Airzom

PDFs · Office files · scans · 200 per batch

Stop retyping what a machine can read

Airzom pulls clean Markdown and real tables out of PDFs, Office files and scans. Queue up to 200 files, close the tab — the work finishes without you.

Read a documentSee pricing

200 free credits when you sign up — 50 pages. No card required.

The Airzom OCR workspace on a phone: the supplier invoice above, its extracted result below.

Not a mockup — that invoice went through the live pipeline at 4 credits a page.

Input

It reads what you already have

No converting, no flattening, no re-scanning. Every accepted format, counted and on the board.

Documents

Page structure survives the read

6
  • PDFPDF
  • DOCXDOCX
  • PPTXPPTX
  • EPUBEPUB
  • ODT
  • RTF

Spreadsheets

Cells come back as tables

2
  • XLSXXLSX
  • CSV

Images and scans

Read as a single page

8
  • JPGJPG
  • PNGPNG
  • HEICHEIC
  • TIFFTIFF
  • AVIF
  • WEBP
  • GIF
  • BMP

Text and markup

Straight through, no OCR needed

4
  • TXTTXT
  • XML
  • LaTeX
  • Jupyter notebooks

The product

Built for the hundred-invoice afternoon

The archive nobody has opened since 2019. The folder of phone photos of receipts. The supplier who still sends PDFs.

Close the tab. It keeps going.

A batch is queued and drained on the server, not in your browser. Start 200 files, walk away, come back to finished work.

A finished batch of four invoices in the Airzom OCR queue, each marked done with its page count and credit cost.

Markdown, tables and JSON

Every page comes back as clean Markdown with its tables still tables — plus the raw JSON when something downstream has to parse it.

The Airzom result panel showing the extracted invoice as rendered Markdown, with the supplier details and line-item tables intact.

Structured output, your shape

Name the fields you want and the result comes back as JSON with exactly that shape. No prompt engineering.

The Airzom response-format dialog, defining invoice_number, supplier_name, issue_date, total_amount and line_items with their types.

You pay for pages that were read

Files with no matching pages are skipped and never billed. Pages the provider fails on are refunded to your balance automatically.

4credits a page
6 with structured JSON

The alternative

Retype it, or read it

The manual path costs an afternoon per stack. This one costs credits by the page.

Retyping by hand
  • Every table rebuilt cell by cell
  • Copy-paste from a PDF flattens the columns
  • A transposed digit nobody catches until it matters
  • One afternoon per stack of scans
With Airzom
  • A 30-page, 56MB scan reads in 2.7 seconds
  • Tables come back as tables, in clean Markdown
  • 200 files queue on the server — the tab is optional
  • Pages the provider fails on are refunded automatically

Capacity

Built for stacks, not samples

The ceilings are on the product page, not in a support ticket.

2.7s
for a 30-page scan
A 56MB supplier PDF, measured against the live pipeline.
200
files per batch
Queued and drained on the server; the tab is optional.
200MB
per file
Uploads go straight to storage, not through the request.
1,000
pages per document
The provider's hard ceiling — refused before it bills, not after.

Pricing

Credits, not a subscription

Buy a pack when you have work. Nothing renews on its own, packs stack, and unused pages are never charged.

Packs
2,000 · 5,000 · 10,000 credits
Valid for
180 days from purchase
Pay by
Card in USD — or VietQR transfer in Vietnam
See the packs and prices

FAQ

Asked before the first batch

Short answers, no sales voice. Payment questions live on the pricing page.

Try it on your own worst document

200 credits is 50 pages. If it does not read your files well, you have not spent anything.

Read a document