A browser agent that never leaves your computer.
Point Valet at Ollama on your machine and the whole loop stays local: the page, the screenshots, the model's answers. No account, no key, no cloud.
Why run it locally
- Private by design
- With a local model, page content and screenshots go to your own computer and nowhere else.
- Free to run
- No API bill and no subscription. Valet is free and so is your local model.
- No setup, no key
- Valet talks to Ollama at localhost:11434 out of the box. Pick it and start.
- Any local server
- LM Studio, vLLM, llama.cpp or a company gateway: anything OpenAI-compatible works.
Set it up in three steps
- Install Ollama from ollama.com and pull a model that supports tool calling, for example
ollama pull qwen3. - Open Valet → Settings → Add a connection → Ollama on this computer.
- Pick the model on its row and give it a task.
Valet drives pages through tool calls, so model choice matters. Smaller models can struggle with multi-step tasks; for pages where layout matters, pick one that can see images. You can keep a cloud model connected too and switch to it when a task needs more.
Does anything leave my computer with Ollama?
Not to your AI provider: the model runs on your machine. Valet itself has no server, so nothing goes to us either. The pages you browse load as usual.
Which local model should I use?
One with good tool calling, such as a recent Qwen or Llama model. Bigger models handle longer tasks better; vision models help on screenshot-heavy pages.
Can I use LM Studio or llama.cpp instead?
Yes. Pick Other OpenAI-compatible server and enter its URL, for example http://localhost:1234.
Is there a cloud option without an API key?
Yes. Sign in with Ollama Cloud, or with a Claude Pro/Max or ChatGPT Plus/Pro plan.
Keep reading
Give it a small errand today.
Free, open source, and it asks before anything that matters.