Tallystick

Deterministic, local-first invoice & e-invoice intelligence. Finance-grade document processing that runs 100% on your machine — no cloud, no API keys, no subscription, no AI hallucinations. Same input, same output, every time.

A tally stick was the original tamper-proof financial record: a notched stick split in two, one half for each party. Verification meant matching the halves. Tallystick does the same for your documents — everything must tally.

What it does

• E-invoice native — reads Factur-X / ZUGFeRD (XML embedded in PDF), UBL 2.1 and UN/CEFACT CII directly: zero OCR, perfect accuracy. Ready for the 2026–2028 global e-invoicing mandates (FR, DE, MY, ID, …).
• Template extraction — digital PDF invoices are extracted with deterministic per-vendor templates (invoice2data format). Teach a vendor once in Teach Studio; every later invoice extracts instantly.
• Finance-grade validation — line math, net+tax=gross reconciliation, duplicate detection, review queue. Nothing silently wrong.
• Automation — watch folder, batch processing, full CLI, local audit trail.
• Export — XLSX, CSV, JSON, QBO (QuickBooks) — plus built-in analytics.

Run

python -m venv .venv
.venv\Scripts\python.exe -m pip install -r requirements.txt
.venv\Scripts\python.exe app.py            # desktop app
.venv\Scripts\python.exe cli.py --help     # headless CLI

CLI

cli.py process <files-or-dir>     # extract + validate + store
cli.py export --format xlsx --out invoices.xlsx
cli.py watch <dir>                # watch-folder daemon
cli.py stats                      # vendor/month totals

Sample data

.venv\Scripts\python.exe sampledata/make_einvoice_samples.py
.venv\Scripts\python.exe sampledata/make_pdf_samples.py
.venv\Scripts\python.exe cli.py process sampledata/out

Architecture

engine/einvoice.py   Factur-X/ZUGFeRD/UBL/CII parsers (lxml + pikepdf)
engine/extract.py    invoice2data template extraction (pdfplumber backend)
engine/teach.py      teach-once template drafting (candidates -> YAML)
engine/validate.py   arithmetic + consistency checks
engine/exports.py    XLSX/CSV/JSON/QBO
store.py             SQLite archive + audit trail
ui/                  PySide6 "Counting House" interface

All processing is offline. The only optional download is an OCR pack for scanned documents (planned; scans are detected and flagged today).

License note

Built on MIT/BSD/Apache dependencies only (pdfplumber, pypdfium2, invoice2data, pikepdf, lxml, pandas). No AGPL components.