Self-host

Two paths

MIT engine (on-prem parsing)

For teams whose compliance policy prohibits sending PDFs to a third party.

pip install bankstract

import io
import bankstract

with open("statement.pdf", "rb") as f:
    buf = io.BytesIO(f.read())

result = bankstract.parse(buf)

The engine runs entirely in process. No network calls, no disk writes. Source and supported banks: github.com/logickoder/bankstract.

Need a bank added? The engine is MIT-licensed and accepts contributions. If you need a specific bank prioritised, contact us about sponsoring the implementation. Sponsored additions ship to the public MIT engine.

Full cloud stack (self-hosted API)

The same AGPL-3.0 code that powers the hosted service. Brings up the worker, the consumer demo, and a Caddy reverse proxy.

git clone https://github.com/logickoder/bankstract-cloud
cd bankstract-cloud/infra
cp .env.example .env
# fill in TURNSTILE_SECRET_KEY and DEMO_API_KEY at minimum
docker compose up --build

The B2B API runs on the same host at /v1. See infra/README.md for domain, TLS, and backup notes.

Self-hosting means you run your own keys, your own audit log, and pay nothing per parse. The privacy posture is identical to the hosted service: PDFs are parsed in memory, only metadata is stored.

AGPL note: API consumers call you over HTTP and do not inherit AGPL. If you fork and host this code as a SaaS, you must open-source your modifications.