- Diagnose whether Ollama is installed, running, and accessible.
- Fix common errors when connecting apps, Claude Code, or gateways.
- Adjust the model and workflow when your machine falls short.
60-Second Diagnosis
ollama --version ollama list ollama ps curl http://localhost:11434/api/tags
Error: connection refused
It means your program is calling localhost:11434, but nothing is listening.
ollama serve
On macOS and Windows, Ollama usually starts as a desktop app. If you closed the app, open it again. On Linux, check the service:
systemctl status ollama sudo systemctl restart ollama
Error: model not found
The name you're using in your app doesn't match the installed model.
ollama list ollama pull qwen3:4b
Ollama responds very slowly
- Switch to a smaller model: 1B, 3B, or 4B.
- Close heavy apps before launching the model.
- Reduce context and input documents.
- Use the hybrid workflow: Claude Code for building, Ollama for processing private or repetitive data.
The app works in the terminal, but not from Docker
Inside a container, localhost points to the container, not your computer. Try Docker's host:
http://host.docker.internal:11434
LiteLLM or a gateway won't connect
First test Ollama directly. Then check the gateway configuration file.
curl http://localhost:11434/api/tags # In LiteLLM, the model is usually declared like this: model: ollama/qwen3:4b api_base: http://localhost:11434
Final checklist
- The ollama --version command works.
- ollama list shows the exact model.
- curl /api/tags responds.
- Your app points to http://localhost:11434 or the correct host if it uses Docker.
- The model fits in your RAM without locking up your machine.