Quickstart¶
This page takes you from nothing to a processed invoice in a few minutes. You don't need an API key or network access after the install.
Install¶
pip install "invoice-agent[api,synth]"
Get the sample data¶
Write the 50 synthetic invoices, purchase orders, and receipts to dataset/:
invoice-agent generate-dataset
Process invoices¶
invoice-agent process dataset/invoices/016_b04.pdf dataset/invoices/005_q05.pdf \
--pos dataset/purchase_orders.json --receipts dataset/receipts.json
016_b04.pdf: NEEDS REVIEW
vendor Brightforge Maschinenteile GmbH
invoice BF-2026/0404
date 2026-03-17
total EUR 409.05
po PO-2026-0114 (3-way)
lines 2/2 matched to PO lines
[!] PRICE_VARIANCE: Line 1 ('Zahnriemen HTD 8M / Timing belt HTD 8M', PO line 1): unit price 61.56 is 8.0% above the PO price 57.00 (tolerance 2.0% or 0.05).
005_q05.pdf: NEEDS REVIEW
...
Add --json for one JSON result per line, and --ledger ledger.json to keep duplicate and
billed-quantity history across runs.
Score the pipeline¶
invoice-agent eval
eval runs all 50 invoices through the pipeline in submission order and prints field accuracy,
decision accuracy, exception precision and recall, and a confusion matrix.
Run the web page and API¶
invoice-agent serve --pos dataset/purchase_orders.json --receipts dataset/receipts.json
Open http://localhost:8000, upload a PDF from dataset/invoices/, and see the extracted
fields, the line matches, and the decision.
Use your own data¶
Pass your own POs and receipts as JSON, CSV, or SQLite. The formats are in
Data loading. To change
tolerances, pass --policy policy.json; see Concepts.