LLM_Deploy
Talk to an engineerMenu

Where your data goes, and who can see it_

The model runs on GPUs inside your network. Prompts and logs stay there, access is controlled by role, and every action is written to an audit log you keep. Switch the monitor to see what a cloud API does with the same traffic.

Boundary monitorDEMO TELEMETRY
Where the model runs
  • Left: your apps. Centre: the /v1 endpoint. Right: GPU nodes. Lattice wall: your network boundary. Outside: the third party.
External endpoints reached06
Packets inside2,841
Probes deflectedPackets left boundary17412

Nothing leaves. Anything probing from outside bounces off the boundary. Switch to Cloud API to see the same traffic cross it.This is a cloud API. Every prompt crosses the boundary to a third party. Switch back to see the same traffic stay inside.

What is logged.

Every request, every admin action and every refusal is written to an audit log that stays in your storage. This is a sample of what a line looks like. The stream below is demo telemetry, not a customer's data.

DEMO TELEMETRY · static

No entries at this level. Set the level to All.

  1. 09:41:01.623INFOgatewayPOST /v1/chat/completions model=qwen3-32b egress=0 B
  2. 09:41:02.484FACTegressnothing leaves your network · external destinations=0
  3. 09:41:03.727OKrbackey …7c02 scope=chat granted · role=app
  4. 09:41:04.625INFOgatewayPOST /v1/embeddings model=mistral-small-3.1 egress=0 B
  5. 09:41:05.979DENYrbackey …a41f scope=admin denied · role=reader
  6. 09:41:06.840FACTrbacrole-based access control on every key and every user
  7. 09:41:08.305OKauditaudit export signed · external destinations=0
  8. 09:41:09.844INFOmodelgpt-oss-120b loaded on 1× 80GB GPU · cache warm
  9. 09:41:10.890FACTauditevery request and admin action is written to the audit log
  10. 09:41:11.825WARNopsdriver update queued on 2 of 4 nodes · window Sun 02:00
  11. 09:41:12.760INFOgatewayPOST /v1/chat/completions model=gpt-oss-120b egress=0 B
  12. 09:41:13.769FACTendpointOpenAI-compatible /v1 endpoint served inside your network

The controls.

Five statements about how a deployment is built. Each one can be checked on your own network during the pilot.

  1. 01Role-based access control

    Every API key and every user has a role. Calls outside the role are refused and written to the audit log.

  2. 02Audit logging

    Every request and every admin action is logged with who, what and where. The log and its signed exports stay in your storage.

  3. 03OpenAI-compatible endpoint, inside your network

    Your applications call a /v1 endpoint on your own network. Change the base URL in the SDK you already use.

  4. 04Air-gap option

    The deployment can run with no route to the internet. Model weights and updates are delivered to you rather than fetched by the cluster.

  5. 05What leaves your network

    Nothing. Prompts, completions and logs are processed and stored on your GPUs and in your storage.

Compliance.

Frameworks our customers ask about. Documentation on request.

  • framework 01 / 04

    HIPAA

    Deployments for healthcare workloads that handle protected health information.

    Scope
    On request
    Date
    On request
  • framework 02 / 04

    SOC 2

    Operational controls for a managed deployment.

    Scope
    On request
    Date
    On request
  • framework 03 / 04

    FedRAMP

    Deployments for US federal workloads.

    Scope
    On request
    Date
    On request
  • framework 04 / 04

    GDPR

    Processing stays inside your network, in the jurisdiction you choose.

    Scope
    On request
    Date
    On request

Documentation on request. Ask an engineer for the scope and date that apply to your deployment.

Support and escalation.

Incidents go to the on-call engineer first. SLA tiers are part of the subscription, so the escalation path and named contacts are written into your agreement. Ask an engineer which tier fits your node count.

99.95%Uptime SLA
The published commitment for managed deployments.
5 minIncident response
From the alert to an engineer working on it.
24/7Support hours
On-call rota shared between the Limassol and Auckland offices.

Who you contract with.

The legal entity behind this service is MAX M LTD, Cyprus. An UpSystems practice — enterprise infrastructure since 2000.

Limassol HQArchaion Solon 1, 2nd floor, Mesa Geitonia, 4006, Limassol
Cyprus
+357 25 123877
Auckland53 Fort Street, Auckland CBD, 1010
New Zealand
+64 9 887 9088

Check it on your own network.

The 72-hour pilot runs on your data inside your network, so egress, access control and the audit log can be tested before you decide.