Limited pilot / Measurable scope

One use case.
A verifiable result.

The first pilot does not try to protect the entire AI landscape. We define the use case, models, data classes, infrastructure and acceptance criteria.

Limited pilotProduction after validation

Start the discussion

Tell us about the first route.

The form does not send data to a CyberPass server. Submitting opens your email client with a prepared message to hello@cyberpass.kz.

  • 01
    Use case

    An application, device group or managed browser channel.

  • 02
    Models

    One local and/or one public OpenAI-compatible endpoint.

  • 03
    Acceptance

    Data rules, latency baseline, events, backup/restore and security gate.

01 / DEFAULT PILOT

A baseline scope
you can validate.

Final scope is confirmed after discovery. The following is a practical starting boundary.

Included

Gateway pilot

  • On-premise Gateway and Control Center
  • One OpenAI-compatible client route
  • One Private and/or Public LLM
  • Text DLP and LLM Guard in observe/block mode
  • Policy decisions, events and SIEM path
  • Compose install, backup/restore, upgrade acceptance
Separate stage

Not included by default

  • Kubernetes HA and multi-site DR
  • Full air-gap lifecycle and offline licensing
  • Validation of a specific IdP/LDAP schema and external HSM/KMS
  • MCP resources/prompts, files and multimodal inspection
  • Transparent TLS inspection
  • Regulatory certification
02 / INPUTS

What to prepare
for the technical meeting.

More precise inputs mean fewer unverified assumptions in the estimate.

01Use case, users and permitted actions.
02LLM endpoints, authentication protocol and network route.
03Data classes, synthetic tests and fail mode.
04Linux host, PKI, DNS, proxy and egress policy.
05SIEM endpoint, transport and retention.
06Throughput, latency baseline and acceptance owner.