With Ollama, GhostNote runs a model on your own machine. No API key. No usage cost. No internet. Nothing you ask ever leaves your computer.
Every other tool in this category sends your screen and your questions to someone else’s server. That is fine right up until it isn’t — a work laptop, an NDA, a medical or legal context, a company that logs outbound traffic, or simply not wanting a third party to hold a transcript of everything you looked at.
Running locally removes the question entirely. There is no request to intercept, no account holding your history, and no bill.
ollama pull llama3.3.GhostNote talks to Ollama on localhost:11434. If Ollama is running, it just works.
Llama 3.1, 3.2 and 3.3, Code Llama, Mistral, Mixtral, DeepSeek Coder V2, Qwen 2.5 Coder, Phi-3, Gemma 2, and LLaVA for vision. Pull whichever suits the work and the machine.
A local model will not match Claude Opus 5 or GPT-5.6 on hard reasoning, and a small machine will be slower. For quick factual questions, code, definitions and language work it is genuinely good — for dense multi-step reasoning, a frontier model is still better.
You do not have to choose permanently. GhostNote switches providers from a menu, so you can run local by default and reach for a cloud model when a question deserves it.
Offline mode works on Ghost Free too — and since there is no API cost, running local costs nothing at all.