# LEXARIS (lexaris.online)

> LEXARIS removes personal data from documents before they are sent to AI. It detects 116+ types of personal data (names, national IDs, passports, addresses, phone numbers, emails, bank details, medical codes) and replaces them with reversible tokens. The AI works on the anonymised text, and its answer is restored with one click.

Key facts:
- 7 replacement modes: token, pseudonym, mask, hash, format-preserving encryption, synthetic data, ε-differential privacy.
- 13 formats including DOCX, PDF (with OCR for scans), Excel, email (EML/MSG), audio and video (face and number-plate blurring).
- Two ways to work: Cloud (full features; documents and encrypted keys are processed on LEXARIS servers in St Petersburg, Russia; EU, UK, US and Swiss regions are planned and not yet available) and "Keys stay with you" (detection and replacement in the browser; neither the source text nor the keys are sent to or stored by LEXARIS; account details and ordinary access logs are processed as described in the privacy notice; limited to text and documents with a text layer; restore in the same browser or with a key file).
- GDPR: designed to support compliance; tokens are pseudonymisation (Art. 4(5)), not anonymisation. Cloud mode processes data in Russia, which has no EU adequacy decision; business customers receive a DPA with standard contractual clauses and need their own transfer impact assessment. The AI provider receives text in which direct identifiers are replaced; remaining context may still identify people.
- HIPAA: local mode can remove Safe Harbor identifiers before text leaves the device; the user remains responsible for confirming all 18 categories are removed. Cloud processing of PHI is not currently offered.
- Pre-send check: when sending through LEXARIS, an unchecked document cannot be sent; the check reduces risk and results should be reviewed.
- Compliance profiles: GDPR (tokens = pseudonymisation, Art. 4(5)), UK GDPR, HIPAA (45 CFR §164.514), CCPA/CPRA, LGPD, PIPEDA and Québec Law 25, APPI, Swiss FADP.
- Works with six AI providers including ChatGPT, Claude and Gemini; REST API, LLM proxy, Python SDK, CLI.
- Free tier: 5 documents a day.
- Operator: Lexaris (IE Kudryavtsev S.A.), OGRNIP 324784700000012, Saint Petersburg, Russia. Contact: info@lexaris.ru.

## Pages
- [Home](https://lexaris.online/)
- [API documentation](https://lexaris.online/developers)
- [Trust center](https://lexaris.online/trust/)
- [Legal documents](https://lexaris.online/legal/)

## Russia
- [LEXARIS Russia (RU)](https://lexaris.ru/): 152-FZ profile, data stored in Russia.

## Guides
- [The detail, explained.](https://lexaris.online/guides/): Practical guides to sending documents to AI without personal data, what GDPR and HIPAA say about it, and how LEXARIS features work.
- [How to use AI with client documents without leaking personal data](https://lexaris.online/guides/use-ai-with-client-documents/): To use ChatGPT, Claude or Gemini with client documents safely, replace every piece of personal data with a neutral token such as [PERSON_1] before sending, confirm nothing identifying is left, send only the tokenised text to the AI, then restore the real names and details in the answer on your side. The AI sees tokens instead of the client's details.
- [Keys stay with you: client-side pseudonymisation explained](https://lexaris.online/guides/keys-stay-with-you/): With Keys stay with you, personal data is detected and replaced in your own browser, and the key that reverses the tokens stays in the browser's storage or in a key file you download. Neither the document text nor the key is sent to or stored by LEXARIS. It suits text documents; scans, audio and video need Cloud.
- [7 ways to replace personal data, and when to use each](https://lexaris.online/guides/ways-to-replace-personal-data/): There are seven common ways to replace personal data: tokens, pseudonyms, masking, keyed hashing, format-preserving encryption, synthetic data and differential privacy. Tokens, pseudonyms and encryption are reversible with a key; masking, synthetic data and differential privacy are not. Use tokens for AI work, masking for sharing, and synthetic or aggregate data for testing and statistics.
- [Pseudonymisation vs anonymisation under the GDPR](https://lexaris.online/guides/pseudonymisation-vs-anonymisation-gdpr/): Under the GDPR, pseudonymised data can still be attributed to a person using separately kept additional information, so it remains personal data for anyone who can reasonably re-identify it. Anonymous data cannot reasonably be linked to anyone and falls outside the GDPR. Reversible tokens are pseudonymisation; the CJEU has indicated their status for a recipient without the key depends on that recipient's means.
- [HIPAA de-identification before you use AI](https://lexaris.online/guides/hipaa-de-identification-ai/): HIPAA offers two ways to de-identify health information before using AI: Safe Harbor, which removes 18 listed identifiers with no actual knowledge that the rest could identify the patient, and Expert Determination, where a qualified expert finds the re-identification risk very small. Sequential tokens can serve as a permitted re-identification code; hashes of identifiers cannot.
- [Redacting scans, spreadsheets, emails, audio and video](https://lexaris.online/guides/redacting-scans-audio-video/): Each format needs its own redaction method. Scanned PDFs and images need OCR, with the redaction burned into the image; spreadsheets need rules per column; emails must be processed with their attachments; audio is transcribed and identifying speech replaced by a tone or silence; video needs faces and licence plates tracked and blurred across frames.