I’ve been using hledger since 2022. I started it as an experiment hoping to understand where my money goes, find some patterns in my spending and save some money eventually. I did not identify many patterns, if any, but I still was able to save some money.

I’m going to describe how and why I find tracking money being useful and then I’ll show you my setup.

Why is it useful to track finances

It seems like, at least in my case, I was able to reduce money spend just by tracking where they go. I blame my subconscious for this. For example I probably noticed that I spend too much on food delivery services and then I started to cook at home more often.

I don’t remember any cases during almost 4 years when I found a spending pattern that was an obvious issue. Never happened. Some things happened though:

• I found some cases when I paid too much fees for a service I used regularly, which then led me to find a replacement for it.
• Found that I pay for telecommunications (internet, phone) too much considering my usage. It led me to close some internet accounts and change my phone package.
• I paid too much for the apartment I rented. I didn’t need hledger to figure this out, but doing summary report one day, I noticed the money I spent on it in 6 months were uncomfortably large even though monthly payment was acceptable by me.
• I’ve noticed coffee prices went up :D

How I track it

I use my phone and laptop with Emacs. Files are synced using syncthing. The whole process is half automated.

On my phone I use Termux and Termux:API package. It allows me to get list of all notifications that I can then parse into hledger records.

I have a script (call ledger-check to execute it) that parses all notifications and if a bank notifications found, saves them into the ledger.journal file (which is a hledger file for current year). It runs every minute on my phone in Termux session so notifications aren’t usually lost.

The script makes sha256 hash from each record and adds it to the hledger file as well so next time the script is ran the records not duplicated.

Each record looks similar to this:

;; automation:bf4e2493fc27a6d17bd173ea9c70620759766f29
;; datetime: 2025-09-25 17:32:42
;; content: Paid $20.65 at Place Name Here AUD Balance: $100.41
2025-09-25 eating out
    expenses:food:eat-out                      20.65 AUD
    assets:banks:bankname1

In this case the title and expense category were recognised automatically because I have this place mentioned in my templates list (see the script). Otherwise it would mark it as expenses:unknown which means I then should go and manually update records on my laptop. This is why I keep content: to each record which is full content of push notification received from a bank app. It helps me to deal with “unknown” records later.

Every week on Sunday I then review my spendings and check if amount of money in my bank account matches with what hledger report tells me. I manually adjust my ledger records if there is a mismatch (using special expense category for this).

Thanks to Emacs and ledger-mode I can just simply run M-x ledger-report and choose among many pre-existing reports. I also made a few of my own.

Final Thoughts

I’ve noticed that parsing bank notifications is easier than looking for ways to integrate with a bank or special bank reports that are often available for download.

Keeping all the records in simple text file has been very useful and makes me feel safe. I believe simple txt format will never go away. If hledger disappears I can easily replace it with ledger or beancount. Even with my own scripts if necessary.

I sometimes read that people use a tool for a week or a month and able to tell how useful it is. I wish I have similar skill. I think test driving this tool for 4 years proves its usefulness for me.