Skip to content


Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


Build Status

A tool to convert bank account CSV files to ledger. Guesses account name via simple Bayesian inference based on your existing ledger file. Read about the motivation, the algorithm and the workflow in this article.

Running and installation

banks2ledger is written in Clojure, and Leiningen is all you should need to get up and running.

After cloning the repository, you might want to run lein uberjar to obtain a self-contained JAR file that can be copied to a system-wide installation location. The program can also be run directly via lein run.


Usage: banks2ledger [options]
  available options (syntax to set: -x value)
    -l : Ledger file to get accounts and probabilities
         default: ledger.dat
    -f : Input transactions in CSV format
         default: transactions.csv
    -e : Encoding of the CSV file
         default: UTF-8
    -a : Originating account of transactions
         default: Assets:Checking
    -F : CSV field separator
         default: ,
   -sa : CSV header lines to skip
         default: 0
   -sz : CSV trailer lines to skip
         default: 0
    -c : Currency
         default: SEK
    -D : Format of date field in CSV file
         default: yyyy-MM-dd
    -d : Date column index (zero-based)
         default: 0
    -r : Payment reference column index (zero-based)
         default: -1
    -m : Amount column index (zero-based)
         default: 2
    -t : Text (descriptor) column index specs (zero-based)
         default: %3
   -ds : Decimal sign character
         default: .
   -gs : Decimal group (thousands) separator character
         default: ,
   -hf : Hooks file defining customized output entries
         default: nil
  -dbg : Include debug information in the generated output
         default: false

banks2ledger will write ledger transactions to standard output. It will not modify the contents of any file unless you use an output redirect (eg. > out.dat) in the shell.

The program expects to be set up so the structure of the CSV can be correctly parsed (no guessing there). It also expects to be able to read your main ledger file containing all those transactions that will form the basis of the Bayesian inference for newly created transactions.

CSV column mapping

The default value for -r (-1) means that in case you don't have a reference column in the CSV, you can simply omit this option and no reference column will be used. Otherwise, set it to the column number for the payment reference, and the data from there will be printed as part of the generated ledger entry.

The -t option takes something called 'column index specs' that warrants further explanation. Since the description string forms the basis of the account inference, and different banks provide different layouts in their CSV files (even multiple possible layouts for the same file provider) this is used as a flexible way to create the descriptor.

The specs string provided for the -t option is a string similar to a printf format string, but allows multiple alternatives to be specified. Alternatives will be tried in order from left to right, and the next alternative is considered only if the current one results in an empty string. Alternatives are separated by the exclamation mark, and individual columns in the CSV are referenced by %n (n is the column number; columns are numbered starting with 0).

Examples to provide as the -t option:

  • "%4": get fourth column
  • "%4 %5": get fourth and fifth column separated by a space
  • "%4!%1 %2 %3!%7": fourth column by default, but if that is empty (contains only whitespace) concatenate the first three columns; if that in turn is empty, take the seventh column.

Amount format

The -ds and -gs options allow parsing almost arbitrarily formatted decimal numbers from the amount column. Their usage is entirely optional. With both options omitted, the accepted numbers are of the usual "Western" format, i.e., a string of digits starting with an optional minus for negative numbers, grouped with optional commas for thousands (or other) separation, followed by an optional fractional part after a decimal dot.

Evident garbage (text that cannot be possibly part of a number) both before and after the number is implicitly discarded, so having a currency as part of the amount field should not be a problem.

Examples of setting -ds and -gs for parsing different amount formats:

Example amount Parse options
"1,234,567.89" (defaults)
"1.234.567,89" -ds ',' -gs '.'
"1 234 567,89" -ds ',' -gs ' '
"123_4567.89" -gs '_'

Custom hooks to generate output transactions

The option -hf allows passing a file containing Clojure code that defines custom hooks. These hooks are invoked when generating the output ledger entries; the hooks have the ability to alter the output for certain transactions. This mechanism provides high flexibility as all data concerning the ledger entry, as well as the full power of Clojure (and several helper functions in the code of banks2ledger) are available to the hook.

An example hooks file is provided as part of the test suite at test/data/hooks.clj. A detailed tutorial on this feature is also available.


banks2ledger development is governed, first and foremost, by the author's own needs. Naturally, pull requests to add features or fix problems by other developers are gladly considered.

Because of the above, the program is not aiming to be complete. In particular, it does not implement every documented feature of ledger-cli file syntax. If you run into problems (i.e. because banks2ledger does not parse your ledger file that is flawlessly parsed by ledger itself) please open an issue with a specific minimal example demonstrating the problem.


Feel free to open a pull request if you find a bug, or have a feature you would like to see included.

There are several unit tests you can run via lein test. Make sure they don't break; also, add coverage for any new functionality you might add or regression tests for bugs you might fix.

The script runs the unit tests, and if they are successful, proceeds with doing some end-to-end testing with "real" files. The input files are under test/data/ along with the reference output, which is used to validate the results. The test script also shows the usual invocation (parameterization) of banks2ledger for differently structured CSV files. For real production usage, it is recommended to roll a Makefile or similar solution to process your input files; see this article for an example.