Workflow
How SafeToAI works
The product job is narrow on purpose: make legal materials safe enough to paste into external AI — without becoming another AI lawyer.
1. Parse on your device
When you choose a DOCX, text-based PDF, TXT, Markdown, or CSV file, SafeToAI extracts plain text in the browser. DOCX uses Mammoth; PDFs use PDF.js against the text layer.
Scanned image-only PDFs are not supported in v1 — there is no OCR yet. If extraction fails, convert to DOCX/TXT first or wait for the OCR roadmap.
2. Detect and review
A rules engine flags emails, phones, SSN/ITIN-like IDs, credit cards, organization-like strings, person-name heuristics, coarse addresses, dates, currency amounts, and case-like IDs.
You can disable false positives, remove spans, or add an exact manual match. Nothing is “auto-sent” to an LLM for detection on the default path.
3. Choose a redaction mode
Pseudonym mode replaces each unique value with a stable token (for example [PERSON_01], [ORG_02]) so models can still track parties and obligations.
Asterisk mode masks characters when you need stronger visual concealment and do not care about AI readability.
4. Export for AI — then optionally restore
Download Markdown and paste into ChatGPT, Claude, or an internal assistant.
Download mapping.json and later restore originals into AI-edited text locally — still without uploading the matter file to SafeToAI.
What we deliberately do not claim
- Not legal advice and not a substitute for counsel review
- Not a guarantee of zero residual PII — you must review detections
- Not certified against a specific bar or jurisdiction rule set