You can use a local model! It's in Settings in a Thread and you can select Ollama.
But that doesn't work for inline edit predictions, right?