Ledger

What it is

Ledger is “a powerful, double-entry accounting system” run from the command line over a plain-text journal. John Wiegley began it in August 2003; it is BSD-licensed C++ and the tool that hledger, Beancount and the other plain-text accounting projects descend from or react to. Its manual is clear about scope: “Ledger will never alter your input file” and “Ledger is only for analyzing the data, not for altering it.” It “provides no bells or whistles” and suggests GnuCash to anyone who wants an entry interface.

How it works

You write transactions in a journal with any editor. Ledger parses it and answers questions: balances, registers, budgets, forecasts, values at market prices, with a value-expression language and automated transactions for anything else. It is optimistic where Beancount is strict: virtual postings are allowed, lots are not matched against earlier additions, and balance assertions follow file order, as Beancount’s author notes.

For bank data, ledger convert turns a CSV file into entries. With --rich-data, each entry carries the original CSV line, the import date and a UUID that is “a checksum of original csv line”; if an entry with that UUID is already in the journal, it is not printed again. That is content-based identity on the raw line, done right. Without the flag, the raw line is not kept.

Version 3.4.0 (2025-10-22, per the NEWS) added --hashes: a SHA-512 chain over transactions, built from each posting’s account, amount and other details plus the previous transaction’s hash, so that “the single hash value of the final transaction is sufficient to guarantee the shape of the entire history leading up to it”. You record Hash metadata on some transactions and Ledger reports a mismatch.

Data and access

The journal is yours and Ledger never writes to it. Reports come out as text, CSV, XML or Lisp, with custom --format strings; there is no JSON output and no network API. Python bindings expose the core, though Beancount’s author found them hard to build, and the unreleased 3.5.0 notes list many fixes for crashes in them.

Agents

Ledger has no AI features and no official MCP server; none from the community was verified for this dossier. plaintextaccounting.org lists ledger-guesser, a neural network that proposes entries resembling past ones. An agent can use Ledger as a capable read-only query engine and write by appending text, with no gate except the person and, now, a hash chain that shows if history was changed afterwards.

Customers

Technical individuals. The repository has 6,053 GitHub stars (2026-10-10); plaintextaccounting.org (2026) counts about 245 committers and a mailing list of about 900, the largest of the three. There are no review-site profiles. Releases slowed: 3.3.2 in March 2023, then 3.4.0 and 3.4.1 in October 2025, with a large 3.5.0 (more than 575 merged pull requests, C++17) in progress.

History

Ledger 2.x added XML output that Ledger could also read back. 3.0 changed defaults; 3.1 came 2014-10-05; 3.2.0 on 2020-05-01; 3.3 on 2023-02-08. hledger (2007) began as a Haskell take on it and Beancount (2008) as a stricter one.

For ownpurse

Ledger is kin and the original statement of the idea: your books are a file you own and the tool only reads it. Three of its choices line up with ownpurse: never write the owner’s data, identify an import by a checksum of the raw line rather than by date, and chain hashes so the last one vouches for the whole history.

Where it stops is where ownpurse starts. The raw record is discarded unless you ask; the chain is opt-in metadata on a file you can still edit; and every source must first be flattened into one journal by hand or by script. ownpurse keeps what each source reported as facts in an append-only, chained log by default and derives the journal, which a Ledger user could still export and read with Ledger.