Stoat › Guides

Run a free, local AI coding agent in VS Code with Ollama

If your code can't leave your machine, or you just don't want another bill, run the model locally. Stoat talks to Ollama (or LM Studio) on your own computer: no key, no account, and nothing is sent anywhere.

1. Install Ollama and a model

Install Ollama from ollama.com and start it (the app, or ollama serve).
Pull a model that supports tool calling, which the agent needs to read and edit files. For example:
```
ollama pull qwen3:14b
```
On a laptop with less memory, try a smaller variant such as qwen3:8b.

2. Install Stoat

Install Stoat. In VS Code open Extensions (Ctrl+Shift+X), search Stoat and install Stoat – AI Coding Agent. (Cursor, Windsurf and VSCodium find it on Open VSX the same way.)
Open your project folder and click the Stoat icon in the activity bar (or press Ctrl+Shift+L).

3. Connect Ollama

In Connect a model, pick Ollama (local). Stoat reads your installed models automatically.
Tick the model(s) you want and click Save & connect.

Using LM Studio instead? Start its local server (Developer tab → Start server, port 1234) and pick LM Studio (local) in Stoat.

4. Work with it

Ask a question in Ask mode, or switch to Agent (Shift+Tab) to let it edit.

When the agent finishes, a review bar lists every file it changed. Click a file to see the diff, then Keep or Undo it (or Keep all / Undo all). Commands such as npm test or dotnet build always wait for your approval first.

Troubleshooting

"Nothing answered at http://localhost:11434/v1": Ollama isn't running. Start the app or run ollama serve.
No models listed: pull one first (ollama pull …), then click Try loading again.
It answers but never edits files: the model may not support tool calling. Try another model, or check you're in Agent mode.

FAQ

Does a local coding agent send my code anywhere?

No. With Ollama or LM Studio the model runs on your computer, and Stoat has no server and no telemetry, so your code stays on your machine.

Which local model should I use?

Pick a model that supports tool calling. Larger models give better results but need more memory; start with what your hardware runs comfortably.

Stoat is free and open source. Small agent, takes on the big ones.

Install for VS Code Open VSX GitHub

More guides: How to use DeepSeek as a coding agent in VS Code · Use your Claude API key in VS Code as a coding agent · Use Gemini, Groq or OpenRouter as a coding agent in VS Code · An open-source Cursor alternative for VS Code