A simple natural language to shell command translator using Ollama with the codestral
LLM.
Currently it supports only zsh and MacOS.
Ollama must be installed and up and running.
The codestral
model must be available.
Run Ollama in the background and pull the codestral
LLM before using the chatbot:
ollama serve &
ollama pull codestral
Build the project by running:
cargo build --release
Run the chat from cargo:
cargo run
Install the chatbot by placing the binary from target/release/chat
in your path.
For now the chat is a simple terminal chatbot. You can chat with it by running:
chat
- Add a non-interactive mode and execute the commands directly
- Add chat history
- Add a feature which explains the generated command
- Make it configurable to use different LLMs
- Make it configurable to support different additional tools installed on the machine
- Add support for more shells
- Add support for other OSs
These are known issue and will be fixed in the future:
- Sometimes the model will not return only the command but also some explanation.
- Very complex workflows might fail to generate the correct command.