I’ve been on Beancount for a few weeks coming from spreadsheets. Honestly, the typing is getting tedious. I can write Python and YAML all day, but manual transaction entry? Feels like the opposite of automation.
I got thinking—ChatGPT can write Python now. Could someone build a bot that turns voice or plain text into Beancount entries? Like “spent $50 at Costco on groceries” becomes a drafted transaction I review before committing to Git.
For developers, plain text workflows are amazing. But we’re also lazy. We want automation where it makes sense.
Has anyone tried this? Or is it a bad idea from a plain text philosophy standpoint? Curious if security/privacy concerns make it not worth it.
Hard pass if it means sending transaction details to OpenAI. Where does your data actually go?
I built a Python script last year that parsed email receipts into Beancount entries. Seemed smart. Turns out the bot categorized “office supplies” as “meals” more than half the time. Took me 3 weeks to untangle the mess.
AI drafts are a shortcut that looks good at first. The real work is validation. You’d still review every entry (defeats time savings), or you’d trust it and miss errors quietly for months.
The benefit isn’t speed. It’s lowering the friction for new users. But that’s a UX design problem, not a bot problem.
You’d still review every entry (defeats time savings)
But even review is faster than typing from scratch, right? If the bot gets 80% of fields right and I just fix the category, that’s still a win. Is that naive?