● API · Business plan

Bank statements in, structured data out.

Send any bank’s statement: a PDF or a spreadsheet in English or Arabic, or a scan in English. Get back every transaction with its category, the balance check, insights, and on request the tamper check and the affordability report: the same results as the app, as JSON. Nothing is kept.

Make an API key Business plan · $79 a month

Base URL: https://docsanalyzer.com
Auth: Bearer key
Files: PDF, scan, photo, XLS, XLSX, CSV · 25 MB
Rate: 60 requests a minute per key
Kept: In memory, one hour at most

What teams build with it

Lenders and brokers. Income, obligations and the debt burden ratio from an applicant’s statements, with the tamper check before a decision.
Accounting firms. Client statements straight into the books: categorized rows that reconcile, in your own system’s format.
Landlords and onboarding. Proof of income and balance, read the same way every time, with nothing stored afterwards.

Quickstart

Make a key on your account page and keep it in DOCSANALYZER_KEY. It is shown once; we store only a hash of it. Then: send, poll, read, delete.

import os, time, requests

BASE = "https://docsanalyzer.com"
H = {"Authorization": f"Bearer {os.environ['DOCSANALYZER_KEY']}"}

# 1. Send the statement
with open("statement.pdf", "rb") as f:
    r = requests.post(f"{BASE}/v1/statements", headers=H, files={"file": f})
r.raise_for_status()
sid = r.json()["id"]

# 2. Poll until it is read
while True:
    res = requests.get(f"{BASE}/v1/statements/{sid}", headers=H).json()
    if res["status"] != "reading":
        break
    time.sleep(2)
if res["status"] == "failed":
    raise SystemExit(res["error"]["message"])

# 3. Use it, then delete it
print(res["statement"]["check"]["status"])          # PASS when every balance adds up
for t in res["statement"]["transactions"]:
    print(t["PostingDate"], t["Description"], t["Debit"], t["Credit"], t["Balance"], t["Category"])
requests.delete(f"{BASE}/v1/statements/{sid}", headers=H)

Endpoints

POST /v1/statements

Multipart form with the document in file, and password for a locked PDF (used once to open it, never kept). Answers 202 with {"id", "status": "reading"}. The pages are checked against your plan before reading starts.

GET /v1/statements/{id}

status is reading, done or failed. Once done, the result below; once failed, error.message, a sentence you can show. A text PDF takes about a second; a scan up to a minute a page. Poll every couple of seconds.

DELETE /v1/statements/{id}

Removes the result at once. Without it, a result goes an hour after it was sent.

The result

Field	What it holds
`statement.transactions[]`	Every row: PostingDate, Description, Debit, Credit, Balance, Currency and Category; Payee, Kind and Reference where they are found; Cells, the row in the statement’s own columns.
`statement.check`	The balance check: status PASS or FAIL, and the arithmetic in detail (opening + credits − debits = closing, and each running balance).
`statement.summary`	What the statement says about itself: holder, account, IBAN, period, currency, opening and closing balance, as statement_fields.
`statement.source_columns`	The column names as the statement prints them, in order.
`insights.currencies.{code}`	Per currency: overview (totals, the daily balance, spending by category, top payees), months, category_by_month, recurring payments and the insights, each with the rows behind it.
`integrity`	The tamper check: level, a plain label, and the signals found (editor in the PDF’s history, later revisions, mixed fonts, boxes over figures, balances that do not add up).
`affordability`	The affordability report: balances, salary regularity, monthly obligations and the debt burden ratio, per currency.
`pages`	Pages counted against your plan for this statement.

{
  "id": "3f2a…", "status": "done", "pages": 2,
  "statement": {
    "transactions": [
      {"PostingDate": "03-08-2026", "Description": "CARREFOUR CITY CENTRE", "Debit": 212.4, "Credit": 0,
       "Balance": 7000.41, "Currency": "AED", "Category": "Groceries", "Cells": {"Date": "03-08-2026", …}},
      …
    ],
    "check": {"status": "PASS", "detail": "Opening 8,420.55 + credits … = 28,864.00; …"},
    "summary": {"statement_fields": [{"label": "currency", "value": "AED"}, …]},
    "source_columns": ["Date", "Description", "Debit", "Credit", "Balance"]
  },
  "insights": {"currencies": {"AED": {"overview": {…}, "months": […], "recurring": […], "insights": […]}}},
  "integrity": {"level": "none", "label": "No signs of editing found", "signals": []},
  "affordability": {"currencies": [{"currency": "AED", "balances": {…}, "dbr": {…}}]}
}

Errors and limits

Every error is {"error": {"code", "message"}}; the message is a sentence you can show your users.

Status	Code	When
400	`bad_request`	Not multipart/form-data, or no field named file.
401	`no_key · bad_key`	No Bearer key, or a key that is wrong or revoked.
402	`upgrade · over_limit`	The plan has no API, or the pages left are not enough (with when they reset).
404	`not_found`	Deleted, someone else’s, or past its hour.
413	`too_large`	Over 25 MB.
422	`password_required · password_wrong`	A locked PDF: send or correct its password.
429	`rate_limited`	Over 60 requests a minute for this key; wait for Retry-After.

Pages count against your plan exactly as in the app: a PDF or photo page is one page, a spreadsheet one page for every 50 rows. There are no webhooks: poll the result.

What happens to the data

The file is read in memory on our own server, by our own models. No cloud AI and no outside service sees it.
Nothing is written to disk. The result is held in memory for an hour at most, or until you delete it.
A locked PDF’s password opens it once and is forgotten.
We keep your account, page counts and a hash of each key; never a statement. See privacy.