Redact information in documents

The redacting functionality allows you to automatically detect and redact sensitive information within uploaded PDF documents. This feature helps you protect personal and confidential data when sharing or processing documents.

How detection works

Intric uses a two-step process to identify and locate sensitive information on each page of a document:

AI detection: The assistant’s selected completion model scans each page individually, identifying entities that match the categories you have selected (such as names, identification numbers, financial data, or medical information).
Coordinate mapping: Once entities are identified, Intric locates their exact positions on the PDF by matching the detected text against the document’s layout. This determines the precise bounding boxes used for redaction.

This process repeats page by page until the entire document has been analyzed. Only a single AI model is involved — the one configured on the assistant you select when starting the analysis.

Detection categories

Before running an analysis, you choose which categories of information to detect. The available categories are:

Personal details — names, email addresses, phone numbers, Swedish personal identity numbers (personnummer), personal dates (e.g. birthdays), and personal locations (e.g. home addresses).
Company names — organization names, Swedish organization numbers, and organization-associated locations.
Locations — street addresses, postal codes, cities, countries, and other geographic references.
Financial — credit card numbers, IBAN/bank account numbers, US Social Security Numbers, monetary amounts, invoice numbers, and VAT numbers.
Medical — medical information, diagnoses, and treatments.
Date/time — all dates and times in any format.
Custom categories — you can also define your own free-text categories for the AI to look for.

When to use PDF redacting

Use PDF redacting when you need to:

Remove personal or sensitive information before sharing a document.
Create redacted versions of documents for compliance purposes.
Process sensitive documents while protecting individual privacy.
Prepare documents for broader distribution within your organization.

How to redact a PDF

1. Upload your PDF to Intric

Upload a PDF file in any conversation.
Wait for Intric to process and extract the text content from the document.
Once processing is complete, click on the PDF to open it in the side panel as an artifact.

2. Configure and run the analysis

Click the settings icon (cogwheel) next to the Redact button to select which detection categories to include.
Optionally add custom categories for domain-specific information.
Select the assistant whose AI model should perform the detection.
Click Redact to start the automatic detection process.
Wait for the analysis to complete (processing time depends on the number of pages and their complexity). You can track progress via the progress bar in the viewer.

3. Review and adjust

Once analysis is complete, review the detected entities and their highlighted positions in the document.
You can manually add, remove, or adjust redaction boxes before applying the final mask.

4. Apply the mask and download

When you are satisfied with the redaction selections, apply the mask to generate the final redacted PDF.
Click the download button to save the redacted PDF to your device. In the redacted version, all marked areas are permanently removed — the underlying text is destroyed and replaced with black rectangles.

Model configuration and requirements

Assistant model: The detection is performed by the completion model configured on the assistant you select. Make sure the required AI model is enabled for your tenant before use.
Text extraction: The document must have extracted text content available (produced during the upload/ingestion pipeline). If a file lacks page-level text, you will be asked to re-upload it.
File size limit: PDFs must be under the configured maximum file size for masking (default: 3 MB).

Availability

PDF redacting is available to all users whenever a PDF with extracted text content is uploaded in a conversation. The redacting button will appear automatically once Intric has successfully processed your document.

Best practices

Review your work: Always review the redacted PDF to ensure all sensitive information has been properly covered. The AI may miss entities or flag false positives.
Format matters: Redacting works best on text-based PDFs with clear, readable content. Scanned documents depend on the quality of text extraction during upload.
Quality counts: For best results, upload high-quality PDFs where text is easily recognizable.
Permanent redaction: The mask is applied by rasterizing pages and drawing black rectangles over detected areas. This permanently removes the underlying text — it cannot be recovered from the redacted file.
Choose categories carefully: Selecting only the categories relevant to your use case reduces noise and improves detection accuracy.