Docs Projects

Document Extraction

Updated 45 minutes ago 4 min read 76 views

What is document extraction?

Extraction uses AI to read your uploaded files and output clean, structured data (JSON, CSV, XLSX, Markdown, XML, or SQL).

It supports:

  • PDF documents (contracts, invoices, reports, forms)

  • Word documents (.docx)

  • Excel spreadsheets (.xlsx, .csv)

  • Images (JPG, PNG — useful for scanned documents)

  • Text files and Markdown

How to run an extraction:

  1. Click + New → New Pipeline

  2. Choose Document Extraction

  3. Upload your files (drag and drop, or click to browse)

  4. Choose your output format (JSON recommended for structured data)

  5. Optionally define a custom schema — tell the AI exactly what fields to extract

  6. Click Extract

The job runs in the background. You will see a live progress bar. When done, the extracted data appears in the project detail page.

Custom schemas:

By default, the AI decides what fields to extract.

To control this:

  1. In the extraction setup, click Define Schema

  2. Add field names and descriptions (e.g. "invoice_number: The invoice reference number at the top of the document")

  3. The AI will populate exactly those fields from each document

Custom schemas are saved as templates you can reuse across projects.

Processing modes:

Small batches process one at a time for fast feedback. Larger batches switch to parallel processing automatically, at a discounted rate — there's no mode to choose, it happens on its own.

What happens after extraction?

Your extracted data file is stored securely in your project. From the project detail page you can:

  • Preview the data inline

  • Download it (JSON, CSV, XLSX, XML, SQL, Markdown)

  • Use it as input for a report

  • Auto-export it to your connected destination if confidence is high enough

Was this page helpful?