Vane Field GuideOfficial project

Local model setup

Connect Vane to Ollama

Vane can use an Ollama model for local inference. The most common connection mistake is giving a Docker container the host machine's localhost address. The Vane README documents the address to use for each operating system.

Documentation-based guide, checked October 22, 2026. This page does not claim a measured model result.

Understand the connection

BrowserVane containerOllama on host

Your browser opens Vane on port 3000. Vane then calls Ollama's API on the host machine. That second connection needs an address the container can reach.

1. Prepare Ollama

Install and start Ollama using the official Ollama download. Make sure a model is available in Ollama before selecting it in Vane. The model you choose affects speed, hardware use, and answer quality, so record its name when you test results.

2. Enter the Ollama API address in Vane

For Windows and macOS, Vane's official troubleshooting section gives this address when Vane is inside Docker:

http://host.docker.internal:11434

For Linux, the README points to the host's private IP and additional Ollama network configuration. Follow that official Linux guidance rather than copying the Windows address.

3. Test the complete path

  1. Confirm Ollama is running and the model you selected exists.
  2. Confirm Vane can list or select that model in settings.
  3. Ask a simple web question and inspect the returned source links.
  4. If the model works but sources do not appear, check the search side separately.

What “local” does and does not mean

Using Ollama can keep model inference on your hardware. Web searches still need information from outside your machine unless you turn off external retrieval. If you connect a cloud model instead, review that provider's data handling and costs.

Sources

Vane README: Ollama connection errors and Ollama download, checked October 22, 2026.