I'm evaluating whether self hosting accounting software is at all feasible to meet CMMC requirements and so far have my sights set on ERPNext. I configured my bank to send me an email alert for every transaction and intend to parse that and appene it to my ledger, but the API to do so is fairly annoying, so hearing that beancount is meant to be automatable is intriguing indeed
BTW the other thing that shocks me about quicken classic is lack of version control - the database can become corrupt and your only recourse is restoring from backup or sending the file to support and having them manually fix it!
I am working on a plugin to pull categories and transactions from the Hibiscus DB (H2 or PG) to Beancount [2]. It is not there yet, but the process overall looks promising.
[2]: https://github.com/Sieboldianus/beancount-hibiscus-importer
>Developer Information (in german)
Which sends to a Javadoc which is actually a 404.
To be honest, I wouldn't work on/with a project that uses not-English as primary language (even if it was my own). Maybe a bit of cleanup and documentation would help make it more accessible.
Actually, a client is migrating their system and needs to redo their bank synchronization. Shall I try with Google Translate, or is there someone who would also work on updating some of the resources to English?
I'm pretty sure FinTS/HBCI is a mostly only used here in Germany sadly. Which is a shame because all this "OpenBanking" stuff requires registering/paying/certificates etc.
If you are okay with Plaid[1], many of their bank connections are now using OAuth-style authentication instead of password sharing. I actually added a new feature called Direct Connect[2] a while back to allow any plaintext accounting book users to pull CSV directly via Plaid API through BeanHub. We don't train AI models with our customers' transactions, and if we want to, we will ask for explicit consent (not just ToS) and anonymize customers' data.
If you're okay with the above, the key to achieving a high automation level is the ability to pull CSV transaction files directly from the bank in a standard format. Maybe you can give it a try. We have 30 days free trial period.
I am not so familiar with the CMMC requirements, as you mentioned, but for us to access transactions from some banks, such as Chase, Plaid requires us to pass an auditing process about our security measurements. Is the CMMC compliance your company needs to meet to take a third-party software vendor into considerations?
[1]: https://plaid.com
[2]: https://beanhub.io/blog/2025/01/16/direct-connect-repository...
In 2024, CFPB had mandated all U.S. banks to open up data history to SaaS/fintech vendors, if the customer gives permission.
The banks having to do it if we want them to, doesn't not mean we have to want them to.
> if the customer gives permission
Of course once permission has been given, it is part of that dataset (whether or not that is made clear by the app) and numerous other datasets (unless the law requires companies not share that data, and companies all fully obey that law, and they don't have any little data accidents).
We originally used Quickbooks Online and I'm glad we decided to switch to ERPNext a few years ago.
...not always, they don't. Some financial institutes provide this, some do not. I know very specifically that some FIs do not have any way to grant read-only access, and that aggregators like Plaid, Flinks, Quicken, etc. will simply fallback on good-old-fashioned "give us your username/password and we'll screen scrape your account info".
I can personally attest to this being true in 2025 in Canada, at least.
- some transactions go through without a text
- some transactions generate two identical texts
- for Forex transactions, the amount in the text and the amount on your statement will not match
- some texts are ambiguous in that they could have been generated by two different kinds of transaction; especially true of deposits & credit card payments
In the end I gave up, accepted that accounting will never be smooth & simple, and now just generate a CSV every month by hand.
Probably built 10+ importers, plus some plugins to do automated transaction annotations.
I have not made any update for many years now, because: - Downloading statements is still a pain, have to manually go through all websites. Banks are bad at making the statements available, and worse making it possible to automate it. - The root of the issue is actually that beancount is too slow. Any change/update takes ages. Python is both a blessing (makes it easy to add plugins/importers etc), and a curse (way slower than some other languages.
I believe the creator of beancount has started working on v3 with a mix of C++/python, relying on protobufs, a C++ core for parsing, etc. AFAIK, that is not production-ready yet.
I've found HLedger and Shake to be fast enough to process almost a decade of finances. Dmitry Astapov has an extremely well produced tutorial workflow[3].
How have you managed the PDF parsing? Mine has become a bit of a mess dealing with slight variations in formatting as they change over time. I've been considering using LLMs but have been nervous about quality.
[1]: https://hledger.org [2]: https://blog.danslimmon.com/2019/07/15/do-nothing-scripting-... [3]: https://github.com/adept/full-fledged-hledger
What is Shake?
If you have 1e5 - 1e6 of lines of transactions, I think a SQLite database would be a huge step forward. If you have much more than that, you probably need an ERP system.
Of course the text files make it ~easy to enter transactions, but maybe there's an elegant way to use those for ingestion only; that does make the system much more complicated to use. That might not be a problem for the kind of person using plain-text accounting over the course of years though.
https://groups.google.com/g/beancount/c/iTdRuvZnE4E
I found the migration pretty confusing and haven't found good documentation on how to go from v2 to v3.
The best I've found is this unofficial write-up from an experienced Beancount user:
Is there a point in migrating already?
The maintainer says here that v2 is officially deprecated:
>You should not use v2 anymore.
https://groups.google.com/g/beancount/c/iTdRuvZnE4E/m/o9V91W...
I can link my secondary bank account to my main bank's app so I can see the balance in one place, but the catch is that I need to refresh this authorization through the app every 90 days.
Ideally, you'd just use your banking credentials to authorise the API access and pull data through that. What this requires in practice, I have no idea but it probably involves a bit of bureucracy.
If you can't, you can try use one of the open banking providers such as TrueLayer, Plaid, Nordigen (seems to be acquired by GoCardless: https://gocardless.com/bank-account-data/), etc. Most have a free/dev tier that nevertheless allows connections to real accounts and might be enough for personal use.
Finally, screen-scraping is potentially an option. One of the few benefits of shifting everything to SPAs is that you generally have clean JSON APIs under the hood that are easier to interface with than "conventional" screen-scraping involving parsing HTML.
My thought for working around tracking new transactions without a third party is to just set up email alerts so I get a notification on every charge, deposit etc and set up some cron job to read new emails and update my books.
Have you considered using Playwright?
I used aider[0] recently to log into my work's payslips and download all the relevant payslips into JSON format (with values encrypted). It took about 3 hours, but that's mostly because of my lack of knowledge of good CSS selectors.
I deal with two banks for credit cards.
One (call it "Blue Bank") allows me to download a statement. I filter out a couple of things (payments mostly), check that it matches the paper statement balance, and post it. About 15 minutes start to finish.
The other (call it "Orange Bank") allows me to download a "statement". I filter out a couple of things (payments mostly), check my previous month's transactions to see which ones at the beginning of the file actually go in the current billing period (not already paid), stare at the last transactions to see which ones actually were posted to the current billing period (not after the cutoff), run the script to check the total (nope, doesn't match) then do that a couple of times until it matches. The time they changed the meaning of the "credit" column from "just confirming this is a credit" to "it's a credit, you need to flip the sign" it was 45 minutes.
But hey, it's all CSV!
My other issue is with stores like Costco that sell both household goods, groceries, clothes, and even misc kids stuff. I like to track each separately. Which means I then need to fetch and analyze the receipts.
That is a reality. To make my life easier, when I check out at a store, I put all my grocery items first on the belt. Then everything else. Usually "everything else" is only a few items. So I categorize those additional items, and then specify "Groceries" for the rest.
Often I buy only groceries, and I throw those receipts away. When I'm in a ledger/beancount session, if I don't have a receipt, that means it was just Groceries.
This method alone really reduced my time dealing with receipts.
I am starting a new business now and intend to see how far I can take plain text accounting in that context. I plan to use mercury for banking and want to automate as much as possible. I would also like to associate invoices that are stored in a self-hosted paperless-ngx instance.
I am looking to deprecate my Quickbooks usage after this year since it is such a pain to split payments into multiple chunks automatically and I don't really know what I am getting for $60/month.
One thing which I disagree with in this article is the focus on file based data storage:
> That makes it 10 times harder because you need to parse the text file, update it accordingly, and write it back. But I am glad I did. That guarantees all my accounting books are in the same open format.
This quote captures my issues with it. It just makes things so much more difficult; and it makes the whole process slower as the file grows. I remember that when I used hledger for tracking my expenses over 3 years, I had to "close books" once a year and consolidate all the transactions for the past year into 1 entry in a new ledger file to keep entry/query operations fast.
I get the sentiment; you want open data formats that remain even after your app is shutdown. But you can get the same by using open formats; maybe a sqlite DB is good enough for that?
The only thing that would be more complicated with a DB is versioning & reviewing commits like this app does; which does seem like a very exciting feature.
When do you split the files? How do you track which data resides in which files? Does one file represent one kind of data (table)? Does it reflect data within a given time range? Do you sometimes need to retrieve data that crosses file boundaries?
You quickly lose the simplicity inherent in saving to just a single file.
Which is where Sqlite shines. It's a single file. But with a full, user defined schema. And can update it and query it incrementally, without having to read and write the entire thing frequently. It handles all of that complexity for you.
> When do you split the files? How do you track which data resides in which files? Does one file represent one kind of data (table)? Does it reflect data within a given time range? Do you sometimes need to retrieve data that crosses file boundaries?
Not really. Splitting anywhere from per-day to per-year is probably fine! Or split arbitrarily and merge the files at runtime. Make it configurable! Add tools to split or merge files, it's really not that hard, a far cry from a database engine.
> You quickly lose the simplicity inherent in saving to just a single file.
No, you really don't.
> Which is where Sqlite shines. It's a single file. But with a full, user defined schema. And can update it and query it incrementally, without having to read and write the entire thing frequently. It handles all of that complexity for you.
That you need a particular tool or library to interact with.
I'm not going to try and sell you on the benefits of using plaintext tools because you've already clearly made up your mind, but there are reasons even if you can't see them. SQLite has like 160k lines of code complexity that isn't necessary and is inherently less composable.
I get where you're coming from. My books are also growing big right now, and indeed, they have become slower to process. Some projects in the community, such as Beanpost [1], are actually trying to solve the problem, as you said, by using an RMDB instead of plaintext.
But I still like text file format more for many reasons. The first would be the hot topic, which is about LLM friendliness. While I am still thinking about using AI to make the process even easier, with text-based accounting books, it's much easier to let AI process them and generate data for you.
Another reason is accessibility. Text-based accounting only requires an editor plus the CLI command line. Surely, you can build a friendly UI for SQLite-based books, but then so can text-based accounting books.
Yet another reason is, as you said, Git or VCS (Version control system) friendliness. With text-based, you can easily track all the changes from commit to commit for free and see what's changed. So, if I make a mistake in the book and I want to know when I made the mistake and how many years I need to go back and revise my reports, I can easily do that with Git.
Performance is a solvable technical challenge. We can break down the textbooks into smaller files and have a smart cache system to avoid parsing the same file repeatedly. Currently, I don't have the bandwidth to dig this rabbit hole, but I already have many ideas about how to improve performance when the file grows really big.
However, I feel like maybe a different approach could be to store all the app state in the DB, and then export to this text only format when needed; like when interacting with LLMs or when someone wants an export of their data.
Breaking the file into smaller blocks would necessarily need a cache system I guess, and then maybe you're implementing your own DB engine in the cache because you still want all the same functions of being able to query older records.
There's no easy answer I guess, just different solutions with different tradeoffs.
But what you've built is very cool! If I was still doing text based accounting I would have loved this.
I dream of a future where anyone can download a ZIP of all the PDFs they've ever received from their bank, drop it onto a local app, and wait while it creates an entire accounting setup for you.
Edit: also not mentioned here is Fava, which is a really nice web UI for Beancount (https://beancount.github.io/fava/). I share a link with my accountants, and they find it convenient (for downloading files, at least).
I use git locally because the ledger is extremely sensitive
At the same time, I did automate 90% of my business beancount import by writing a custom stripe importer that imports transactions from stripe. As for the expenses, I still enter them manually in the aforementioned 20-30 minute session.
That sounds a lot - I spend less.
Consider entering the transactions into a software like KMyMoney and write a script to export to beancount format. Entering/importing is a lot easier in a SW like KMyMoney (e.g. it does decent matching of the new transaction to prior transactions of similar amounts).
These days, though, my process is more manual. Around every 24 or 48 hours, or immediately after making a transaction, I'll record the transaction in my ledger, which I store in Google Sheets (!) instead of a .ldg file. No more CSVs, no more pure functions of bank output.
Sometimes I miss the .ldg format. But I don't really miss maintaining the automated system. Google Sheets isn't as expressive as Ledger, but it is sufficient for my needs. Charting is a bit easier. YMMV!
To me, the most essential pivots to get me back into personal accounting were to record expenses both manually and ~daily. If I were to return to ledger again — which I might! — I'd focus on those aspects more than the automation.
https://github.com/zdw/ledgercalc
And fed it with pile of scripts that extracted bank PDFs -> Text -> ledger entries, and shoved it all in git.
This looks like it some superset of that, but in general I found the files + text to ledger process to scratch a great itch.
https://plaintextaccounting.org/ is the resource for most of them, and has good resources for making them work for you. It's not for everyone, though, many people just prefer spreadsheets or apps, and that's fine.
So BeanHub is built on top of Beancount and uses double-entry accounting. It's one of the benefits of double-entry accounting. Many accounting software are not good at dealing with multi-currencies or custom currency. With Beancount, you can define any commodity you want, create transactions, and convert them with different currencies easily. For example, you can define a commodity TSM and create transactions[1] like this:
2025-01-01 commodity TSM
2025-03-05 * "Purchase TSMC"
Assets:US:Bank:WellsFargo:Checking -2,000 USD @ 100 TSM
Assets:US:Bank:Robinhood 20 TSM
[1]: https://beancount.github.io/docs/beancount_language_syntax.h...
I feel that there are some people who legitimately enjoy looking at money on spreadsheets and implementing budgets/categorizing purchases and I think that YNAB is great for those people, but I personally hate even THINKING about money, let alone interfacing with budgeting software every couple days
How is that different from "give each dollar a job"? The only difference I see is that it forces you to make the categories add up to the amount of your paycheck.
I copy over all the previous months budget amounts and tweak it to match the current paycheck, if it's different. And frankly don't care too much if it's a few dollars off.
I do this in a spreadsheet. I had written an app for budgeting but it was too much hassle keeping it up to date. New versions of Mac OS would break it in subtle ways and wasn't worth the effort of all the bug fixes.
YNAB with Credit Cards was difficult for me, as was envelope-based budgeting (what "give every dollar a job" is called) because I also was used to the typical "set limits on categories and stay within them"-style budgeting (Like Mint, at least Mint way back when it first came out, I haven't touched it in years). YNAB is very different in that it doesn't let you allocate dollars that are not in your bank account. You can't say "$300 for eating out" unless you have $300 in your bank account and YNAB doesn't care that you might have that money available by the time you want to spend it, it forces you to allocate the money you actually have and every time you get paid you allocate it into categories with the long-term goal of getting a month (or months) ahead in your budgeting (not spending the money you made this month on stuff you need in this month).
Credit cards were also hard to wrap my head around. In a debit-only world it all made sense but CC's complicated things for me. I really enjoyed Nick True's videos on Youtube, they helped me with this a lot but a simplified way to think about this is:
* You put $200 in your "groceries" category (aka envelope)
* You go to the store and spend $60 (on eggs I assume?) and pay with your CC
* In YNAB-land you will record that transaction (or it will be auto-imported) and you will assign it to the groceries category but since YNAB knows you spent this on a CC (you always record which account the transaction happened on) it essentially takes $60 out of the "Groceries" envelope and moves it to the "Chase Sapphire" (or whatever you name it) envelope. You set aside the money for your CC purchases at time of purchase and then when the CC bill comes due it's paid out of that "envelope".
In this way YNAB has become a layer on top of all my finances and I care little about how much money is in any given savings/checking account since YNAB tracks everything. I just make sure there is enough to cover CC payments (there always is) or any big transfers I want to make (like moving money to a HYSA).
I've automated as much as I can with YNAB but I do spend 30min-1hr every few weeks (this is not what they recommend, but it works for me) reconciling my accounts. I totally get if people don't want to do that or don't see the value in it. Personally I love knowing where every dollar of mine is and tracking every purchase/transfer.
I really feel like I ought to be able to do them myself - it's just following some rules, and my accounts aren't that complex. Still, it was just enough of a pain that it was easier to hire someone overseas for cheap (especially since I know what the business' numbers should roughly come out to, so I can validate their work).
But as I've been using the latest AI models, I really feel like this is something that's going to be fully automated by AI in the next 1-2 years (at least for my very simple use case). It's simple enough that an AI agent should pretty trivially be able to fetch the docs from the various places that I sell upload them, categorize transactions (this is already basically automated by rules for me anyway) and then do whatever it is that bookkeepers do to close the books.
I can't help but think that bookkeeper is not going to be a profession in five years, and I'm just not sure where those people go. It's not like automating bookkeeping will expand the economic pie enough to create new jobs - I don't believe it's a bottleneck to anything at this point.
Many customers have asked me about AI offerings, and I am considering them. While this is doable with modern LLM technologies, I need to consider many issues.
The first is that nobody, myself included, likes their data being part of someone else's machine-learning training pipeline. That's why I promised my users that I wouldn't use their data for machine learning training without asking for explicit consent (and, of course, anonymization will be needed).
While I know everything involved in AI sounds cool, do we really need LLM for a task like this? Maybe a rule-based import engine could kill 95% of the repeating transactions? And that's why I built beanhub-import[1] in the first place. Then, here comes another question: Should I make LLM generate the rule for you or generate the final transactions directly?
Yet another question is, everybody/every company's book is different from one to another. Even if you can train a big model to deal with the most common approaches, the outcome may not be what you really need. So, I am thinking about possibly using your own Git history as a source of training data to teach machine learning models to generate transactions like you would do. That would be yet another interesting blog post, I guess if I actually built a prototype or really made it a feature for BeanHub. But for now, it's still an idea.
While my "books" are synced via Quickbooks, they (accountants) really seem to love having the PDF in hand. I just need the PDFs and they do not send them via email...
It's not yet ready for release, but I should be ready for beta-test within 2 months. If you're interested I would be happy to add you to my list of "people to notify when I am ready to beta test"
The upside of a "real database" like postgres is when you need multiple people in the books making changes at the same time. Until your accounting department grows into multiple people, Beancount, hledger or any plain text accounting system would be fine.
For me, the main benefit is tax lot matching and an auditable trail of sales should the IRS come knocking. It's impossible to do this properly in a spreadsheet (I mean... it is possible, but no one will complain if it's done wrong until it's too late). With beancount, matching is much more straightforward. I have a python script that does it automatically for each sale using normal FIFO, tax loss minimization, etc. Due to beancount's internal checks, I am certain that these are correct once they're entered into the ledger. There is no chance of failure, and if the IRS ever asks me to justify a capital gain / loss figure, it's very easy to point out that I keep track of all my shares and have for the past several years and I maintain a solid record of the entire history of each lot.
You give beancount the balances at the end of each month straight from your bank / broker statements, and it throws an error if your transactions don't match up.
You won't make a mistake using any of them.
I used and still use both, since they share the file format. Beancount is more opinionated, with slightly different file format, so I didn't bother adding it.
But I did write an importer for the csv files my bank provides and with smart_importer I don’t even have to categorize the statements anymore (although there are mistakes sometimes). I don’t gather metrics though , I use fava to have a visual view of my books.
I usually spend half an hour per month maintaining the ledger .
However, contrary to the article's automated method, my workflow is to manually input transactions every day (or every couple days, depending on what's going on) and balance my accounts. It's a bit of a ritual, but I like having a really good handle on day-to-day spending. Plus, I find ~10 minutes once a day way easier than (e.g.) ~3 hours once a month, even if it's the same amount of time overall.