Built for GDPR, HIPAA and CCPA · Keys can stay in your browser
Protect the data. Keep the meaning.
Take personal data out of a document before you send it to AI, then put it back into the answer with one click.116+ types of personal data are replaced with reversible tokens before any text reaches ChatGPT, Claude or Gemini. In Cloud mode, documents and keys are processed on our servers in St Petersburg, Russia; EU, UK, US and Swiss regions are planned and not yet available. In “Keys stay with you” mode they never leave your browser.
●This demo runs entirely in your browser. Nothing is sent anywhere. The product adds neural entity recognition and 116+ data types.
The Client [PERSON_1], passport [PASSPORT_1], residing at [ADDRESS_1], phone [PHONE_1], IBAN [IBAN_1], and the Provider, represented by [PERSON_2], agree as follows…
Pre-send check
✓Check passed
Residual re-identification risk
low
Person2
Passport1
Address1
Phone · IBAN2
116
Data typesNER + rules116+ types
⇄
TokensReversible
🔑
KeysRe-identification keyWith you
✓
Pre-send checkGO / NO-GO
§
ProfilesArt. 4(5) · §164.514GDPR · HIPAA · CCPA
🛡
SecurityAES-256 · audit log
Send a contract to AI and you also send your client’s name, passport, phone and bank details. With LEXARIS, only tokens.
How it works
The document is full of personal data.Names, passport, address, phone, bank details, email.
LEXARIS swaps it for tokens.The keys to those tokens stay with you.
The AI only sees tokens.ChatGPT, Claude, Gemini: any model works on the anonymised text.
The answer comes back with your data.One click, and the tokens become names and numbers again.
SERVICES AGREEMENT No. 17
London3 March 2026
The Client John Smith[PERSON_1], passport 533 401 872[PASSPORT_1], residing at 221 Baker Street[ADDRESS_1], phone +44 20 7946 0958[PHONE_1], IBAN GB29 NWBK 6016 1331 9268 19[IBAN_1], email john.smith@example.com[EMAIL_1], and the Provider agree as follows…
AI answer
Draft letter: Dear John Smith[PERSON_1], please settle the outstanding invoice within 14 days. Kindly reply to 221 Baker Street[ADDRESS_1].
PPerson
John Smith→[PERSON_1]
Names, including initials and nicknames.
NER; variants are linked to one token across the document.
IDPassport
533 401 872→[PASSPORT_1]
Passports, IDs, driving licences.
Format and checksum validation per country.
AAddress
221 Baker Street→[ADDRESS_1]
Full or partial addresses.
Compound addresses, postcodes, geo-coordinates.
☎Phone
+44 20 7946 0958→[PHONE_1]
Any phone format.
International formats normalised.
€Bank details
GB29 NWBK …→[IBAN_1]
IBANs, cards, account numbers.
IBAN and card checksums; tax and social-security IDs.
@Email
john.smith@example.com→[EMAIL_1]
Emails, handles, profile links.
Identical values share one token across the document.
Keys stay with youin your browser or a key file
Where the keys live
Your keys. Not even we hold them.
Choose where the keys to your tokens are kept: in our cloud in St Petersburg, Russia, or only in your browser. EU, UK, US and Swiss regions are planned and not yet available.
Two ways to work. In "Keys stay with you" mode, detection and replacement run on the device and the re-identification table stays with the user. LEXARIS receives neither the document content nor the keys, so it is not a processor of that data. Account details and ordinary access logs are processed as described in the privacy notice.
LEXARIS Cloud
Every feature
Documents are processed on our servers in St Petersburg, Russia. Keys are stored there too, encrypted. EU, UK, US and Swiss regions are planned and not yet available.
Document→St Petersburg→Tokens → AI
116+ data types, neural recognition
Scans (OCR), audio and video
History, team workspaces, audit log
Restore from any device
Processor terms: GDPR Art. 28 DPA with standard contractual clauses. Hosting in Russia (no EU adequacy decision); EU, UK, US and Swiss regions planned, not yet available. No cloud processing of PHI.
Keys stay with you
We store neither your data nor your keys
Detection and replacement happen in your browser. The key stays on your device, or in a file you download.
Document→Your browser→Tokens → AI
Your document text and keys never leave your device
Fewer residency questions: your data never reaches us
Text and documents with a text layer only
No scans, audio or video
Restore in the same browser or with the key file
The key is kept in browser storage (IndexedDB) or a file, never in cookies, so it is never sent to our servers
116+types of personal data
From names to IBANs.
Passports, national IDs, bank details, addresses, medical codes and more, across many countries.
National identifiers and tax IDs; IBAN and card checksums; 200+ drug names and ICD-10 codes; 12+ secret types such as AWS and GitHub keys.
Replacement modes
Seven ways to replace data.
Pick the one that fits the job, from working with AI to publishing and statistics.
Reversible:yes, via the lookup tableUse it for: The same value always gives the same hash, so you can match records, count unique customers and join exports without revealing the data. Need the values back? Keep the lookup table.
Legal status: HMAC-SHA-256 with a secret key. The hash itself is one-way: the original cannot be computed from it, and without the key short values (phone numbers, IDs) cannot be brute-forced either. Values can only be restored from a “hash → value” lookup table. If you keep one (in our cloud or in “Keys stay with you” mode) this is pseudonymisation and the data remains personal data. Without a lookup table the hash cannot be reversed, but anyone holding the secret key can test candidate values, so it remains pseudonymisation until the key is destroyed. Not a HIPAA re-identification code.
Reversible:noUse it for: Statistics and reports with a mathematical privacy guarantee.
Legal status: ε-differential privacy: calibrated noise; individual records are not disclosed.
13 formats
Any format. One tool.
Supply Agreement No. 214.docxSaved
FileHomeInsertLayoutReferencesReview
Times New Roman12BIU
SUPPLY AGREEMENT No. 214
London12 March 2026
This Agreement is made between Northwind Trading Ltd, acting through its director Sarah Mitchell[PERSON_1] (the “Supplier”), and John Smith[PERSON_2], passport 533 401 872[PASSPORT_1], residing at 221 Baker Street, London NW1 6XE[ADDRESS_1], phone +44 20 7946 0958[PHONE_1] (the “Buyer”).
1. SUBJECT MATTER
1.1. The Supplier shall deliver the Goods to the Buyer, and the Buyer shall accept and pay for them on the terms set out in this Agreement.
1.2. Payment shall be made to GB29 NWBK 6016 1331 9268 19[IBAN_1] within five business days.
✓ Anonymised
AI answer · sees only tokensAnswer with your data
The Buyer [PERSON_2]John Smith pays to [IBAN_1]GB29 NWBK 6016 1331 9268 19 within 5 business days. ⚠ No delivery date — add clause 1.3.
DOCX · contractsText and layout stay exactly as they wereStyles, tables, headers, footers and comments preserved
discharge_summary_scan.pdf1 / 3100%OCR
St Aidan’s General Hospital · Department of Medicine Riverside Road, Leeds
DISCHARGE SUMMARY
Patient:Emily Clarke
Date of birth:14/05/1987
Address:12 Park Lane, Leeds LS1 4AB
NHS number:943 476 5919
Phone:07700 900461
Diagnosis:J06.9 — acute URTI
Admitted:02/03 – 09/03/2026
Consultant ____________
✓ Redacted on the scan
AI answer · sees only tokensAnswer with your data
[PERSON_1]Emily Clarke was treated for J06.9. A fit note can be issued.
PDF and scansFinds personal data even in a photo of a pageOCR plus redaction burned into the image, not just the text layer
staff_2026.xlsx
B3fxJohn Smith
A
B
C
D
E
1
Name
NI number
Phone
Salary, £
Team
2
John Smith[PERSON_1]
QQ 12 34 56 CQQ ** ** ** C
07700 90046107*** ***461
52,000
Sales
3
Emma Clarke[PERSON_2]
QQ 98 76 54 AQQ ** ** ** A
07700 90012307*** ***123
48,500
Logistics
4
Mark Reid[PERSON_3]
QQ 55 44 33 DQQ ** ** ** D
07700 90098707*** ***987
61,000
Legal
5
Priya Shah[PERSON_4]
QQ 11 22 33 BQQ ** ** ** B
07700 90055507*** ***555
57,200
Finance
6
Tom Walsh[PERSON_5]
QQ 77 88 99 AQQ ** ** ** A
07700 90031407*** ***314
66,000
IT
StaffSalaries+
✓ Per-column rules
AI answer · sees only tokensAnswer with your data
Highest salary: [PERSON_5]Tom Walsh (IT, £66,000). Company average: £56,940.
Excel and CSVPersonal columns hidden, numbers kept for analysisPer-column rules: mask, token or hash; totals and formulas untouched
Mail · Inbox
John SmithCase documents
County CourtNotice of hearing…
AccountsInvoice #58, 10 March
Emma ClarkePower of attorney
Case documents — claim K40CL123
JS
John Smith[PERSON_1]john.smith@example.com[EMAIL_1] → counsel@lawfirm.example
Hello, please find my passport scan and the signed authority attached. My new address is 14 Queen St, Bristol BS1 4DJ[ADDRESS_1]. Best number: 07700 900461[PHONE_1].
Draft reply: “[PERSON_1]John Smith, documents received. We will call you on [PHONE_1]07700 900461 before 6 pm today.”
Email EML · MSGThe message and its attachments in one goHeaders, body and attachments in any supported format
Client call.mp300:42 / 02:14
● transcribing
00:03 · CHi, my name is John Smith[PERSON_1].
00:09 · CYou can call me back on 07700 900461[PHONE_1].
00:15 · CMy date of birth is 14 May 1987[DOB_1].
✓ Speech anonymised
AI answer · sees only tokensAnswer with your data
Call summary: [PERSON_1]John Smith asks for a call back on [PHONE_1]07700 900461 about the contract.
AudioCall transcribed, personal details bleepedTranscription plus replacing speech segments with a tone or silence
▶00:0700:242 faces · 3 plates
AI answer · sees only tokensAnswer with your data
18:42:07 — a red car [PLATE_1]LK19 XRT drove towards the exit; parked next to it: [PLATE_2]BD51 SMR. One pedestrian, face hidden.
VideoFaces and number plates blurredFaces and licence plates tracked across every frame, pixelated or blurred
Compliance profiles
Choose your regulation. We apply its rules.
Each profile sets which data counts as personal, which mode to use, and where the keys should live.
Profiles set entity types, the default replacement mode and the key location for each regime. They also document the legal status of the output: pseudonymised or anonymised.
EU
GDPR — EU / EEA
European Union, Iceland, Liechtenstein, Norway
Tokens count as pseudonymised data: still personal data for you, but the AI provider receives text without direct identifiers. Context left in the text may still point to someone, so review it before sending. In 'Keys stay with you' mode your document text and keys never reach us.
Tokenisation with a separately held key is pseudonymisation (Art. 4(5)) and remains personal data for the controller holding the key (Recital 26). Whether it is personal data for a recipient without the key depends on that recipient's reasonable means of re-identification (EDPB Guidelines 01/2025 (version for public consultation); CJEU C-413/23 P, EDPS v SRB). Under EDPS v SRB, your own obligations are assessed from your perspective as the controller holding the key. LEXARIS Cloud currently processes data in Russia, which has no EU adequacy decision. Pseudonymisation is a recognised safeguard under Arts. 25, 32 and 89.
Recommended: Token for AI work; synthetic or differential privacy for analyticsKeys: Your browser (recommended), or our cloud in Russia (EU region planned)
References
GDPR Art. 4(5) — pseudonymisation; Recital 26 — anonymous information
Arts. 25, 32 — data protection by design; security of processing
Arts. 44–49 — transfers to AI providers outside the EEA
Art. 28 — processor terms (cloud mode)
EDPB Guidelines 01/2025 on pseudonymisation (version for public consultation)
EU AI Act Art. 10(5) — special-category data for bias detection in high-risk AI systems, subject to safeguards incl. pseudonymisation
UK
UK GDPR & Data Protection Act 2018
United Kingdom
Same logic as the EU: tokens are pseudonymised data, and only you can turn them back into names.
Pseudonymised data remains personal data for the holder of the key (UK GDPR Art. 4(5)); the ICO's anonymisation and pseudonymisation guidance applies the 'motivated intruder' test to recipients.
Recommended: Token for AI work; mask for sharingKeys: Your browser (recommended), or our cloud in Russia (UK region planned)
References
UK GDPR Art. 4(5), Arts. 25, 32
Data Protection Act 2018
ICO guidance on anonymisation and pseudonymisation
US
HIPAA — US health data
Covered entities and business associates in the United States
Remove patient identifiers before protected health information reaches an AI tool. Only you can re-link the tokens. You still need to check that every identifier, including dates, is gone.
De-identification under 45 CFR §164.514(b) by Safe Harbor (removal of the 18 identifiers, no actual knowledge of re-identifiability) or Expert Determination. Tokens not derived from PHI and kept confidential can serve as a re-identification code under §164.514(c). Local mode lets you remove Safe Harbor identifiers before text leaves your device; you remain responsible for confirming all 18 categories, including dates other than the year, are removed. Cloud processing of PHI is not currently offered.
Recommended: Token or mask; local mode for PHIKeys: Your browser (cloud processing of PHI is not offered)
References
45 CFR §164.514(a)–(c) — de-identification and re-identification codes
45 CFR §164.502(d)
45 CFR §164.504(e) — business associate agreements
CA
CCPA / CPRA — California
Businesses handling California residents' data
Tokenised data is 'pseudonymized' under California law: still personal information for you, and the AI provider receives no direct identifiers; 'deidentified' status additionally needs the CCPA's public-commitment and contract conditions.
Cal. Civ. Code §1798.140 distinguishes pseudonymized from deidentified information; deidentification additionally requires reasonable technical measures, a public commitment not to re-identify, and contractual controls on recipients.
Recommended: Token for AI work; synthetic for testingKeys: Your browser (recommended), or our cloud in Russia (US region planned)
Cal. Civ. Code §1798.100 ff. — business obligations
BR
LGPD — Brazil
Brazil
Reversible tokens are pseudonymisation; truly anonymised data falls outside the LGPD.
Anonymised data is not personal data unless the process can be reversed (art. 12); pseudonymisation is defined in art. 13 §4. Reversible tokens therefore remain personal data for the key holder.
Recommended: Token for AI work; mask or synthetic for anonymisationKeys: Your browser (recommended), or our cloud in Russia (Brazil region planned)
References
LGPD art. 5 III, XI — anonymised data, anonymisation
LGPD art. 12 — anonymised data
LGPD art. 13 §4 — pseudonymisation
CA
PIPEDA & Québec Law 25 — Canada
Canada
Québec law distinguishes de-identified from anonymised information; tokens are de-identified and only you can reverse them.
Québec's private-sector Act (as amended by Law 25) separates de-identified information (s. 12: still personal information) from anonymised information (s. 23: irreversible, under regulatory criteria). PIPEDA's safeguards principle applies to the key.
Recommended: Token for AI workKeys: Your browser (recommended), or our cloud in Russia (Canada region planned)
References
Québec Act respecting the protection of personal information in the private sector, s. 12 — de-identified information
Québec Act respecting the protection of personal information in the private sector, s. 23 — anonymised information
PIPEDA Schedule 1, Principle 4.7 — safeguards
JP
APPI — Japan
Japan
Japan has special rules for 'pseudonymously processed' and 'anonymously processed' information; tokens are the former. Pseudonymously processed information generally cannot be provided to third parties, so AI use must fit an entrustment arrangement.
Pseudonymously processed information (arts. 41–42) may be used internally with restrictions and generally cannot be provided to third parties (art. 41(6)), so disclosure to an AI provider must fit an entrustment arrangement (art. 27(5)(i)); anonymously processed information (arts. 43–46) is subject to disclosure and non-re-identification duties.
Recommended: Token for internal AI workKeys: Your browser
References
APPI arts. 41–42 — pseudonymously processed information
APPI arts. 43–46 — anonymously processed information
CH
revFADP — Switzerland
Switzerland
Anonymised data is outside the FADP; tokens remain personal data for you, and only you hold the key.
The revised FADP (in force 1 September 2023) applies to data relating to identified or identifiable persons; privacy by design and by default (Art. 7) favours pseudonymisation before disclosure to AI providers.
Recommended: Token for AI workKeys: Your browser (recommended), or our cloud in Russia (Switzerland region planned)
References
FADP Art. 5 — definitions
FADP Art. 7 — data protection by design and by default
FADP Art. 16 — disclosure abroad
Russia: 152-FZ profile, data in Russia — lexaris.ru
Up to €20m or 4%
The maximum GDPR fine, whichever is higher.
GDPR Art. 83(5): up to €20 million or 4% of total worldwide annual turnover, whichever is higher.
0 bytes
of document content reach us in "Keys stay with you" mode. Your account details and ordinary access logs are processed as described in the privacy notice.
0 bytes of document content. Detection runs in your browser; neither the source text nor the re-identification table is sent to us. Your account details and ordinary access logs are processed as described in the privacy notice.
GO / NO-GO
When you send through LEXARIS, an unchecked document cannot be sent. The check reduces risk; review the result before sharing.
Pre-send check with a residual re-identification risk score. When you send through LEXARIS, an unchecked document cannot be sent. The check reduces risk; review the result before sharing.
Summaries, analysis and drafts with AI, without disclosing client data.
Presets for contracts, M&A and court documents.
Less risk to privilege
In "Keys stay with you" mode, client data never leaves the lawyer’s computer.
Key in the browser or a file.
People data
Personnel files, payroll and applications, anonymised before analytics or AI.
HR preset.
Evidence for audits
An audit log of every anonymisation, exportable as PDF or CSV.
Ready for your DPO.
Patient confidentiality
Records, prescriptions and discharge notes, analysed without exposing patients.
HIPAA profile, 200+ drug names, ICD-10 codes.
Scans and images
OCR and redaction directly on the image.
Healthcare preset.
API in minutes
REST API, SDK, CLI, webhooks and batch processing.
pip install anon-client
RAG and secrets
Clean data for RAG pipelines, and find AWS and GitHub keys in code and logs.
12+ secret types.
For counsel and DPOs
The legal status of tokens.
Pseudonymisation vs anonymisation, controller and processor roles, transfers, health data and a policy clause you can adopt.
GDPR Art. 4(5) · Recital 26
Tokens are pseudonymisation
Replacing identifiers with tokens while the key is held separately is pseudonymisation, not anonymisation.
For the controller holding the key, the data remains personal data.
For a recipient without the key, the question is whether it has reasonable means to re-identify (EDPB Guidelines 01/2025 (version for public consultation); CJEU C‑413/23 P, EDPS v SRB).
Irreversible modes (mask, synthetic, differential privacy) can reach anonymisation, subject to a re-identification risk assessment.
GDPR Arts. 28, 44–49
Roles and transfers
Cloud mode: LEXARIS acts as processor under a DPA with standard contractual clauses. Processing and hosting are in St Petersburg, Russia, which has no EU adequacy decision, so you will need your own transfer impact assessment. EU, UK, US and Swiss regions are planned and not yet available.
Keys stay with you: detection runs on the device, and LEXARIS receives neither document content nor keys.
The provider receives text in which direct identifiers are replaced. That can reduce transfer risk, but under EDPS v SRB your own obligations are assessed from your perspective as the controller holding the key, and remaining context may still identify people.
45 CFR §164.514
Health data (HIPAA)
Safe Harbor requires removal of the 18 identifiers and no actual knowledge of re-identifiability. Expert Determination is the alternative. Tokens not derived from PHI and kept confidential can serve as a re-identification code (§164.514(c)). Local mode lets you remove Safe Harbor identifiers before text leaves your device. You remain responsible for confirming all 18 categories, including dates other than the year, are removed. Cloud processing of PHI is not currently offered.
Privilege and confidentiality
Professional secrecy
Legal professional privilege, medical confidentiality and banking secrecy exist independently of data-protection law. Anonymising before AI reduces the risk of disclosure. Local mode means no third party, including LEXARIS, receives the source data.
Template
Clause for an AI acceptable-use policy
Documents containing personal data or confidential client information may be submitted to generative AI services only after they have been pseudonymised using LEXARIS. Re-identification keys are held [in the organisation’s LEXARIS workspace (hosted in St Petersburg, Russia) / only on the employee’s device], and only authorised staff may restore the original data. Each document must pass the pre-send check before submission.
Pricing
Start for free.
Placeholder amounts for this design. Final prices, including VAT where applicable, will be published before launch.
LEXARIS removes personal data from documents before they are sent to AI. It finds 116+ types of personal data (names, IDs, addresses, phone numbers, bank details, medical codes and more) and replaces them with reversible tokens. The AI works on the anonymised text, and its answer is restored with one click. In "Keys stay with you" mode, detection runs in your browser, and neither your document text nor the keys ever reach our servers.
Can I get the original data back?
Yes. In token mode, the AI’s answer is restored with one click. Pseudonym and format-preserving encryption are reversible with the key too. A hash is one-way by itself, but values can be restored if you keep the lookup table. Mask, synthetic and differential-privacy modes are irreversible by design, for publishing and analytics.
Where are my data and keys stored?
In Cloud mode, documents and keys are processed on our servers in St Petersburg, Russia, and keys are stored encrypted; EU, UK, US and Swiss regions are planned and not yet available. In "Keys stay with you" mode they never leave your browser: the key stays in browser storage or in a key file you download.
Which AI tools does it work with?
With six providers, including ChatGPT, Claude and Gemini, and with any model through the API and LLM proxy.
Is it GDPR-compliant?
LEXARIS is designed to support GDPR compliance: tokenisation is pseudonymisation under Art. 4(5). Cloud mode currently processes data in Russia, which has no EU adequacy decision; business customers receive a DPA with standard contractual clauses, and you will need your own transfer impact assessment. In "Keys stay with you" mode, personal data from your documents never reaches us.
Is tokenised data still personal data?
For you, yes: holding the key makes it pseudonymised personal data (Recital 26). For a recipient without the key, such as an AI provider, it depends on that recipient's reasonable means of re-identification (EDPB Guidelines 01/2025 (version for public consultation); CJEU C‑413/23 P).
Does sending tokens to a US AI provider count as a transfer?
The provider receives text in which direct identifiers are replaced. That can reduce transfer risk, but under EDPS v SRB your own obligations are assessed from your perspective as the controller holding the key, and remaining context may still identify people. The final assessment depends on the data and your processing model.
Can it be used for HIPAA de-identification?
Local mode lets you remove Safe Harbor identifiers before text leaves your device. You remain responsible for confirming all 18 categories, including dates other than the year, are removed. Cloud processing of PHI is not currently offered.