Use AI on Egyptian customer data without sending it to the cloud.

AI keeps working on real customer records. Protecto hides Arabic and English personal data in real time, across agents, MCP calls, and RAG, and controls who sees the real values. Deploy on-premises, in your private cloud, or air-gapped.

Where each value goesIllustrative

Customer nameStays in your data center

محمد حسن

National IDStays in your data center

29001011234567

Complaint textGoes to the model as tokens

[PERSON_9e2] reported a failed transfer.

Real values stay inside your environment

Trusted by regulated enterprises & agentic platforms

Case study result

Bank Muscat, Arabic and English data

Read the case study →

99%

Recall

96%

Precision

4 weeks

To a working proof

The problem

Your team wants AI on real customer data. Compliance says it stays inside.

The need

Your team wants AI

Claims, loan files, complaints, and call notes are where AI saves the most time. They are also full of names, national ID numbers, and account details, in Arabic, English, and a mix of both.

The challenge

The cloud is off limits

Compliance says customer data cannot go on the cloud. Your contracts and sector rules may say the same, and for banks, Central Bank of Egypt rules apply on top. Sending the data to an outside AI service, or to a cloud privacy tool, is off the table.

The gap

The workaround fails

DLPs and firewalls often miss sensitive information in Arabic, including names and 14-digit national IDs. When they do catch it, they blank it out and strip the context the model needs, so answers fail, even on a model you host yourself.

The risk

Inside is not the same as safe

Once AI is running, real names and ID numbers land in vector stores, agent memory, and logs. Developers, vendors, and every team with access can read them, and the model cannot tell who should see what.

How it works

Real data stays inside. AI sees protected values.

A bank’s service team asks AI to summarize a customer complaint. The customer’s name, ID, and phone become tokens before any model sees them, and real values return only where policy allows.

Deploy where your data lives: on-prem, private cloud, or air-gapped.

Choose your isolation

On-premises

Runs on your servers.

Private cloud

Runs in your cloud account, in the region you choose.

Air-gapped

No outside link. The model runs inside the same boundary.

Add it without changing your apps

One API, or a proxy

Call Protecto before the model, or route model traffic through it.

Your apps stay as they are

Same logic and workflows. No new interface to roll out.

Start with one flow

One use case, one data sample. Add flows as reviewers sign off.

Control inside your network

Same record, different viewers. Each sees only what policy allows.

Many people and tools touch AI data, even in a closed network. Protecto keeps real values out of the model, the vector store, and the logs, and releases them only to the people your policy names.

protecto-vault · inside your environmentIllustrative

$ agent: summarize complaint #4821

mcp tool call · crm.lookup(customer)

detect PERSON محمد حسن → [PERSON_9e2]

detect ID 29001011234567 → [ID_5b2]

detect PHONE +20 100 123 4567 → [PHONE_8d1]

model ← "Customer [PERSON_9e2], ID [ID_5b2] …"

model → "[PERSON_9e2] reported a failed transfer."

unmask case_handler → محمد حسن

unmask analyst → tokens only

audit every decision logged

Real values stay inside your environment

AI model

Always works on tokens

Name [PERSON_9e2]
ID number [ID_5b2]
Phone [PHONE_8d1]

Analyst

Trends and summaries, no identities

Name [PERSON_9e2]
ID number [ID_5b2]
Phone [PHONE_8d1]

Case handler

Needs the name to act on the case

Name محمد حسن
ID number [ID_5b2]
Phone [PHONE_8d1]

Compliance officer

Full view for reviews and requests

Name محمد حسن
ID number 29001011234567
Phone +20 100 123 4567

Every unmask is a logged decision, so your team can show who saw which values, and when.

For your reviewers

Five rules that decide where AI can run, and what Protecto does about each.

Cross-border violations carry imprisonment of at least three months and/or fines of EGP 500,000 to 5,000,000, doubled for repeat offences (Law 151 of 2020, Arts. 42 and 48). References are to the Personal Data Protection Law promulgated by Law No. 151 of 2020.

Rule	Requirement	Protecto solution
Transfer, storage and sharing abroadLaw 151/2020 Arts. 14, 15, 16	Personal data may be transferred, stored or shared outside Egypt only if protection abroad meets the law's level and the Center grants a licence or permit. The definition of a cross-border transfer includes storing and processing.	Protecto runs inside your network, with token mappings and keys in your own data center, so real values never leave. Models see protected values only. If a model runs outside Egypt, those values can still count as personal data, so Articles 14 to 16 apply to that path.
Sensitive personal dataLaw 151/2020 Art. 12	Health, genetic, biometric, financial, religious, political and criminal data, and children's data, need a licence from the Center and explicit written consent.	Protecto detects these categories and tokenizes them before AI sees them, so models and most tools work on tokens, not the sensitive values.
Security and recordsLaw 151/2020 Arts. 4, 5	Controllers and processors must apply technical and organizational protections, keep records of processing including cross-border transfers, and prove compliance on inspection.	Runtime masking and policy enforcement across prompts, MCP calls, APIs, RAG, agent context, logs and more. An audit log of every decision stays on your servers.
Vendors and processorsLaw 151/2020 Art. 5	Processors act on written instructions and must not disclose the data or the results of processing outside the permitted cases.	Detection, token mappings, and audit logs live on your servers. The model never holds the real values.
Breach noticeLaw 151/2020 Art. 7	Notify the Center within 72 hours of a breach, and data subjects within three days after that.	Real values stay out of models, vector stores, and logs, so there is less data to expose in an incident. The audit log shows who saw what.

Banks and other CBE-supervised entities: personal data held by entities under Central Bank of Egypt supervision is outside the PDPL if CBE rules on handling personal data are observed (issuing Art. 3(6)). The CBE rules apply instead, and Protecto’s on-premises and air-gapped options fit a boundary your team controls.

The whitepaper maps every article and every data path, with what may leave Egypt and on what footing.

Case study

How Bank Muscat got to a working proof in four weeks.

Financial services

Data residency rules blocked cloud AI. A mix of Arabic and English data made it harder.

Bank Muscat wanted to adopt AI but could not send customer data to top cloud models, and its Arabic, English, and mixed-language text made sensitive data hard to find. With Protecto, sensitive data is masked by default and unmasked only under policy. The team had a working proof of concept in four weeks.

99%

Recall on sensitive data, including Arabic edge cases

96%

Precision, so useful text is not blanked out

4 weeks

From start to a working proof of concept

Bring one use case and one Arabic data sample. We will show it protected, running on your own servers.

Resources

Go deeper.

Whitepaper · 28 pages

Egypt PDPL for Enterprise AI and Agents

Protect personal data while enabling useful AI workflows under Egyptian requirements. A path-by-path map of what may leave Egypt, and on what footing.

Blog

What Is a Data Privacy Vault?

How vault-based tokenization keeps sensitive data out of AI workflows.

Read the post

Guide

How to Ensure Data Privacy with AI

A step-by-step guide to protecting personal data in AI projects.

Read the guide

Industry

Sovereign AI for banks

How banks use leading LLMs while keeping sensitive data inside their boundary.

See the solution

FAQ

Questions security and compliance teams ask first.

Does Egyptian law say customer data cannot be on the cloud?

Not as a blanket rule in the law text. Article 14 of the PDPL allows personal data to be transferred, stored or shared abroad only if protection there meets the law’s level and the Center grants a licence or permit. Sensitive data needs its own licence under Article 12. Banks supervised by the Central Bank of Egypt follow CBE rules for the data they hold (issuing Article 3(6)), and contracts with your customers are often stricter than the law. A private cloud or data center inside Egypt is not a cross-border transfer under the Article 1 definition. Confirm your position with counsel and your sector regulator.

Can Protecto run with no internet connection?

Yes. Air-gapped is one of the ways to deploy Protecto, alongside on-premises and private cloud. In an air-gapped setup the AI model has to run inside the same boundary, so you host it yourself, for example with Ollama. Nothing in the flow needs an outside connection.

How hard is it to add Protecto to our existing apps?

Protecto is an API product. Your team calls the API before the model, or routes model traffic through Protecto as a proxy, and your existing app logic stays as it is. There is no new interface to roll out. Start with one AI flow and one data sample, then add more flows as reviewers sign off.

If we already keep data in our own systems, why mask it?

Because AI moves data out of those systems. A request, a retrieved document, or an agent’s tool call can carry real names and ID numbers to a model, a vector store, or a log, and many people and tools can read them. Protecto swaps sensitive values for protected ones at the start, so each model, person and tool sees only what your policy allows.

Does Protecto work on Arabic text and Egyptian identifiers?

Protecto detects 50+ entity types out of the box and supports custom entities, so you can add identifiers specific to your business.

What types of sensitive data does Protecto detect?

Protecto detects 50+ entity types out of the box, covering personal data (PII), health data (PHI), and payment card data (PCI). If your data includes identifiers that are not covered by default, you can add custom entities. Detection works in Arabic, English, French, German, Spanish, Dutch, and Japanese, with more languages on request.

Will masking make the AI answers worse?

Basic redaction can, because a blank placeholder tells the model nothing. Protecto uses consistent, format-preserving tokens, so the same person maps to the same token and the model can still tell who is who. Test it on your own data before relying on any figure.

Is tokenized data still personal data?

Tokenization is pseudonymization, not anonymization. Whoever holds the token mapping can reverse a token, so treat tokenized data as personal data unless counsel advises otherwise. In a self-hosted setup, the token mappings and keys stay inside your own data center, so real values never leave Egypt. If you use a model hosted outside Egypt, the protected values it receives can still count as personal data, so that path falls under Articles 14 to 16. Confirm your position with counsel.

Use AI on Egyptian customer data without the public cloud.

Bring one AI use case and one Arabic data sample. We will show it running on your own servers, what your reviewers can see, and what Protecto changes.