swimmingbrain
Joined 7 months ago
Joined 7 months ago
It connects TeXbrain with braiNOTES over a local websocket. You give your agent tools such as "render page 3 of this notebook" or "insert this LaTeX at the end of this .tex file" and it reads the handwriting and compiles it.
The instructions for VS Code, Cursor, Codex, Claude Code or whatever you use are in the repository. Basically, everything runs on your machine, except of the model.
If you have any questions feel free to ask (or feedback/bugs!)
It compiles on its own a second after you stop typing, and there's also an "Auto" button next to "Compile" if you prefer to press Ctrl+Enter yourself.
The TFM error usually means a font file wasn't available. However, the service worker validates the downloads before caching, so any error pages or empty bodies get rejected (which makes posoned cache unlikely), but a stale state may still be possible.
Hard refresh bypasses the service worker, I patched around that last month but apparently not well enough.
Also, better error logs and reporting is already on my ToDo-list, sorry for that.
I'll make New Project also fall back the same way, so it works in Firefox and Safari without the detour.
The format and packages have to agree, and when I pulled in newer packages against that old format, some things broke (hyperref had changed its aux label format, for which I ended up backporting a padding fix, and babel was missing some aliases in the mismatched version, e.g. brazil-portuges).
And yes, I'd be very happy to help with that and will get in touch via the volunteers page!
That said, most of this repo is solving the wrong problem. "Answer before reasoning" actively hurts quality, and the benchmark is basically meaningless. But the anti-sycophancy rules should just be default. "Great Question!" has never really helped anyone debug anything.