AI productivity without the data risk.

Null anonymizes your sensitive data in real time, before prompts reach the model. The model sees tokens. Your people see originals. Compliance sleeps through the night.

Book a demoTry in browserNo card. EU-hosted.
Claim file · draft

Allianz Versicherung received claim #S-2024-48113 from Mara Hoffmann on March 3, 2026. The insured vehicle was registered at Leipziger Str. 14, 10117 Berlin. Witness Jonas Becker (policy VK-993-0144) filed the statement. Medical records for IBAN DE89 3704 0044 0532 0130 00 are pending review.

7 entities detectedConfidence 0.96
Latency +138ms

Built for the most regulated teams in Europe.

Insurance, healthcare, legal, financial services, public sector. If your data can't cross a border, Null is how you still ship.

Customer stories

We couldn't put GPT in front of case workers without a wall between the model and our data. Null is that wall, and it's invisible to the people using it.

Chief Information Security Officer, Top-3 German insurer

GDPRBDSGEU AI ActISO 27001ISO 42001Hosted in Frankfurt

Three things Null promises. Without exception.

01 · Perimeter

Zero raw sensitive data forwarded.

Structure-aware document intelligence detects names, addresses, dates, insurance numbers, contract IDs, and internal codes, not just regex. The model receives deterministic Vault tokens.

PERSON_A·S7FT9CDDATE_X·91VVZAIBAN_E·P2N81HADDR_C·KM3R4J+ 3 more
02 · Re-identification

Local re-identification.

The model's response is resolved on the user's device. No sensitive data in server logs, no sensitive data in inference traces, no sensitive data in training data.

03 · Sovereignty

Made in EU, for the EU.

Frankfurt routing. GDPR, BDSG, EU AI Act, ISO 27001, ISO 42001. No detour through US clouds. Pseudonymization meets Art. 32 “state of the art” by design.

EU

A prompt, end to end. In 138 milliseconds.

Every request follows the same four steps. No raw sensitive data crosses the perimeter at any point, and every reveal is audited for seven years.

Inbound prompt

// raw, as typed by the user
> Summarize the claim filed by Mara Hoffmann on
  March 3, 2026 against Allianz (case
  #S-2024-48113). Flag liability risks.

// 7 sensitive data entities detected · blocking inference

Detection graph

person.full_name    "Mara Hoffmann"        0.98
person.full_name    "Jonas Becker"         0.96
org.insurer         "Allianz"              0.99
case.id             "S-2024-48113"         0.99
date.claim_filed    "March 3, 2026"        0.94
address.postal      "Leipziger Str. 14…"   0.93
bank.iban           "DE89 3704 0044 …"     1.00
Step 1 of 4 · Prompt interceptedCumulative latency: +12 ms

Reads every document. Redacts what matters. Anonymization calibrated to what your workspace calls sensitive: names, IDs, internal codes, anything you define.

Confidence scoring

Every detection, with its doubt.

Each entity comes with a confidence score, alternatives, and provenance, NER, structural signal, or custom rule. Thresholds tunable per entity type.

Plain-language rules

Describe what to hide, in words.

“Internal project codes starting with PRJ-” and our engine builds the detector. No regex gymnastics, no ML pipeline to wire up.

Structure-aware

Roles, not just names.

Null knows the difference between plaintiff and defendant, case worker and claimant, by reading the document's structure, not just its words.

Scoped reveal

Session. Case. Global, with approval.

Scoped re-identification: per conversation, per case file, or global with admin sign-off. Every reveal is audited, every time.

Immutable audit

Seven years. Exportable on request.

DSGVO-grade append-only log. Who revealed what, when, under which scope. Exportable on subject access request.

Model allow-list

Admins decide which models run.

Workspace admins curate the model roster. Regions, versions, training opt-out enforced upstream. No shadow-AI.

Enterprise only. Regulated first.

Insurance · by the numbers
38%of adjuster time spent reading docs

Claim files. Underwriting memos. Medical reports. Adjusters move faster when the model reads for them, and compliance stays quiet because names, IBANs, and case IDs never leave the perimeter.

Detected entity types
Claimant namePolicy numberCase IDIBANIncident addressWitness statementInsurer org

What people do with Null on a Tuesday morning.

Case worker · Insurance

Summarize a claim file without ever seeing a cleartext name.

Drop a 40-page file into chat. Ask free-form questions. GPT-5.1 answers, from tokens, not identities.

Summarize liability exposure in CASE_B·F8QT2P.
Underwriter · Insurance

Search 200 pages of contract for risk clauses, inside the perimeter.

Indemnity, force majeure, subrogation. The engine finds them. Your counterparty's name never leaves your walls.

Find unusual indemnity caps signed by PERSON_A·S7FT9CD.
Compliance · Financial services

Review suspicious activity with AI, and with a clean audit trail.

Every reveal is logged, scoped, and traceable. DSGVO Art. 32 “state of the art” pseudonymization, out of the box.

Flag anomalies in IBAN_E·P2N81H across Q1.

Frankfurt routing. EU standards. No detour through US clouds.

GDPRArt. 32 pseudonymization
BDSGFederal data protection act
EU AI ActRisk-tier compliant
ISO 27001Information security
ISO 42001AI management system
DE · FrankfurtHosted in Germany

Privacy as a utility. Priced like one.

Business

For regulated teams.

€31per seat / month, billed annually
  • Structure-aware detection
  • Scoped re-identification (session · file · global)
  • All frontier models, Null-managed
  • SSO (SAML / OIDC), 7-year audit log
  • DPA, ISO 27001 / 42001 attestations
Start 14-day trial
Enterprise

Dedicated deployment.

Customon request
  • On-premises or private cloud
  • SCIM provisioning, self-hosted model routing
  • Custom SLA (99.95%+), named security engineer
  • Procurement, vendor-review, red team support
  • Unlimited seats
Talk to sales

Free 14-day trial included. Starter tier from €15/seat. See full pricing →

Give your team AI. Keep the data where it belongs.

Book a demoTry in browser