Accountant24 - Local-first AI agent for personal accounting

Your data stays as plain text files on your machine. Works with any LLM, including local ones. Log spending, import statements, ask questions, teach the agent, track changes with git, and more.

https://github.com/user-attachments/assets/02a0f190-b560-4617-b1a8-8fc124417801

Session Snapshot #1  ·  Session Snapshot #2  ·  Demo Ledger

What you can do

Log spending in natural language

I spent $45 at Whole Foods yesterday

The agent figures out the details and adds it to your ledger as a proper double-entry transaction.

Import from files

Here is my March bank statement, please add missing transactions: ~/Downloads/statement-march.pdf

Add transaction from this receipt: ~/Desktop/receipt.jpg

Drop in a PDF bank statement, an invoice, or a photo of a paper receipt. The agent extracts the text, pulls out every transaction, and adds them to your ledger. The original file is archived in your workspace so you can refer back to it later.

Ask questions about your money

How much did I spend on food this month?

What's my net worth?

Show me all transactions from my last trip to Italy

The agent checks your ledger and gives you a clear answer.

Teach the agent as you go

I'm visiting Italy from April 10th to 15th.
Tag all transactions during this period with the 'trip_italy' tag.

My food budget is $600/month, please notify me when I get close.

Tell the agent once — it remembers and applies it when it matters.

Track every change with git

Every modification is auto-committed to a local git repo. Review your history anytime, roll back mistakes, or push to a private repo for backup.

Quick start

Install

brew install machulav/tap/accountant24

Run

a24

On first launch, Accountant24 creates a ~/Accountant24/ workspace with a pre-configured set of accounts, initializes a git repo, and you're ready to go.

Log in & pick a model

Use /login to connect your LLM provider, then /model to pick a model — and you're ready to start chatting.

Go fully local (optional)

Want your financial data to never leave your machine? Run a local model with Ollama.

1. Download and install Ollama.

2. Pull a Gemma 4 model:

   ollama pull gemma4:31b  # requires ~24 GB+ RAM
   

3. Create a ~/Accountant24/models.json file to register the Ollama provider:

   {
     "providers": {
       "ollama": {
         "baseUrl": "http://localhost:11434/v1",
         "api": "openai-completions",
         "apiKey": "ollama",
         "models": [
           {
             "id": "gemma4:31b",
             "name": "Gemma 4 (31B)"
           }
         ]
       }
     }
   }
   

4. Restart Accountant24, then use /model to select the Gemma 4 model — and start chatting. Nothing leaves your device.

How it compares

Why I built this

For years, I managed my personal finances with YNAB. It worked — but every transaction was a manual chore. I'd fall behind, dread catching up, and put it off for weeks at a time.

One weekend I started playing with hledger and Claude Code, just to see if an agent could handle the bookkeeping for me. I wrote a couple of small skills, and to my surprise, it actually worked — and it was genuinely fun to use.

So I kept going. I packed everything into a standalone agent, tuned it for this one job, and started using it every day. I'm honestly happier with it than any tool I've used for my money before.

If it works this well for me, maybe it'll work for you too. That's why I'm releasing it as an open source project.

Why this stack

Each piece of this puzzle does one thing really well.

hledger is a mature accounting engine with proper double-entry bookkeeping. It's fast, reliable, and stores everything in plain text files you fully own. The catch: it has a steep learning curve — journal syntax, report commands, filter expressions. Not something most people want to deal with.

LLMs are great at understanding what you mean in plain language. But they hallucinate numbers and can't do accounting on their own. Left alone with a ledger, they'd quietly corrupt your books.

pi is the agent framework that glues everything together — sessions, tool execution, LLM communication. It's small, well-designed, and easy to extend.

git tracks every change to plain text files — which is exactly what a ledger is. The agent commits your changes as you go, so mistakes are easy to undo, and pushing to a private remote gives you backups.

Put them together and each piece covers the other's weakness. You speak naturally, the LLM figures out what you mean, hledger keeps the math honest, git makes every change traceable, and pi orchestrates the whole thing. That's the combination I ended up with after trying a few alternatives — and it's been working well ever since.

Credits

Accountant24 wouldn't exist without these projects:

• pi by Mario Zechner — a minimal but powerful framework for building AI agents.
• hledger by Simon Michael — the accounting engine that makes proper double-entry bookkeeping possible.

Both are remarkable pieces of software.

Contributing

This is a personal project I use every day, and I'd love to hear from anyone else using it. Bug reports, ideas, feedback, pull requests — all welcome. If you're planning a bigger change, open an issue first so we can talk it through.

License

MIT