Your data. Your rules. Any AI model.
JanusTB sits between your company’s data and every AI model. Each request is checked — who is asking, what data is sent, where it may go — then allowed, redacted, routed or blocked. Every decision is recorded.
- Live in production
- Hosted in France
- SSO with Google & Microsoft
- Request Summarise the Q3 financial report
- Identity alice · Finance
- Data Confidential · financial
- Policy confidential stays in EU
- Decision
- Model Mistral · EU
- Audit evt_8f3a29 recorded
Works with
Company data now flows to AI from everywhere. Nobody controls it.
Apps, agents and RAG assistants send data to AI providers through APIs. Employees paste it into chatbots. Each connection has its own rules — and no one can say which data went to which model.
0%
of companies whose AI apps were breached had no proper AI access controls
IBM Cost of a Data Breach, 2025
0%
of AI data-policy violations in Europe involve customer, personal or financial data
Netskope, 2026
0%
of AI data-policy violations worldwide involve source code
Netskope, 2026
Data exposure
Sensitive data reaches external providers with no check on what it contains or who sent it.
Model lock-in
Each app is wired to one provider. Changing model means rebuilding the integration.
No central visibility
Security and compliance teams cannot see who sent which data to which model, under which rule.
Every request goes through five steps.
Scroll to follow one request through JanusTB.
Pay supplier invoice, IBAN FR76 3000 6000 0112 3456 7890 189 Pay supplier invoice, IBAN [REDACTED]
personal data → redactREDACTED { "event": "evt_2c81f0",
"user": "alice@acme.eu",
"decision": "redact",
"entities": ["iban"],
"model": "mistral-eu" } - 01
Intercept
Apps and agents call one JanusTB API instead of each provider. Employees are covered by the browser extension.
- 02
Identify
JanusTB knows who is asking: single sign-on identity, role and organisation — or the app’s scoped key.
- 03
Inspect
The content is scanned for personal data, secrets and confidential information: IBANs, card numbers, national IDs, API keys and more.
- 04
Decide
The policy engine applies your rules and returns one decision: allow, redact, route or block.
- 05
Record
The decision, the policy and the model are written to the audit trail — ready for your auditor or your SIEM.
The layer between enterprise data and AI.
Every flow — from apps, agents, RAG assistants and employees — goes through one control plane before it reaches a model.
See what leaves before it leaves.
Employees paste sensitive data into AI chats every day. The JanusTB browser extension checks each message before it is sent.
- Works on ChatGPT, Claude, Gemini and Mistral
- Detects IBANs, card numbers, national IDs, secrets and more
- Block, warn or log — your policy decides
- Fail-closed: if JanusTB is unreachable, nothing is sent
- Available on the Chrome Web Store
Build once. Choose your model. The policy decides.
Change the request and watch JanusTB decide. The rules shown are an illustrative example of a policy you would write.
policy: data-to-ai-v1rules: - if: content.secret action: block - if: classification == confidential allowed_models: [mistral-eu, azure-openai-eu] otherwise: route mistral-eu - if: content.personal_data action: redact - default: allow
Answers from your documents. Only what each user may see.
Every passage is checked against the user’s identity, role and the document’s classification before it can enter the model context.
Relevant context, authorized access.
Retrieval uses the same policy engine as the gateway. Confidential documents are never sent to external embedding providers.
Know what your AI is doing.
One view of every request, decision, model and policy — with evidence you can hand to an auditor.
Generate a PDF of decisions, policies and models for a period — ready for your auditor.
Or pull audit events into your SIEM (NDJSON).
Control you can show to your auditor.
Hosted in France
Production runs on OVHcloud in France.
Single sign-on
Google Workspace and Microsoft Entra ID, with optional enforced SSO.
Fail-closed
If JanusTB is unreachable, requests are blocked, not let through.
Scoped API keys
Each key has one purpose. An extension key cannot read audit data.
Audit export
PDF evidence bundle and SIEM export. Exports never include prompt text.
GDPR ready
Data processing agreement and privacy policy available.
Starts on European cloud. Your infrastructure next.
OVHcloud · France
Our production environment today.
● LiveAWS
Deploy where your infrastructure lives.
RoadmapAzure
For Microsoft-centred companies.
RoadmapOn-premise
Private deployment for large accounts.
RoadmapTake control of your AI stack.
We are working with a small group of European companies as design partners: three months free and direct access to the founders.