This isn't exactly what you're looking for but I found that by creating a rules document that contains mapping instructions "Any transaction from x credit this account and debit this account. List the payee as <business/person> and add this tag " for example and then upload into ChatGPT and then mass upload all transactions from my credit card for the month, and ChatGPT perfectly for creating the beancount entries.

For anything unknown I created an expense called TBD and instructed it to labely mystery transactions with that.

It works really well.

Bought food in supermarket with credit card
Payment from work received
Currency exchange
Moved money from savings to checking account

For these scenarios, if you type in a payee, Fava will complete the transaction with the last transaction to that payee which should cover this AFAICT. If that doesn't work for you, ideas for improvements or other ways to reuse data from existing transactions would be welcome.

You could also build an extension that uses the Fava json_api to add transactions to your Beancount files.