Available now
Supported inputs and limits
Each tool validates input before processing. This reference lists the current web-product formats and operational limits, which are separate from the future public API contract.
Input matrix
| Workflow | Accepted input | Current limit |
|---|---|---|
| Pasted text detection | Plain text | At least 12 words; account-specific character cap |
| URL analysis | One public HTTP or HTTPS page | 40,000 readable characters per fetched page |
| Document extraction | PDF, DOCX, TXT, MD, CSV | 20 MB; up to 120,000 extracted characters |
| Document OCR | PNG, JPG, WebP, TIFF, BMP | 20 MB |
| Image detection | JPG, JPEG, PNG, WebP | 15 MB |
| EU AI Labelizer | JPG, PNG, WebP | Up to 20 images, 25 MB each |
How document extraction chooses a path
Born-digital PDF, DOCX and text files use native extraction first. Scanned or layout-heavy material can move to the OCR path so the detector receives readable text instead of an empty document.
The extraction limit and the account character limit are applied separately. A large document may be read successfully and then shortened to the amount accepted by the selected tool.
Validation and conversion
- The server verifies the MIME type and the file signature, not only the filename extension.
- JPG, PNG and WebP images are normalized before model inference where needed.
- Password-protected or malformed documents may need to be exported again before extraction.
- Old binary
.docfiles should be converted to.docxbefore upload.