Guardian Layer Overview
Guardian Layer is Custodian Labs' privacy data transformation layer for sensitive text and documents.
Use it when your application needs to detect or de-identify sensitive information before content is sent to an AI model, stored, shared, or used in another workflow.
What You Can Do
Guardian Layer helps you:
- Detect sensitive information in text.
- Transform sensitive values into safer alternatives.
- Redact sensitive values when they should be hidden.
- Limit processing to selected sensitive data categories.
- Process plain text and supported files.
Start Here If
Use Guardian Layer when you need to:
- Prepare user input before sending it to an LLM.
- De-identify customer records, support tickets, notes, or internal documents.
- Remove or transform sensitive values before sharing data with another system.
- Build privacy-aware review, moderation, or data preparation workflows.
If you are building a conversational assistant with memory, examples, and RAG, start with the Python SDK Overview instead.
Main Concepts
These concepts appear throughout the Guardian Layer documentation.
| Concept | What it means |
|---|---|
| Detection | Finds sensitive content without changing the input. |
| De-identification | Produces safer output by changing detected sensitive content. |
| Supported entities | The sensitive data categories Guardian Layer can target, such as names, email addresses, phone numbers, and IDs. |
| Transform | Replaces sensitive values with safer alternatives while keeping the text readable. |
| Redact / masking | Hides sensitive values with visible masking marks. |
| Placeholder-style labels | Label-style replacements such as [PERSON]; treat this as implementation-dependent unless explicitly exposed by the API. |
Supported Inputs
Guardian Layer can be used with:
- Text strings.
- CSV, XLSX, and XLS files.
- DOCX files.
- PDF files with extractable text.
- TXT files.
Use the file-specific pages when integrating document workflows.