Ollama.

Start Ollama

1. Install Ollama using its official macOS instructions.
2. Start the Ollama app and leave it running. If using a command-line installation, start its local service with ollama serve.

The default connection is local, so Gyro does not need an API key for Ollama.

Download a model

Choose a model that fits your Mac’s memory. The release documentation uses this small model to check the connection:

ollama pull qwen3:0.6b

Run the command in your own terminal and wait for the download to complete. A small model is useful for setup, but model capability and speed vary. Larger models require more memory.

Refresh and select

1. Open Settings → Providers in Gyro. Use Refresh models beside Ollama.
2. In your project session, choose Ollama and a downloaded model in the chat picker.
3. Run the first-task example.

Connected: your downloaded model appears in the picker and returns a response while Ollama is running.

Local connection limits

For this release, the documented local-model connection uses a loopback Ollama service on your Mac. Start with the default local address. Remote model API connections use the separate API-key provider path.

Local inference avoids a cloud model service for that connection. Agent tools can still access files, run commands, or make network requests according to the task and approvals. Downloading a model requires network access. Read the provider privacy details.

If models are missing

Confirm that Ollama is running and that a download has finished, then refresh the list in Gyro. Check missing-model troubleshooting before reinstalling.