Your receipts, an LLM, and a spreadsheet walk into a bar
A self-hosted accounting tool that lets AI read your invoices so you don't have to.

What it does
TaxHacker is a self-hosted web app for freelancers and small businesses who’d rather not manually type receipt data into spreadsheets. You upload photos or PDFs of receipts, invoices, even handwritten notes; the app uses an LLM to extract dates, amounts, vendors, line items, and whatever else you need, then stores it all in a structured, filterable database. It also converts currencies using historical exchange rates—including crypto—and exports to CSV when tax season arrives.
The interesting bit
Most receipt-scanning tools lock you into their cloud and their categories. TaxHacker flips that: you can self-host it, plug in your own OpenAI, Gemini, or local Ollama endpoint, and write custom AI prompts for every field. Need to extract a project code or a VAT ID that the app doesn’t know about? Add a custom field, write a prompt, done. The README even warns that local LLM OCR quality “is not guaranteed,” which is a refreshingly honest admission.
Key highlights
- Supports 170+ fiat currencies and 14 cryptocurrencies with historical-rate conversion
- Docker Compose setup with PostgreSQL; runs on port 7331 by default
- Custom fields, categories, and projects with per-field LLM prompts
- Full-text search and bulk operations on extracted data
- “No AI-slop PRs” contribution policy
Caveats
- Explicitly marked as early development; “use at your own risk”
- Requires PostgreSQL 17+, Ghostscript, and GraphicsMagick for PDF processing
- Self-hosted mode is where local LLMs and full prompt customization live; unclear how much is available in a non-self-hosted version
Verdict
Freelancers drowning in paper receipts who want privacy and don’t mind self-hosting should take a look. If you need battle-tested, accountant-approved software today, wait a few releases.