Setting Up Rustledger for Personal Accounting on Android

 · Gifted

I've been meaning to get serious about personal accounting for a while. I tried beancount about a year ago, got the basics working for both my personal finances and my transport business, then life happened and I stopped updating it. The friction of opening a file, formatting entries correctly, and running checks manually was just too much after a long day.

What changed this time is that I now have an AI agent managing the workflow. But before I get to that — here's how I set up rustledger, which replaced beancount as my ledger engine.

Why Rustledger

Beancount is Python-based, which means a virtual environment, pip, and a runtime that adds weight to an already constrained Android setup. Rustledger is a Rust implementation of the same double-entry accounting system — fully compatible with beancount's file format, but distributed as a single static binary.

No Python. No venv. No pip. Just download and run.

The other reason: rustledger ships an MCP server. That means my AI agent can query the ledger, run balance reports, and record transactions directly. More on that in a future post.

Installing on Android (proot Debian)

I run everything inside proot-distro Debian on Termux. Standard ARM64 Linux binaries work fine here, but static builds are more reliable — they don't depend on shared libraries that may or may not exist in the proot environment.

First, find the right release asset:

curl -s https://api.github.com/repos/rustledger/rustledger/releases/latest \
  | grep "browser_download_url" | grep "aarch64"

You'll see four aarch64 options — GNU and musl, for Linux and Darwin. Pick the musl build — it's statically linked and works cleanly in proot:

curl -L https://github.com/rustledger/rustledger/releases/download/v0.15.0/rustledger-v0.15.0-aarch64-unknown-linux-musl.tar.gz -o rledger.tar.gz
tar xzf rledger.tar.gz
mv rledger /usr/local/bin/
rledger --version

That should print rledger 0.15.0. Installation done.

The File Format

Rustledger reads standard beancount format. Every entry is plain text — accounts are opened explicitly, transactions are double-entry, and everything must balance. A minimal personal file looks like this:

option "operating_currency" "NGN"

2026-01-01 open Assets:OPay
2026-01-01 open Assets:Cash
2026-01-01 open Expenses:Food
2026-01-01 open Expenses:Transport
2026-01-01 open Income:Salary

2026-05-19 * "Daily expenses"
  Expenses:Transport    2,000.00 NGN
  Expenses:Food         5,000.00 NGN
  Assets:Cash

The last line with no amount is the balancing entry — rustledger fills it in automatically.

Validating the Files

Before doing anything with a ledger file, run the check command:

rledger check personal_2026.bean

Rustledger gives precise error messages with line numbers. Common issues when starting out:

  • Account never opened — you used an account in a transaction before declaring it with an open directive
  • Transaction does not balance — debits and credits don't add up
  • Wrong currency — a typo like NGM instead of NGN

I had all three when cleaning up my 2025 files. The error output is clear enough that fixing them is straightforward.

My Setup

I maintain two separate files:

personal_2026.bean — daily personal expenses, salary, passive income, and an envelope budgeting system where I allocate cash to sub-accounts (food, transport, airtime, etc.) at the start of each week.

business_2026.bean — my Korope transport business. Daily income, fuel, driver salary, repairs, and a weekly profit allocation policy: 40% owner salary, 35% retained earnings, 15% emergency fund, 10% savings.

A third file main.bean just includes both:

include "personal_2026.bean"
include "business_2026.bean"

This is the file the MCP server points to, so the agent can query across both contexts in one call.

What's Next

The real unlock is connecting rustledger to my AI agent via the MCP server. Instead of manually editing bean files, I tell Joe what I spent during the day and he records it, validates the files, and commits to git. The ledger stays accurate without any manual formatting work on my end.

That setup — and what it actually looks like in practice — is what the next post will cover.