Run your own AI provider

Deploy an OpenAI-compatible AI endpoint on Cloudflare with cloudflare-ai-deployer and point Notebooker at it.

Notebooker can use any OpenAI-compatible endpoint as an AI provider. If you’d like to run the models yourself — for cost control, data residency, or just to tinker — the open-source cloudflare-ai-deployer project deploys one to your own Cloudflare account in a single command.

This is the same kind of gateway Notebooker runs in production, made public so you can run your own.

Get cloudflare-ai-deployer on GitHub →

What it deploys

A Cloudflare Worker that exposes an OpenAI-compatible REST API backed by Cloudflare Workers AI:

GET  /v1/models               list configured models
POST /v1/chat/completions     chat / language model
POST /v1/embeddings           embeddings
POST /v1/audio/transcriptions speech-to-text
POST /v1/audio/speech         text-to-speech

Each endpoint is enabled only if you’ve configured a model for it.

Deploy it

You’ll need: Node.js 18+, a Cloudflare account (the free plan is enough), your Account ID, and an API token with Workers Scripts: Edit, Workers AI: Read, and User Details: Read permissions.

git clone https://github.com/Notebooker-ai/cloudflare-ai-deployer
cd cloudflare-ai-deployer
npm install
cp .env.example .env
# edit .env and set CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID
npm run deploy

On the first deploy it auto-generates an API key (saved back to your .env) and prints your worker’s base URL and bearer key:

Base URL:        https://cloudflare-ai.<account-id>.workers.dev/v1
Bearer API key:  3f2a1c…

Keep both — you’ll paste them into Notebooker.

Choose your models

Models are configured in models.json. Each key is an endpoint type, and the value is a Cloudflare Workers AI model id:

{
  "chat": "@cf/meta/llama-3.3-70b-instruct-fp8-fast",
  "embedding": "@cf/baai/bge-base-en-v1.5",
  "text_to_speech": "@cf/myshell-ai/melotts",
  "speech_to_text": "@cf/openai/whisper-large-v3-turbo"
}

Omit any key to disable that endpoint. Browse the full catalog and copy any model id from the Cloudflare Workers AI model directory, then re-run npm run deploy.

Point Notebooker at it

In Notebooker:

  1. Go to Settings → API keys and Add credential → OpenAI Compatible.
  2. Fill in:
    • Name — e.g. “My Cloudflare AI”
    • API key — the bearer key from the deploy output
    • Base URL — https://cloudflare-ai.<account-id>.workers.dev/v1
  3. Save.
  4. Go to Settings → Models, Add model, choose provider OpenAI Compatible, and enter the model name matching a key in your models.json (e.g. chat, embedding, text_to_speech, speech_to_text).

Now select your model wherever Notebooker asks which model to use — for chat, embeddings, podcasts, and creators.

Verify the endpoint any time with:

curl https://cloudflare-ai.<account-id>.workers.dev/v1/models \
  -H "Authorization: Bearer <your-api-key>"