Summary
A browser-based tool that replaces sensitive text with labels before it is sent to AI assistants.
About
Clean before AI adds a local privacy layer to workflows that use external AI assistants by replacing detected personal and organizational data with labels before submission.
Clean before AI detects names, national ID numbers, IBANs, addresses, phone numbers, and company names in text, then replaces them with labels such as [PERSON_1]. The labels are sent to ChatGPT, Claude, or Gemini, and the corresponding real data is restored in the response. Processing runs locally in the browser without a server or account and works offline. Detection combines checksum-validated rules with a small local AI model and supports 17 European languages. A Chrome extension can clean text as it is pasted. The product is free.
Key Features
Detects names, national ID numbers, IBANs, addresses, phone numbers, and company names.
Replaces detected values with labels and restores the corresponding data in AI responses.
Processes text in the browser without a server or account and supports offline use.
Combines checksum-validated rules with a small local AI model.
Supports 17 European languages and offers a Chrome extension that cleans text as it is pasted.
Real-World Use Cases
Privacy-conscious AI-assisted work
Reduce exposure of personal or organizational identifiers when employees paste text into ChatGPT, Claude, or Gemini.
Handling customer communications
Mask contact details and other detected identifiers in customer messages before using an AI assistant to draft or summarize responses.
Reviewing documents with AI
Replace detected identifiers in text before submitting documents or excerpts for AI-assisted analysis.
Business Opportunities
Safer employee use of public AI assistants
Problem: Employees may include personal or organizational identifiers when using external AI assistants, creating privacy and data-handling concerns.
Solution: Use local text cleaning to replace detected identifiers with labels before content is submitted to supported AI assistants.
Estimated ROI: Potentially positive through reduced manual redaction effort and lower exposure of detected identifiers; actual returns depend on usage and organizational risk.
Execution: Introduce the browser workflow and Chrome extension for teams that permit external AI use, and establish guidance on which information may be submitted.
Privacy-aware AI workflows in regulated sectors
Problem: Professionals in sectors handling sensitive records may avoid AI assistance or spend time manually removing identifiers before use.
Solution: Apply browser-based detection and replacement to supported identifiers before text reaches an AI assistant.
Estimated ROI: Potentially moderate where it reduces repetitive redaction and enables limited AI-assisted work; no compliance outcome or risk elimination is established by the provided information.
Execution: Evaluate detection coverage and restoration behavior against representative text, then define approved use cases and review requirements before broader adoption.