Published a prototype of the ai-cli console utility for embedding large language models (GitHub Models, OpenAI, Groq, DeepSeek, etc.) into the pipeline for calling commands on the command line. The utility takes a request from arguments or an input stream and sends it to the selected large language model, and sends the resulting response (command, message, data) to the terminal, file, clipboard or standard output. The project is written in the Rust language and distributed under the MIT license.
The main difference from its analogues is that ai-cli is not an AI agent and never executes commands automatically: the utility prints commands in the terminal, emulating keyboard input, after which the user can edit them and press Enter to run. All configuration, operation history, buffer, provider settings are saved in regular YAML text files. Actions with the AI response are determined by the user’s configuration; the utility does not require the installation of any specific terminal emulator.