Hard limits
- PDFs must be under 500MB and 30 pages or fewer. Files that exceed either limit will not process.
- Password-protected PDFs cannot be parsed. If your vendor sends encrypted files, ask them to remove the password before sending — or build a separate decryption step before your Parabola flow.
Best practices
Parse only the pages you need
Parse only the pages you need
The more pages included, the longer the run. If your data appears on consistent pages across documents, use Page Filtering in the advanced settings to limit what the AI reads. Fewer pages means faster results — especially at scale.
One table per step
One table per step
Each Extract from email step can only extract one table at a time. If you need to pull data from multiple distinct tables within the same PDF, you’ll need a separate step for each table.
Always audit your output before going live
Always audit your output before going live
AI extraction is powerful but not infallible — especially with complex or inconsistently formatted documents. We strongly recommend reviewing the step’s output in Parabola with a few representative sample files before connecting it to a production destination.
Consolidate results across multiple runs
Consolidate results across multiple runs
One file is processed per run. If you’re parsing many PDFs over time, connect your flow to a Parabola Table or an external sheet so results accumulate rather than being overwritten. Without a persistent destination, each new run replaces the previous output.
Use the file URL for audit trails
Use the file URL for audit trails
When you enable Email content + attachment and pull in the
file_url, you get a shareable link to the original PDF. Include this in your destination data so you can always trace any row back to its source document — useful for invoice audits, dispute resolution, and compliance workflows.Use field-level instructions for better results
Use field-level instructions for better results
If a specific column or key isn’t returning the right data, the fix usually lives at the field level — not in the general Fine-tuning section. Add clear descriptions, example values, and targeted instructions directly on the column or key that’s underperforming. The more specific the guidance, the better the AI will do.
A note on handwritten documents
Parabola can extract handwritten data, but results depend on legibility. If you’re working with handwritten PDFs:- Switch the text parsing approach to OCR in the advanced settings
- Add clear descriptions and example values for every column and key
- Start with a small batch of sample documents and review the output before running at scale
Quick reference
| Constraint | Limit |
|---|---|
| Max file size | 500MB |
| Max pages | 30 |
| Max emails in queue | 1,000 |
| Files processed per run | 1 |
| Tables extractable per step | 1 |
| Password-protected PDFs | Not supported |