Each layer follows the same structure: what the specific risk is, what we do about it, and what technical solution stands behind it. The technical details can be expanded.
01Physical layer
The server is not sitting in an office closet
A large share of the most expensive data losses do not come from an attack but from a server room that flooded, overheated or was broken into. An office server cabinet is protected by luck, not by security: everything is fine with it until something goes wrong.
The environments we run are housed at two independent data center sites in Budapest, in physically protected data halls with redundant power and data center cooling. The hardware is enterprise-grade: redundant power supplies, ECC memory, RAID-protected storage.
Fire, water damage and power outages do not start out as security issues. They end up as one.
Infrastructure details on the Technology page →
Technical details
- Sites: two data center sites in Budapest, independent of each other
- Data hall: physically protected environment, redundant power, data center cooling and temperature monitoring
- Certification: both sites are in ISO/IEC 27001:2022 certified data centers, in dedicated, locked racks
- Hardware: enterprise server fleet with redundant power supplies, ECC memory and RAID-protected storage
- Backbone: redundant 2×10 Gbps—losing one uplink or one switch does not mean a service outage
- For on-site systems: monitoring of the uninterruptible power supply (UPS), with orderly shutdown during a power outage
02Network edge
A dedicated gateway per client, not a shared template
On a shared firewall, other people’s rules sit right next to yours. A port opened for another client, a misconfigured forward or a mail server IP with a damaged reputation becomes your risk too—even if you did everything right.
With us, every client gets its own gateway: the firewall, NAT and port rules belong only to your environment, the VPN endpoint is dedicated, and the public IPv4 address comes from our own address range and is assigned to you by name.
Having your own public IP also means your email reputation does not depend on what another client does.
Which vendors stand behind the network →
Technical details
- Gateway: dedicated firewall, NAT and VPN endpoint per client, with its own rule set
- Public IP: dedicated public IPv4 address from our own address range
- VPN: site-to-site and client access—IPsec, WireGuard, OpenVPN, SSTP
- Devices: MikroTik RouterOS, Cisco, plus Fortinet, pfSense and OPNsense firewalls
- Certificates: management of public and internal TLS certificates, with expiry monitoring
- On site: Wi-Fi infrastructure with guest network separation
03Segmentation
A breach of the public layer should not reach the database
In most serious incidents, what matters is not that the attacker got into the first machine but how far they got from there. On a flat network, a compromised web server has a direct line of sight to the file server and the database.
With us, the internal layer (database, application) and the public layer (web, mail) are strictly separated, and client segments are isolated from each other with VLAN-based or physical segmentation. Where the nature of the data requires it, the environment cannot be reached from the internet at all—for our clients who handle sensitive data, this is already working practice, not a plan.
In an audit, “logically separated” is often not enough. That is why we offer physical segmentation too.
The isolation diagram and real-world examples →
Technical details
- Two-tier design: separate internal (database, application) and public (web, mail) network layers
- Client isolation: a dedicated segment per client, with VLAN-based or physical segmentation
- Full isolation: on request, a dedicated virtual or physical switch, a private subnet, and an environment not reachable from the internet
- There is no crossover between segments—one client’s environment cannot see another’s
04Identity
A stolen password alone should not be enough
Signing in is exactly as easy for an attacker as for a colleague—if only a password stands in the way. The second common mistake is not at sign-in but after it: in most companies, nobody can say exactly what a given account actually has access to.
That is why we ask for two things at once: multi-factor authentication for sign-in, and a documented, maintained permission structure behind it. The accounts and devices of joining and departing employees are added and removed by a defined process—not from memory.
Permissions become a risk when nobody knows anymore who has access to what.
What an office AI assistant actually sees—and why it is about permissions →
Technical details
- Authentication: multi-factor authentication, centralized permission management
- Directory: Active Directory domain, Group Policy, redundant domain controllers
- Lifecycle: account and device management for joining and departing employees, with a documented permission structure
- Cloud: Microsoft 365 and Entra ID administration, email and phishing protection
- Only authorized Triox staff have access to the infrastructure, to the extent required for operations
05Endpoint
The attack does not start on the server
In the vast majority of cases, the entry point is a workstation: an opened attachment, an outdated browser extension, an application that has not been updated in two years. The server is only the second stop.
Endpoints run protection that is monitored from a central console, and patch management is a scheduled, tested and traceable process—tied to a maintenance window, not ad hoc clicking. Devices are inventoried, and their lifecycle is tracked.
What nobody watches keeps working—until it doesn’t.
Endpoint monitoring is part of managed IT operations →
Technical details
- Endpoint protection: deployment and monitoring from a central console, on workstations and servers
- Patch management: scheduled installation of Windows and Linux patches, tied to a maintenance window (WSUS- or RMM-based)
- Remote management: RMM: endpoint and server monitoring, remote intervention, device inventory
- Device lifecycle: inventory, warranty and support tracking, replacement planning
06Data
Ransomware goes after the backup too
The first target of an encryption attack is not the data but the backup. If the backup sits on the same network, in the same rack or in the same building as the production system, a single event takes out both.
That is why, with us, backup is not a by-product of the production system but separate infrastructure: multiple, physically independent targets at two sites. Critical workloads also have an off-site VM replica—it does not need to be restored, only started.
The question is not whether there is a backup. It is where it is, and how long it takes to be up and running.
A test proves that the backup can actually be restored →
Technical details
- Backup platform: Veeam Backup & Replication and Veeam Agent; alternatively Synology, Iperius and BackupChain solutions
- Targets: primary repository and separate physical secondary backup targets, on independent devices, at two sites
- 3-2-1 rule: 3 copies, on 2 different media, 1 off-site
- Replication: off-site, ready-to-run VM replica for critical workloads
- Application-aware backup: database- and mail-system-consistent backups, not just file-level copies
- Verification: daily verification of backup jobs—not just that they started, but that they also completed
07Management plane
What cannot be reached from the internet cannot be hacked from there
The security of an infrastructure is often decided not at the application but by where the management interfaces of the servers, the hypervisor and the network devices can be reached from. This is the shortest path to the entire environment—and an exposed management interface can be probed by the whole internet.
Triox’s management network is dedicated and not reachable from the internet. No management interface is exposed that a password would have to protect. Client access goes through VPN, via a dedicated endpoint per client.
The best-protected management interface is the one that is not exposed.
Who has access to my data?—FAQ →
Technical details
- Out-of-band management: dedicated management network, not reachable from the internet
- Hardware level: iDRAC and iLO—the server can be managed and monitored even if the operating system does not start
- Early warning: power supply, fan, memory and disk faults are in many cases detected before an actual failure
- Client access: through VPN, via a dedicated endpoint per client
- Up-to-date, transferable and auditable system documentation—not in the operator’s head