Repository navigation
Troubleshooting
Start with the smallest repeatable case: one synthetic or non-sensitive PDF, a short page range, and only the categories needed. Record the product version, platform, settings, and exact error before changing several options at once.
- Confirm that the expected category is selected.
- Reduce the minimum width and height if the element is small.
- Increase DPI for a low-resolution page.
- Lower the confidence threshold slightly. A lower threshold can add false positives.
- Enable OCR if the missing content is text in a scan.
Feature callout: Table crops are returned as individual cards. If the table row is absent, check the Table category and thresholds; if the card exists but needs machine-readable content, enable structured table extraction.
- Try one or two pages first.
- Lower the image resolution.
- Select fewer categories.
- Split a large PDF into ranges such as 1–50 and 51–100.
- Re-enter the password for a protected PDF.
- Check whether the monthly page allowance has been reached.
If the same page fails with default options, preserve the exact page number and sanitized error text for a bug report. Do not attach the source PDF if it contains private information.
Raise the confidence threshold in small increments and select only the needed categories. Increase the minimum dimensions to remove tiny detections. If several cards describe the same area, return the IoU threshold to its default before trying a small adjustment.
- Confirm that OCR is enabled and that the mode matches the task.
- Use page mode for Advanced OCR.
- Increase DPI for faint or very small text.
- For Standard OCR, verify that the requested Tesseract language is installed.
- For complex columns, irregular tables, or handwriting, try Advanced OCR when the plan supports it.
- Test a short page range before processing the full document.
OCR failure does not necessarily stop element detection. A result can contain image crops even when no transcription is available.
In the web app, Advanced OCR requires a supported plan and a configured service. A temporary 503 response can indicate that the service dependency is unavailable.
In the desktop app:
- Open Settings and check the model status.
- Confirm that there is enough free disk space.
- Retry on a stable internet connection.
- Keep the app open until the download and verification finish.
- Restart the app after a completed model download if the status does not refresh.
Include the app version, operating system, processor/GPU, and sanitized error in a public bug report. Keep license and device identifiers private.
Cloud results expire after the retention period shown for the plan. If Store File was disabled, the files were available only in the immediate response. An extraction record may remain visible after managed files expire. Rerun the extraction to recreate files that were not downloaded in time.
| Status | Check |
|---|---|
| 400 | Form field spelling, page range, password, and Advanced OCR page mode |
| 401 | Authorization header and whether the token was revoked |
| 403 | Page allowance and plan access for OCR or batch extraction |
| 415 | PDF filename and content type |
| 500 | Smaller page range and a later retry with backoff |
| 503 | Advanced OCR or another service dependency may be temporarily unavailable |
See the API guide for request examples.
Run the server command in a terminal, validate the client configuration, and fully restart the client. Confirm that PDF_EXTRACTOR_API_KEY is inside the server's environment configuration. See the MCP guide for complete checks.
- Product area: web, desktop, API, MCP, or documentation
- App/client version, operating system, and browser where relevant
- Exact steps and extraction options
- Expected and actual behavior
- Page count and whether the file is scanned or password-protected
- Sanitized error messages or screenshots
- Whether the problem reproduces with a synthetic PDF
Report a bug after searching existing issues. If the behavior is a missing capability instead of a defect, request an enhancement.
GitHub Issues are public. Never attach private PDFs or include document content, personal information, passwords, API tokens, license keys, device identifiers, local paths, or billing details. Use private support when the problem cannot be safely reproduced or sanitized.
Last reviewed: August 21, 2026.