How PDF Probe works
By Clioos 团队
PDF Probe reads the tables out of your PDF and gives you a clean Excel file. No cloud service — everything runs on the same server that serves this site, and your file is deleted within 1 hour. Signing in is optional for the free tier; buying pages needs an account.
Digital PDFs: read from the text layer
Most PDFs produced by a computer (bank downloads, invoices, reports) contain a real text layer. We open the file with pdfplumber and check that text layer first. Tables are then extracted with two strategies — one that follows the visible ruling lines, one that follows the text positions — and we keep whichever one fills more cells. If neither finds enough rows, we fall back to reconstructing lines from the individual words on the page and parsing those into rows.
Scanned PDFs: rendered and read with OCR
Image-only PDFs — scans, phone photos of statements, faxed pages — have no text layer, so a normal converter returns nothing. We detect that (a page with almost no text) and switch methods: each page is rendered at 300 dpi with pypdfium2 and read with Tesseract OCR, running on this server. If the first OCR pass finds too few rows, a second pass runs with different settings and we keep the better result. Cells the OCR is unsure about are highlighted red in the preview, so you know exactly where to check.
Bank statements: balances are reconciled
Bank statements are the hardest case: Withdrawals/Deposits two-column layouts, running balances, headers that repeat on every page. We handle single-amount and two-column layouts, recompute the running Balance column from the statement's own beginning balance, and compare the computed ending balance against the ending balance printed in the file. The difference is shown in the preview before you download — if it is not zero, a row was probably misread, and the red-highlighted cells tell you where to look.
What you get
The download is an .xlsx with a Transactions sheet (Date, Description, Amount, Balance) and a Summary sheet, or the same data as .csv. Amounts are real numbers and dates are real dates, so the file imports cleanly into Excel, Google Sheets or accounting tools. Multi-page files are merged into one sheet (up to 20 pages and 10 MB per file).
Privacy: deleted within 1 hour
Uploads, rendered page images and results are deleted automatically within 1 hour, and the upload folder is excluded from all backups. No cloud OCR service is involved — your file never leaves this server. Files are never used to train models and nobody opens them.
What we do not promise
No converter is perfect. Clean digital statements extract well in our tests; real-world files vary — photos with glare, heavy tilt, very low resolution or unusual layouts will be worse, and scanned pages are always less accurate than digital ones. That is why the preview and the balance reconciliation exist: always check them before using the numbers. This is a probe, not audited software.
Pricing
The first 10 pages per browser are free. Paid packs are one-time — pages never expire: $9 for 100 pages, $29 for 500, $79 for 2,000.