RediscoveringMacher

Today I rediscovered Macher¹, a wrapper around gptel.

That is, actually, I just went back to Emacs and I always knew I want to go back to knowing how to use gptel. And I was confident macher was just a set of tools for the completions api to be invoked with. I was wrong.

First I launched gptel². I remembered it to be a buffer chat interface that first thing asks for a buffer (usually to create a new buffer for the chat), has some clickable options like where should it write the responses and which buffers should be the context.

And I saw it outputs Markdown, for example a code block that starts with "reasoning:". But I recalled enjoying something being Org Mode!

It reached me that I should use M-x macher-implement³. In a project.el project per the buffer I call from⁴, supposedly — it puzzled me at first when I still thoughts the default way should be to just use its tools from gptel! But I got reminded that these are usually just a directory root that version control (like .git) was detected in.

There, there's Org Mode⁵. It even continues to work after a "Save As…" which also assigns that buffer to a file so subsequent C-x C-s save it to the same place.

Although, first time I invoked a big prompt, my system had to kill Emacs after everything ran out of resources at some point during the reasoning phase… actually the main issue was that I ran it from a Dired buffer. Next time I was more cautious.

Macher tools have no code execution. This is all just file manipulation, bounded to the project. And every time it finishes a turn… you get a diff buffer. To apply the hunks. No changes are done if you don't apply the hunks, not even if you invoke a next prompt within the chat buffer.

At the end of its response, it gives you an empty Org asterisk heading. You type another prompt in that context… and you hit C-c RET.