Autocomplete.sh is a command-line autocompletion tool that brings large language models into the terminal, with the goal of letting users look up --help and man pages less and focus more on completing tasks.
Purpose It provides AI-driven command suggestions directly in the terminal. Users only need to type <TAB><TAB>, and the tool will call an LLM (using OpenAI by default) to return the most appropriate command suggestion, thereby reducing the steps of copying and pasting between CoPilot or ChatGPT.
Main capabilities
- Context-aware: it takes into account the terminal state, recently executed commands, and command-related --help information, thereby giving more accurate and faster suggestions.
- Flexible models: it supports multiple models such as OpenAI, Groq, Anthropic, and Ollama, allowing users to choose between fast and cheap or powerful performance as needed.
- Secure: it supports local LLMs and redacts prompts to protect sensitive information.
- Efficient: it caches recent queries, improving speed and convenience.
- Cost-controllable: it monitors the scale and cost of API calls.
Usage Type <TAB><TAB> in the terminal to trigger suggestions. The tool generates suggestions by combining the machine environment, recently executed commands, the contents of the current directory, and command-specific help information. Users can choose the model to use through a configuration command and can also view the complete prompt.
Tips
- For command arguments: for example, after typing
ffmpeg # reformat video to fit youtube, press <TAB><TAB>. - For complex tasks: for example, after typing
# create a github repo, init a readme, and push it, press <TAB><TAB>. - Settings can be updated through a command.
Applicable scenarios
- Data engineers: efficiently operate on datasets.
- Backend developers: quickly deploy updates.
- Linux users: smoothly navigate the system.
- Terminal beginners: build confidence in the command line.
- Efficiency seekers: simplify repetitive tasks.
- Documentation readers: quickly understand commands.
Installation and maintenance It supports local installation, and development versions can also be installed from local files. The project is currently maintained by Sean Kruzel (@closedloop) of Closedloop.tech, and contributions and bug fixes are welcome. The project uses the MIT license.
Limitations Complete information is not currently provided in the materials regarding specific installation commands, configuration details, supported model versions, API cost calculation methods, and platform compatibility.
