Cloudflare Workers AI

Cloudflare Workers AI runs a catalog of open models — Llama, Gemma, Qwen, Mistral, and more — on Cloudflare’s network. You can connect it as a service and its models join the picker alongside everything else.

Cloudflare works a little differently from the other providers. Its OpenAI-compatible endpoint has no “list my models” call, so instead of importing a whole provider’s models at once, Catalyst gives you a search-and-add flow: you search Cloudflare’s catalog and add the models you’ll actually use, one at a time.

Get your Account ID and API token

You need two things from the Cloudflare dashboard.

1. Account ID. In the Cloudflare dashboard, open Workers & Pages (or your account home). Your Account ID is the hex string in the page URL, and it’s shown on the account overview — copy it.

2. API token. Create an API token with Workers AI permissions. The dashboard’s Workers AI token template is the simplest choice — it grants read access to the model catalog and permission to run models, which is exactly what Catalyst needs. Copy the token when it’s shown (you won’t be able to see it again).

Caution

Treat the API token like a password. Catalyst stores it on your workspace and uses it only to call Cloudflare on your behalf — it’s never sent to your browser (see Your keys stay yours below).

Add the Cloudflare service

1. Open AI Services from the sidebar and click Add Service.

2. Choose the Cloudflare tab.

3. Paste your Account ID and API Token. Optionally give the service a display name (it defaults to “Cloudflare Workers AI”). Catalyst builds the inference endpoint for you — https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1 — so there’s nothing else to configure.

4. Click Add Service. The service is created, but it has no models yet — that’s the next step.

Screenshot to add

A capture of the Cloudflare tab in the Add Service dialog goes here. It needs a live Cloudflare account, so we’ll add it later.

Search the catalog and add models

Open the new service. Because Cloudflare has no bulk import, you’ll see an Add Cloudflare Models panel instead of the usual “Refresh” button.

1. Type in the search box to search Cloudflare’s catalog — try a family name like llama, gemma, or qwen.

2. Click Add next to each model you want. Added models show a Added marker so you know they’re already in your list.

There’s no “import all” — you pick the models you’ll use. Add as many as you like; you can come back and search for more any time.

Screenshot to add

A capture of the Add Cloudflare Models search panel (a search term and a few results with their Add buttons) goes here. It needs a live Cloudflare account, so we’ll add it later.

Tip

The search shows text-generation models — the ones you chat with. Cloudflare’s embedding, image, and speech models aren’t listed here.

Use your models

Each model you add behaves like any other model in Catalyst. It appears in the chat model picker — select one and start chatting — and it’s available to workflow LLM and agent nodes too. Set a default or pin the ones you reach for most.

Your keys stay server-side

Cloudflare models are added as standard proxied models, so your API token stays on the workspace and is never exposed in the browser — the model is called through Catalyst, not directly from your screen. That also makes them safe for an admin to curate into the public catalog for everyone on the workspace, the same as any other provider.

Next

• Adding a provider — connect OpenAI, Anthropic, Mistral, Ollama, or one of ~20 other providers.
• Models & providers — see everything your workspace offers and set your defaults.