Beancount export

[criemen]

Hi,

would you be open to a beancount exporter module? I've added that in my (private) fork, and am still iterating on it, but if you deem it useful, I'd try to upstream it over time.

Tally has now inspired me to set up a flow of bank export -> categorization -> export into beancount.
Some transactions need manual post-processing, but so far it's been surprisingly robust, and more powerful than the insights tally offers.

[davidfowl]

What's the beancount format? Something custom?

[criemen]

It's a plaintext-based accounting software, see https://beancount.github.io/docs/.

In the end, transactions look like this:

2025-12-01 * "Netto" "Supermarket:Netto Netto XXX"
  Expenses:Supermarket:Netto                         38.25 DKK
  Assets:Bank:MyPersonalBank:MyAccount

[criemen]

The philosophical question that I'm asking is:
Do you rather want to be a reusable rule engine, or a frontend to categorized transactions, which requires having a rule engine?

I'm highly interested in the rule engine part, including the LLM classification, and it's working great for my purpose. I less interested in presenting the transactions, because for my bookkeeping, I want to add more data by hand, data from other sources, and query/slice/dice the data in more ways than the static HTML interface allows me to.

[davidfowl]

The front end will remain readonly. The rule engine is the core of the system + LLM based classfication. That was the goal of having lots of different formats. The html output will continue to evolve but purely as a really simple way to further process information.

BTW you can attach more data to transatctions using the supplemental data feature (stil nascent but useful).