@yasunogithub Do you know how other CLI tools interface with LLMs? Is your approach in this PR inspired by similar approaches in other tools?

@yasunogithub Do you know how other CLI tools interface with LLMs? Is your approach in this PR inspired by similar approaches in other tools?

@cruessler
Honestly, I don't have deep knowledge of how other CLI tools specifically interface with LLMs. But based on what I researched, there seem
to be a few common patterns:

1. Direct API calls - Tools like aicommits make HTTP requests directly to OpenAI/Anthropic
2. Local model integration - Tools use ollama or similar for local LLMs
3. Shell command execution - Some tools spawn subprocesses to call LLM CLIs

Our approach is simple but flexible - we execute any shell command you configure. So you could:

• Call curl to hit OpenAI API directly
• Use ollama run llama3.2 for local models
• Run custom Python/Node scripts
• Chain multiple commands with pipes

The template variables like {staged_diff} give the commands git context, and the RON config lets you set up multiple helpers with hotkeys.

I chose this approach because any CLI command works - users aren't locked into specific LLM providers or APIs. Want to try a new AI
service? Just update your config file.

Thanks for the context! I was mainly asking in order to avoid reinventing the wheel in case there are already existing patterns out there.

Thanks for the context! I was mainly asking in order to avoid reinventing the wheel in case there are already existing patterns out there.

Thanks, I agree—this implementation is simple but still consistent with the common patterns, so I don't feel like reinventing the wheel.