A terminal that runs local Ollama models at the prompt
Some code should never leave the machine. Ghost in Shell lets you pick a local Ollama model in Settings and get the same AI-at-the-prompt experience as with GPT or Claude, with nothing sent anywhere.
Set up in two minutes
- Install Ollama for macOS and let it run in the background.
- Open Ghost in Shell, go to Settings and choose Local. The models Ollama already has are listed; others can be browsed and installed from the same screen.
- Pick one. That is the whole setup. There is no key to paste and no account to make.
Any model id Ollama accepts works, so a small fast model for command writing and a larger one for explanations are both fine. Ghost keeps a list of recent models for quick switching.
What you get offline
- Ask at the prompt. Type "list the ten largest files under this folder" and a staged command appears on your prompt. Type "what does this error mean" after a failed build and the answer prints inline.
- Cmd K. Describe a command and it streams into your prompt for review.
- Typo rescue. A line the shell would reject ("nmp run dev") gets a second look instead of dying as command not found.
Every answer is given your shell, your directory, your git branch and the kind of project you are in, so the local model has the same context a cloud model would.
Privacy, stated plainly
With a local model, nothing leaves your Mac. Not the question, not the directory listing, not the recent commands. Ghost is local-first even with cloud models (context travels only with a question you ask, and only your directory, recent commands and a slice of output), but with Ollama there is no network call at all. You can confirm this by unplugging: everything keeps working.
Read and edit the prompts
Local models are more sensitive to prompt wording than frontier models. Ghost shows you the full system prompt behind each AI behavior in Settings and lets you edit it, add house rules or reset it. Local models are given their whole prompt, uncut, so what you see is what the model gets.
Mix cloud and local
Switching models is a dropdown. Many developers keep a local model as the default and switch to Claude or GPT for a hard question. Keys for the cloud providers stay on this Mac only.
Frequently asked questions
Which Ollama models work with Ghost in Shell?
Any model Ollama can run. They are listed, browsed and installed from Ghost's Settings screen; you can also type a custom model id.
Is the terminal usable fully offline?
Yes. With a local Ollama model selected, the AI features work with no network connection.
Does Ghost send anything to Ollama's servers?
No. Ollama runs on your Mac and Ghost talks to it locally.
Can I use GPT or Claude as well?
Yes. Settings lists OpenAI, Claude and Local side by side. Switch any time.