PII Masker for AI Prompts

What the PII Masker Detects

The masker runs two cooperating layers, both entirely in your browser. The deterministic regex layer runs first and catches structured PII with high precision. The optional on-device NER layer then catches unstructured PII like personal names that no pure regex can reliably recognise.

How to Use the PII Masker

1. Paste the prompt you intend to send to an AI assistant into the input area.
2. Pick a detection mode: Regex + NER for maximum coverage, or Regex only for instant deterministic masking.
3. Click Mask PII. Detected items are replaced with stable placeholders such as [NAME_1], [EMAIL_1], and [PHONE_1].
4. Click Copy masked text and paste it into ChatGPT, Claude, Gemini, or any other LLM tool.
5. When the AI replies, open the Restore panel, paste the response and the mapping JSON, and recover the real values locally.
6. Click Reset when you are done — that wipes the mapping from sessionStorage so it never persists.

Key Features

• Two-layer detection — deterministic regex (with Luhn and mod-97 validation) plus on-device BERT NER for unstructured names, orgs, and locations.
• Stable placeholders — repeated values get the same token, so the LLM can reason about them as proper nouns.
• Restoration round-trip — paste the AI reply back into the Restore panel along with the mapping JSON to recover real values locally.
• sessionStorage only — the mapping is wiped when the tab closes; nothing ever lands in localStorage or hits a backend.
• Offline after first load — the ~110 MB NER model is cached in IndexedDB on first use; the regex layer works offline immediately.
• Regex-only fallback — skip the model download entirely if you only need to mask structured PII.

When to Use a PII Masker

• Drafting customer-support replies — strip customer names, emails, and order numbers before asking ChatGPT to rephrase the reply.
• Sharing log snippets — remove IP addresses, internal hostnames, and user identifiers before pasting a stack trace into Claude.
• Polishing job-application materials — let an AI critique your resume content without exposing the recruiter's contact details.
• Summarising private emails — keep sender / recipient identities out of an AI summarisation prompt.
• GDPR / HIPAA / SOC 2 reviews — provide an auditable in-browser path for any team member who needs to use an external LLM with internal content.

Why On-Device Detection Matters

Cloud-based PII redaction services typically require uploading the very text you are trying to protect. That defeats the threat model. This tool uses Transformers.js and a Web Worker to run a quantised BERT NER model entirely on your device — the prompt never crosses the network, and the restoration mapping stays in your tab's sessionStorage. Pair it with the upcoming Prompt Secret Scrubber for AWS / GCP / OpenAI API keys, and the Document PII Redactor for full .txt / .md / .docx / .pdf files.

Validation Details

• Credit cards. Numbers between 13 and 19 digits are accepted only if they pass the Luhn (mod-10) checksum, eliminating most false positives on order numbers and reference codes.
• IBANs. Validated with the ISO 13616 mod-97 algorithm — a country prefix and matching checksum digits are required.
• IPv4. Each octet is range-checked (0–255), so timestamp-like sequences are rejected.
• NER spans. Aggregated with the Transformers.js simple strategy so multi-token entities (e.g. "Jane Doe", "Bank of America") are merged into a single placeholder.