cmd-ai translates plain English into shell commands, aiming to sit somewhere between a search engine lookup and blindly letting an AI execute whatever it wants. Typing something like list all running Docker containers or remove all .DS_Store files recursively gets translated into an actual command, with built-in safety filters and an explanation of what it will do before it runs.
Every AI-generated command gets logged, with a timestamp and exit status, to a local history file, and the tool supports autocompletion to speed up repeated use. It defaults to local Ollama models, which keeps queries off third-party servers unless a developer explicitly configures OpenAI, Gemini, or Claude instead.
Maintained by a single developer, Broda Noel, under the MIT license, it's a small project, around 22 GitHub stars and seven forks, still under active development with open issues around model-swapping and other improvements rather than a mature, widely-adopted tool.