Skip to main content

Cloudflare Workers AI Provider Guide

Open-model inference at the edge via Cloudflare Workers AI

Overview​

Cloudflare Workers AI serves Meta Llama, Mistral, and other open models from Cloudflare's global GPU cluster. NeuroLink talks to the OpenAI-compatible endpoint.

Key Facts​

  • Protocol: OpenAI-compatible (/v1/chat/completions)
  • Default base URL: https://api.cloudflare.com/client/v4/accounts/{accountId}/ai/v1
  • Default model: @cf/meta/llama-3.3-70b-instruct-fp8-fast
  • Streaming: Yes
  • Tool calling: Limited (model-dependent)

Quick Start​

1. Get Credentials​

You need both:

  • A Cloudflare Account ID (Cloudflare dashboard → right sidebar)
  • A Workers AI API token with the Workers AI Read & Write permission (Profile → API Tokens → Create Token)

2. Configure​

CLOUDFLARE_ACCOUNT_ID=your-account-id
CLOUDFLARE_API_KEY=your-workers-ai-token
CLOUDFLARE_MODEL=@cf/meta/llama-3.3-70b-instruct-fp8-fast

3. Generate​

import { NeuroLink } from "@juspay/neurolink";
const ai = new NeuroLink();
const result = await ai.generate({
  provider: "cloudflare",
  input: { text: "Why is Cloudflare's edge network significant?" },
});
console.log(result.content);

Supported Models (sample)​

Model IDNotes
@cf/meta/llama-3.3-70b-instruct-fp8-fastDefault
@cf/meta/llama-3.1-70b-instructLlama 3.1 70B
@cf/meta/llama-3.1-8b-instructFast tier
@cf/meta/llama-3.2-11b-vision-instructVision-capable

Browse: https://developers.cloudflare.com/workers-ai/models

CLI Usage​

pnpm run cli generate "..." --provider cloudflare

Provider Aliases​

AliasExample
cloudflare--provider cloudflare
cf--provider cf

Configuration Reference​

Environment VariableRequiredDefault
CLOUDFLARE_ACCOUNT_IDYes—
CLOUDFLARE_API_KEYYes—
CLOUDFLARE_MODELNo@cf/meta/llama-3.3-70b-instruct-fp8-fast

Also configures decisions​

CLOUDFLARE_API_KEY and CLOUDFLARE_ACCOUNT_ID also configure a separate provider, the Cloudflare Clef provider (cloudflare-clef), which serves decide() and emits no text. It is the last decision provider NeuroLink falls back to: built-in features that call decide() use the first one configured, in the order TypeSafe, Laya, XOR, Perplexity, then Clef. So a host that set these two variables only for this text provider, and has no other decision provider configured, has Clef as its default decision provider, and no switch turns that off while they are set.

credentials.cloudflare does not configure decide; Clef reads its own slice, credentials.cloudflareClef. The text provider itself is unchanged: it still serves generate() and stream() with the same variables, models and base URL. See One token, two providers.

See Also​