PUBLIC TECHNICAL DOCUMENT · TES-BTD-001

Read-Only Architecture Whitepaper

Security architecture of the TES Remote DCIM Platform (module GPUasS AI-Client Engine v.1.0) and its on-site Edge Collector. Intended for client system administrators, information-security officers, auditors and bank compliance teams.

DOCUMENTTES-BTD-001
VERSION1.0
CLASSIFICATIONPublic
OWNERTech-Engineering Solutions LLP
Ask a security question

01Scope and design principle

The platform follows a Hub-and-Spoke topology. Each client facility (spoke) runs one or more Edge Collectors that only read telemetry from local equipment and only push it to the central hub operated from Astana, Kazakhstan (UTC+5). The hub has no technical channel back into the client network.

Core guarantee: the Edge Collector exposes no listening network ports and contains no command handlers. The hub physically cannot send a control command, reboot a server, change GPU clocks or power limits, or otherwise influence client hardware.

02Edge Collector runtime hardening

ControlImplementation
PackagingLightweight Docker container, Python 3.12. Footprint: 1 vCPU, 1 GB RAM, 10 GB disk.
UserUnprivileged non-root user (--user 10001:10001), all Linux capabilities dropped, no-new-privileges.
FilesystemRead-only root filesystem; the only writable volume is the local spool /var/lib/tes/spool.
Network inboundNone. No published ports, no listening sockets, no remote shell, no command handlers.
Network outboundHTTPS (TLS 1.2+) to hub.tesmonitoring.com:443 only. Can be pinned by the client firewall to a single FQDN.
SecretsPer-site HMAC key /etc/tes/site.key, mounted read-only, unique for each site.

03Transport, authentication and integrity

X-TES-Signature = hex( HMAC-SHA256( site_key, timestamp + "." + body ) )

POST /api/v1/ingest/batch HTTP/1.1
Host: hub.tesmonitoring.com
X-TES-Site: SITE-EU-01
X-TES-Collector: COL-MODULAR-01
X-TES-Timestamp: 2026-10-28T14:30:00Z
X-TES-Signature: 8f9c2b4e7a1d...
Content-Encoding: gzip

04Resilience: 72-hour local spool

If the site uplink fails, the Local Spool (store-and-forward) buffers telemetry on disk for at least 72 hours. When connectivity returns, batches are replayed in order; idempotency on batch_id guarantees that no sample is lost or duplicated.

05Data minimisation and edge pseudonymization

06Hub-side tenant isolation

Data is organised in the hierarchy Tenant → Site → Room → Rack → Device → Metric. Each tenant's telemetry, ML models, reports and RAG knowledge base (local LLM, no external clouds) are logically isolated. Reports are exported as PDF / CSV / JSON with SHA-256 checksums for audit trails.

07Supported read-only sources

AdapterEquipment
nvidia-smiNVIDIA GPU accelerators (query mode only)
DCGM ExporterNVIDIA DCGM Prometheus endpoint (HTTP GET, local)
IPMI / RedfishServer BMC sensors (read-only account recommended)
SNMP / ModbusPDU, UPS, CRAC/CRAH, dry coolers (SNMP GET / Modbus read functions)
CSV importHistorical data and third-party systems

08What the hub can NOT do

Collector updates are performed only by the client's administrator by pulling a new signed image version (check & run).

09Document control and contact

Tech-Engineering Solutions LLP (ТОО «Техно-инжиниринговые решения»), BIN 221240000460
Z00Y7B8, Astana, Nura district, Turan Ave., building 46/1, office 613, Republic of Kazakhstan
Web: tesmonitoring.com · Email: info@tesmonitoring.com · Phone: +7 911 283 53 20