tallyA local rule engine for transaction classification. Pair it with an LLM to eliminate the manual work.
Works with Claude Code, Codex, Copilot, Cursor, or a local model
Tally has no AI built in. It's a CLI you can pair with an agent to categorize your transactions.
The Problem
WHOLEFDS MKT 10847 SEATTLE WA
AMZN MKTP US*2K7X9
SQ *JOES COFFEE SEATTLE
Bank transactions are cryptic. Bank categories say "Shopping" when you want "Kids > Clothing" vs "Home > Furniture". They lump restaurants together but you need "Coffee" separate from "Fast Food".
The Solution
LLMs have read the internet. They know what "WHOLEFDS MKT" is. They understand "SQ *" means Square. Tell them your rules in plain English:
"ZELLE to Sarah is babysitting → Childcare"
"COSTCO with GAS is fuel, otherwise groceries"
The AI writes pattern-matching rules to a CSV. No database. No CMS. Just a text file you control.
Install
Linux / macOS
curl -fsSL https://tallyai.money/install.sh | bash
Windows PowerShell
irm https://tallyai.money/install.ps1 | iex
Python (uv)
uv tool install git+https://github.com/davidfowl/tally
Workflow
1
tally init
Creates config folder with AGENTS.md instructions
↓
🤖
AI Agent Loop
Your AI reads AGENTS.md then:
tally discover --format json → finds unknown merchants
AI identifies merchants, writes rules to CSV
Repeat until unknowns are minimal
↓
2
tally run
Generates HTML spending report with all transactions categorized
Open Source
Tally is MIT licensed. Contributions welcome!