export class SupportAgent extends ChatAgent {
  getModel(): LanguageModel {
    return providers.gemini("gemini-2.5-flash");
  }

  getFastModel(): LanguageModel {
    return providers.anthropic("claude-haiku-4-5");
  }
}

providers.anthropic(...) is typed from @ai-sdk/anthropic, and providers.gemini(...) is typed from @ai-sdk/google, so model names follow the SDKs instead of hand-maintained lists in this package.

Configuration

createAiGatewayVertexProviders accepts the shared Cloudflare AI Gateway settings plus the Google Vertex settings needed for Anthropic:

interface AiGatewayVertexProvidersOptions {
  cloudflareAccountId: string;
  cloudflareAiGatewayId: string;
  cloudflareApiToken: string;
  googleCloudProjectId: string;
  googleApiKey?: string;
  location?: string;
  anthropicVersion?: string;
}

Defaults:

• location: "europe-west1"
• anthropicVersion: "vertex-2023-10-16"

Required values should usually come from Worker bindings rather than being read inside the library. That keeps this package testable and usable in different runtimes.

googleApiKey is passed to @ai-sdk/google for Gemini. If your Cloudflare AI Gateway is configured to authenticate upstream requests itself, the library sends "unused" to satisfy the Google SDK's required API-key header.

Exports

import {
  createAiGatewayVertexProviders,
  createAnthropicVertexProvider,
  createGeminiProvider,
  type AiGatewayVertexProvidersOptions,
  type AnthropicVertexModelId,
  type AnthropicVertexProviderOptions,
  type GeminiModelId,
  type GeminiProviderOptions,
} from "@economic/agents-providers";

Use createAiGatewayVertexProviders for normal app setup. Use createAnthropicVertexProvider or createGeminiProvider only if you want one provider by itself.

Current Providers

• anthropic(modelId) routes Anthropic Messages models through Cloudflare AI Gateway to Google Vertex AI.
• gemini(modelId) routes Gemini models through Cloudflare AI Gateway to Google AI Studio.

More providers can be added to the same factory without changing the way callers pass shared Cloudflare/Vertex configuration.