Security through architecture

Not policy. Not passwords.
The architecture itself.

A typical AI agent has no internal boundaries — one client's data mixes with another's. In S.V.I., every department is physically isolated: its own server, its own permissions, its own context. Any exchange goes through Mai — the single point of control.

🇩🇪 EU/GDPR via Frankfurt Physical data isolation NDA on every contract Full audit trail
The key difference

Typical AI mixes data. S.V.I. doesn't.

When a company runs one 'universal' AI, all data flows through one model: finance sees sales, support sees marketing. Our model is different: every department is isolated on its own server. Cross-department requests go through Mai — and she decides whether to release the data.

Typical AI agent Risk

ChatGPT, Copilot, and the rest — one model for everything.
  • One context for all tasks — finance sees sales data, sales sees marketing
  • Cross-session leakage: the model 'remembers' prior conversations from other customers
  • No hard role separation — any request can pull any data
  • Logs and audit live on the vendor's side, not yours
  • Protection is policy and user-level limits, not infrastructure

S.V.I. (HandOfHands) Protection

5-tier hierarchy, one server per department, Mai as the gateway.
  • Every department runs on its own physical server. Access to another department's data simply doesn't exist at the infrastructure level
  • One client = one dedicated server. No overlap with other clients — anywhere
  • Cross-department exchange goes through Mai — she validates the request and decides what to release
  • Full audit trail: who, when, what data they touched — visible in your dashboard
  • Protection is built into the architecture — not a setting, but the shape of the system itself
Four pillars of security

Not one safeguard — four layers at once.

They work together. A breach would have to break the infrastructure, fool Mai, bypass the audit trail, and crack the encrypted backups — all at once. Beyond the scope of a single client's contract, that's technically impossible.

01

Physical data isolation

Every corporate department runs on its own server with its own database. Finance can't see sales. Marketing can't see support. This isn't an access setting — it's separate infrastructure. A request for another department's data simply can't execute without explicit permission.

02

Mai — the single request gateway

When a booking agent needs accounting data, it doesn't go fetch it directly. The request lands with Mai. She checks: is the request legitimate, is this the right set of fields, can they be returned in this context? Only then does data flow — and never more than necessary.

03

Full audit trail

Every agent action, every cross-department request, every Mai decision — logged and visible in your dashboard. Who, when, what data was touched, and what Mai approved. No 'black box': fully transparent to you, tamper-proof for everyone else.

04

One client = one server

Your contract is your own physical server. Your data never overlaps with another client's — not in training, not at inference, not in logs, not in backups. When you leave, the server gets wiped clean. No 'residue in the model'.

How it works in practice

A cross-department request, in practice

A booking agent takes on a guest and needs prepayment info from accounting. Instead of going direct, the request routes through Mai. She decides whether to release the data — now, to this agent, in this context.

Dept A · Booking isolated server 1 Mai GATEWAY 2 Dept B · Accounting isolated server 3 1 · Request 2 · Mai filter 3 · Filtered response audit · all 3 steps logged · MAI=ALLOW · 12ms

The accounting agent has no way to know which guest is involved or why the prepayment is needed — it sees only the requested field. The booking agent has no access to the invoice, the payer, or the payment history. Mai sees the whole exchange and logs every step.

Compliance & Legal

Jurisdiction, contracts, infrastructure.

EU data is processed and stored in Frankfurt (GDPR). A separate NDA for every contract. Backup infrastructure across multiple regions. Everything your legal team will want to review.

NDA on every contract
Client names and identifying details are not published. Standard practice for enterprise deals.
GDPR · Frankfurt
All EU data is processed and stored in the Frankfurt data center. Full compliance with GDPR Art. 28/32/44.
AI agents monitor & auto-recover
A dedicated team of AI agents monitors every server 24/7. On failure — auto-restart, auto-migration to backup, auto-restore from backup. No action required on your side.
9 servers · 3 continents · failover
Phuket and Frankfurt are primary. Seven backup servers across Asia, Europe, and the US. If a region goes down, traffic shifts to the nearest backup automatically. Your customers see uninterrupted service.

Want to see exactly how this is built?

Request the Security Brief for your legal team or CISO. Under NDA we walk through the access model, audit, isolation, incident history, and SLA — everything you need to sign off on an enterprise rollout.

Request Security Brief via Mai Email us