Today was a good day to clean up quite a few old ideas. A mix of AI backend, markdown frontend, and finally using Menai for real!
Google Gemini reasoning output
I had a long-standing note to allow Google Gemini to generate thinking/reasoning output. Today I implemented it.
This was a very lazy approach: I simply copied and pasted the API page contents from the Gemini website, explained what I wanted, and Claude Sonnet 4.6 one-shotted the implementation.
Markdown blockquotes
When I last updated the Markdown parser, it was set up to handle blockquotes, but there was no front-end rendering of the blockquotes. For about a year I've had a GitHub issue reminding me to fix this.
This one was a little tricky. Qt doesn't really have a nice way to express this concept so in the end we implemented a custom painter to draw the vertical bar for the blockquote.
Along the way we discovered a couple of parsing bugs and a couple of rendering bugs. These are now resolved, and what was immediately satisfying was to discover that Claude had actually used blockquoted sections in a conversation an hour or so earlier!
Using Menai
A couple of months ago, I'd built a Menai pipeline runner that lets Menai's functional code connect with a Python-built pipeline running framework that can do I/O. Today Claude and I updated things so AIs can directly use Menai to edit files (via the filesystem AI tool) and editor buffers (via the editor AI tool).
This is a very early implementation, but a nice proof of concept. There's a lot more to be done in this area, but the design does start with a "dry run" feature that allows an AI to check any edits prior to requesting they be approved by the user.