Connect a model

Valet works with the AI model you bring. Open the side panel (the Valet icon in the toolbar), then Settings (the sliders icon) → Add a connection, and pick a provider. Once it's connected, pick a model on its row.

Sign in with a subscription

Claude Pro/Max

Pick Claude Pro/Max and click Sign in with Claude. claude.com opens to approve access; it then shows a code. Paste the code in Valet and click Finish.

ChatGPT Plus/Pro

Pick ChatGPT Plus/Pro and click Sign in with ChatGPT. Valet shows a code and opens openai.com: enter the code there. Valet connects as soon as you have.

Ollama Cloud

Pick Ollama Cloud and click Sign in with Ollama. ollama.com opens to pair this browser with your account, like ollama signin. Click Connect there, then come back. No API key needed.

Use an API key

Pick the provider, paste a key you created on its site, and click Connect. Keys are stored encrypted in the extension and sent only to that provider.

Cloudflare Workers AI and AI Gateway

These addresses include your Cloudflare account id (and, for AI Gateway, your gateway's name). Valet fills in the server URL with {account_id} in it: replace it with yours from the Cloudflare dashboard, paste an API token, and click Connect.

Run it yourself

Ollama on this computer

Install Ollama, pull a model (for example ollama pull qwen3), and pick Ollama on this computer. Valet talks to it at http://localhost:11434, with no setup and no key. Nothing leaves your machine.

Any other OpenAI-compatible server

LM Studio, vLLM, llama.cpp, a company gateway: pick Other OpenAI-compatible server, enter its URL (for example http://localhost:1234), and a key if it needs one.

Which model should I pick?

Valet drives pages through tool calls, so a strong model helps: a recent Claude, GPT, Gemini or Grok model works best. Smaller local models can struggle with tool calling. For screenshot-heavy pages (maps, canvases), pick a model that can see images.

Something else?

Open an issue on GitHub.