Automation for Beancount.

Features (for rationale see Notes):

• network downloads (via APIs),
• automated categorization,
• insertion of new transactions.

Supported data sources:

• Fio banka,
• Banka Creditas,
• or any other source implementing the API Importer Protocol.

Project status: rough prototype; tested on Linux only.

Create a config file and save it alongside your existing ledger:

$ ls
beanclerk-config.yml  my_ledger.beancount

Import new transactions:

$ bean-clerk import --from-date 2023-01-01
Account: 'Assets:Banks:Fio:Checking'
  New transactions: 3, balance OK: 9830.00 CZK
Account: 'Assets:Banks:Fio:Savings'
  New transactions: 0, balance OK: 50001.97 CZK

2023-01-01 * "Initial import date for Beanclerk"
  id: "dummy"
  Assets:Banks:Fio:Checking   0 CZK
  Assets:Banks:Fio:Savings    0 CZK

Once Beanclerk encounters a transaction without a matching categorization rule, it prompts you for resolution:

$ bean-clerk import
Account: 'Assets:Banks:Fio:Checking'
...
No categorization rule matches the following transaction:
2023-01-03 *
  id: "10000000002"
  account_id: "2345678901"
  account_name: "Pavel, Žák"
  bank_id: "2010"
  bank_name: "Fio banka, a.s."
  type: "Příjem převodem uvnitř banky"
  specification: "test specification"
  bic: "TESTBICXXXX"
  order_id: "30000000002"
  payer_reference: "test payer reference"
  Assets:Banks:Fio:Checking  500.0 CZK

Available actions:
'r': reload config (you should add a new rule first)
'i': import as-is (transaction remains unbalanced)
...

pip install beanclerk

Confirm successful installation by running:

bean-clerk -h

Beanclerk automates some areas not addressed by Beancount:

1. Network downloads: As financial institutions start to provide access to their services via APIs, it is more convenient and less error-prone to use them instead of a manual download and multi-step import from CSV (or similar) reports. Compared to these reports, APIs usually have a stable specification and provide transaction IDs, making the importing process (e.g. checking for duplicates) much easier. Therefore, inspired by Beancount Importer Protocol, Beanclerk proposes a simple API Importer Protocol to support virtually any API.
2. Automated categorization: With growing number of new transactions, manual categorization quickly becomes repetitive, boring and error-prone. At the moment, Beanclerk provides a way to define rules for automated categorization. However, it might be interesting to augment it by machine-learning capabilities (e.g. via the Smart Importer).
3. Insertion of new transactions: Beanclerk appends transactions to the Beancount input file (i.e. the ledger) defined in the config. It saves the step of doing this manually. (With reporting tools like Fava I don't care about the precise position of a new transaction in the file.) Consider to keep your ledger under a version control to make any changes easy to review.

I started Beanclerk to try out some Python packages and programming concepts. Actually, there are a couple of interesting projects of similar sort:

• beancount-import: Web UI for semi-automatically importing external data into beancount.
• finance-dl: Tools for automatically downloading/scraping personal financial data.
• beancount_reds_importers: Simple ingesting tools for Beancount (plain text, double entry accounting software). More importantly, a framework to allow you to easily write your own importers.
• smart_importer: Augment Beancount importers with machine learning functionality.
• autobean: A collection of plugins and scripts that help automating bookkeeping with beancount.

Set up a development environment:

pipenv sync --dev
pipenv run pre-commit install

Run tests:

pytest

Follow Conventional Commits.

Following the Beancount license, this code is distributed under the terms of the "GNU GPLv2 only".