Ollama setup
Ollama runs large language models locally on your machine. For solo engineers, it is the recommended default: no API key, predictable privacy, and zero per-token cloud cost.
Install Ollama
- Download Ollama from ollama.com for Windows or macOS.
- Run the installer and start the Ollama app (or ensure the background service is running).
- Default API URL:
http://127.0.0.1:11434
Verify from a terminal:
curl http://127.0.0.1:11434/api/tags
You should get JSON (possibly an empty models list before your first pull).
Pull a model
Choose a model that supports tool calling for the best agent experience. Examples:
ollama pull llama3.1:8b
ollama pull qwen2.5:7b
Larger models need more RAM and GPU memory. Start with a 7B–8B model on a typical laptop.
List installed models:
ollama list
Configure MBD Mind
- Open Settings (
Ctrl+,) → Connection. - Under Ollama, confirm the URL matches your server (default
http://127.0.0.1:11434). - Click Refresh models. Available models from Ollama should appear.
- In the model toolbar on the Agent view, select Ollama and your pulled model.
Org admins can set a default Ollama URL in the admin console; desktop still lets you override the URL locally in Connection settings.
Troubleshooting
| Symptom | Fix |
|---|---|
| Connection refused | Start Ollama; confirm nothing else uses port 11434 |
| No models in dropdown | Run ollama pull <model> then Refresh models |
| Slow or weak replies | Try a smaller model, enable GPU if available, or add a cloud provider |
| Agent ignores tools | Pick a model known for tool use; check Activity log for errors |
Local-only mode
In Settings → AI Providers, you can enable local only to block cloud LLM calls entirely. Useful for air-gapped or export-controlled environments.
Next steps
- Getting started with the desktop app — full setup checklist
- Connecting a model — switch models during a session
- Cloud providers — optional upgrade for stronger models