Skip to content
Wavertex
Documentation menu

Cortex overview

Verified with Cortex on October 2, 2026

Cortex is a self-hosted AI gateway. It sits between the people in your organization and language models, whether they are hosted in the cloud or on your own servers.

What Cortex does

Every exchange follows the same path:

  1. Authentication. The person signs in with your organization's identity provider (OIDC or SAML) or with a local account protected by MFA.
  2. Redaction. The prompt goes through DLP before any provider. Sensitive values are replaced by pseudonyms such as [PERSON_1]. If DLP fails, the request is blocked.
  3. Routing. Cortex sends the redacted prompt to the model chosen by the administrator, with failover when a provider does not answer.
  4. Rewriting. The reply is rewritten before it is displayed or stored: the person who wrote the prompt gets their own values back.
  5. Audit. The exchange is written to the audit journal in the same transaction. A request that cannot be audited does not happen.
Cortex screenshot: the administration console activity dashboard, with request volume, redacted items and redactions by entity type
The administration console: your organization's activity, with no conversation content.

Components

ComponentRole
Web interfaceConversation, administration console, auditor console
GatewayAuthentication, DLP, pseudonymization, routing, audit
PostgreSQLOne schema per tenant, content encrypted at rest
AI node (optional)Entity recognition, local-LLM detection, knowledge base

Pseudonymization

Within a conversation, a value always keeps the same pseudonym: the model can reason about [PERSON_1] without ever seeing the name. The mapping between pseudonyms and real values never leaves the gateway.