TREN
Fluxis

Fluxis Gateway

The control point through which all of the institution's language model and agent traffic passes. Sensitive data in every request is detected and masked, the request is routed to an on-premises or external model according to the sensitivity of its content, masked data is unmasked only for the authorised user when the answer returns; usage policies, cost and quality monitoring and the audit trail are gathered in the same component.

Solution

What we provide

Fluxis Gateway routes all of the institution's AI traffic through a single control point. Sensitive data in every request is detected and masked, the request is routed to an on-premises or external model or blocked according to its sensitivity, masked data is unmasked for the authorised user when the answer returns, and every decision is recorded for audit. AI use becomes governed rather than banned.

  • Sensitive data never leaves the institution: masked on the way out, unmasked on return
  • Policy decides which request goes to a local model and which to an external one
  • Every interaction is recorded to answer who, when and which data
  • Cost, usage and quality are monitored from one screen

Fluxis Gateway sits between the institution's applications, developer tools and agents and the language models. Every request is first classified: personal data (national ID, IBAN, card, name, address, phone), customer and account information and confidential internal content are captured with Turkish-specific detection rules and model-based recognition. Detected data is masked with consistent placeholders that preserve context; in the model's answer the same placeholders are converted back to real values according to the user's permissions (reversible masking). The routing decision is content-based: policy sends high-sensitivity requests to the on-premises model, the rest to an external model selected by cost and quality, or blocks them entirely. Agent tools based on the Model Context Protocol (MCP) pass through the same control. All interactions are recorded as evidence for audit, together with the masking and routing decisions.

  • Turkish personal and sensitive data detection; consistent, reversible masking
  • Unmasking in the answer only for the authorised user
  • KVKK and BDDK policy sets, institutional policy definition
  • On-premises model / external model / block decision based on content sensitivity (policy-based routing)
  • Authorisation and audit for MCP agent tools
  • Prompt injection and data leakage protection
  • Cost, usage and quality monitoring; audit reports