Skip to content
Book a call
IT EN
Privacy by design Coming soon

The intelligence is external. Your data is not.

Every time an agent asks an external AI model for help, Agentic Darts detects and replaces in the text the data that identifies people and companies. The model reasons about neutral placeholders. The real data stays within your perimeter.

Roadmap feature: currently being built into the platform.

The problem we solve

The most capable AI models are external services. Using them without precautions means sending third parties customer names, tax codes, IBANs (International Bank Account Numbers), contract terms, health or commercial information. This is where the most serious risks for a company arise: breaches of the GDPR (General Data Protection Regulation), loss of trade secrets, customer disputes, fines.

How it works, in five steps

1

Detect

Before every call to an external model, a dedicated engine finds the confidential data in the text and attachments: first and last names, tax codes and VAT numbers, IBANs and cards, emails, phone numbers, addresses, licence plates, dates of birth, case and contract numbers, plus the categories you define (project names, confidential prices, strategic customers).

2

Replace

Each item is replaced by a consistent placeholder: "Mario Rossi" becomes [PERSON_1], his IBAN becomes [IBAN_1]. The text remains understandable for the model, but no longer says anything about anyone.

3

Safeguard

The mapping between placeholders and real data is kept only within your perimeter, encrypted, and deleted at the end of processing. Technically it is reversible pseudonymisation: only your company can link the placeholders back to the real data.

4

Reason

The external model receives only the text with the placeholders: it classifies, summarises, drafts, replies.

5

Restore

When the answer comes back, the placeholders become real data again, inside your systems.

Four levels of protection

Automatic pseudonymisation

of every request to external models, with rules per data category.

Data that must never leave

for the most delicate categories (for example health or judicial data) processing is mandatorily routed to models installed in your data centre.

Local models when needed

the platform also works with models run on-premise (in your data centre), with no traffic to the outside.

Traceability

the audit log records every call to the models (which agent, which model, which data categories were replaced), without storing the data in clear.

Without Agentic Darts

Full texts with names, tax codes and contract terms sent to external AI services. No trace of what went out and to whom.

With Agentic Darts

The model receives only placeholders; the map stays within your perimeter; every call is recorded; the most sensitive data never leaves.

GDPR

For your DPO (Data Protection Officer)

  • Pulse acts as data processor under Art. 28 GDPR, with a data processing agreement (DPA, Data Processing Agreement) available.
  • Up-to-date list of sub-processors, including the AI model providers.
  • Choice of model providers with processing in the European Union and no retention or training on customer data, where provided by the contracts.
  • Support for the data protection impact assessment (DPIA, Data Protection Impact Assessment) and technical documentation for IT, available on request.
  • Live demonstration: a real request before and after data replacement.
A note on the term: from the GDPR standpoint, replacement with placeholders is reversible pseudonymisation, not anonymisation. For your company the data remains personal data; towards the model provider the text contains no identifiers. No recognition system is infallible on free text: the residual risk is managed with rules per category, routing to local models and periodic review of missed cases.

Want to see what really reaches the model?

During the call we show you a real request, before and after replacement.

Book a call