Command AI Resolver (cmd-ai-resolver)

Description

cmd-ai-resolver is a command-line application that translate your ideas into shell commands using AI. It identifies special AI instruction tags (e.g., <AI>your ai prompt</AI>) within your shell commands and sends the entire command-line context along with the extracted prompt to an OpenAI LLM, and replaces the tag with the LLM-generated shell command segment. Tool is configured to be used with the VISUAL environment variable, so you can use an well-known keyboard shortcut (C-xC-e) to resolve your command-line requests with AI. When there are no AI tags found in the command-line it will run the passthrough command, so you can still use it as a regular editor for your command-line.

Features

• based on the OpenAI LLM API with an option to specify a custom base URL OPENAI_BASE_URL (for proxies or self-hosted compatible endpoints)
• supports the gpt-4.1-mini model by default, configurable via the OPENAI_BASE_MODEL environment variable
• option to passthrough the command-line without AI tags to the editor to preserve the same experience as before with VISUAL
• debug logging support with -d flag to see the AI processing steps
• bash and zsh compatible

Installation

Currently, the application is run from the source or by building a binary.

Usage

1. Prepare environment variables for OpenAI API at least the API key:

export OPENAI_API_KEY="your_openai_api_key_here"

2. Prepare your shell by creating the wrapper script cmd-ai-resolver with pass-through functionality to your editor (e.g., vim):

cat $HOME/cmd-ai-resolver-wrapper.sh
#!/usr/bin/env bash

$HOME/cmd-ai-resolver --pass-through vim $1

3. Set the VISUAL environment variable to use the cmd-ai-resolver-wrapper script:

export VISUAL="$HOME/cmd-ai-resolver-wrapper.sh"

4. Test the setup by running resolving a command with AI (C-xC-e):

ls -l /some/directory | <AI>filter for text files and show only the last 5 entries</AI>

TODO

• Multiple AI Tag Processing: Currently, cmd-ai-resolver processes only the first <AI>...</AI> tag found in the input file. Future versions aim to support processing multiple AI tags.