Home

This page says what wisp is, how to install it in four lines, and where to find everything else.

wisp shows grey ghost-text command suggestions in zsh. A small code model running on your own machine writes them, and it reads what you can see: the terminal output above the prompt, your shell history, the files in the current directory and your git branches. After a command finishes, wisp predicts the next one; while you type, it completes the line.

% ls
Makefile  docker-compose.dev.yml  docker-compose.prod.yml  src/
% docker compose -f docker-compose.p|rod.yml
                                    └─ grey suggestion; → accepts it

The file name was never in your history. wisp completed it because ls had just printed it.

Quick start

brew install llama.cpp                      # or a llama.cpp release on Linux
curl -fsSL https://raw.githubusercontent.com/lolabradora/wisp/main/install.sh | sh
wisp setup                                  # model download + background service
echo 'eval "$(wisp init zsh)"' >> ~/.zshrc

Open a new shell and run wisp doctor to check the result. Installation explains each step.

Which terminal do I need?

wisp runs in any terminal emulator on macOS or Linux, as long as the shell is zsh. What changes is whether it can read the screen.

Where your shell runs	Setup	Context the model gets
Inside tmux, in any terminal (Ghostty, iTerm2, Alacritty, Terminal.app, ...)	None	Screen, history, files, git branches, session commands
WezTerm	None	Screen, history, files, git branches, session commands
kitty (experimental)	Two lines in kitty.conf	Screen, history, files, git branches, session commands
Any other terminal, with WISP_SCREEN_CMD set	A command that prints the visible screen	Screen, history, files, git branches, session commands
Any other terminal, nothing configured	None	History, files, git branches, and the commands this session ran with their exit codes

kitty support has not been verified on a real kitty install yet. Details for every row are in Terminal support.

Pages

• Installation: prerequisites, wisp setup, model choice, upgrading, uninstalling
• Terminal support: screen sources and how to check which one is active
• Configuration: environment variables, pausing, using another llama-server or model
• How it works: architecture, the prompt, token healing, caching and cancellation
• Privacy and security: what the model sees and what stays on disk
• Troubleshooting: symptoms, causes and fixes
• FAQ: short answers to common questions
• Contributing: dev loop, repo layout, where to start