Skip to content
Wavertex

Open-source AI gateway · Self-hosted

The AI gateway that protects your data before it leaves your perimeter

Cortex sits between your people and language models. It redacts and pseudonymizes what must not leave, restores the values on the way back for their author, and keeps a verifiable audit record of every exchange.

Get startedTalk to the team

Self-hosted · MIT licence · Runs on local models

Works with

  • OpenAI-compatible APIs
  • Anthropic
  • Ollama
  • vLLM
  • OIDC
  • SAML

01 · Protect

Nothing sensitive leaves unseen

Every prompt goes through DLP before any provider. A detection error blocks the request instead of letting it through.

  • Patterns and secrets: structured personal data, vendor keys and tokens
  • Named-entity recognition (Presidio) on the optional node
  • Detection by a local LLM, with nothing sent outside
  • Reversible pseudonymization: the model sees [PERSON_1], the author sees the real name
Cortex screenshot: a conversation whose prompt held a name, an email and a phone number, with the indicator of sensitive items redacted before reaching the model

02 · Govern

A record where any tampering shows

Every exchange is written in the same transaction as the request. No audit, no request.

  • Hash-chained journal under a per-tenant key
  • Integrity verifier and filterable auditor console
  • Full CSV and JSON export
  • Retention policies and two-step confirmed erasure
Cortex screenshot: the audit journal, with chain integrity, filters, and events with their actors and destinations

03 · Control

Your providers, your rules, your costs

Cortex speaks the OpenAI format to your applications and routes to the providers you choose, cloud or local.

  • Ordered failover chains, retries and replica load balancing
  • Encrypted per-tenant response cache
  • Per-request cost estimate and budgets enforced before sending
  • Virtual API keys with rate limits
Cortex screenshot: usage and costs over the last 30 days, grouped by user

04 · Know

Your documents, for the right people

The knowledge base indexes your documents and brings them into the conversation, within each person's access rights.

  • Collections with per-department access
  • Knowledge-manager role
  • Document extraction and embeddings on the optional node
Cortex screenshot: knowledge base collections and their department access

05 · Deploy

On your own servers

Cortex installs with Docker on your own servers. No Wavertex service is required to run it.

  • Docker images and an interactive setup assistant
  • Can run entirely on local models (Ollama, vLLM)
  • Tenant isolation: dedicated PostgreSQL schema and row-level security
  • Encryption at rest with one key per tenant
Terminal
# Public repository opening soon./scripts/setup-dev.shdocker compose up -d --build

06 · Sovereignty

Your data stays under your jurisdiction

Cortex is designed to help you meet Québec's Law 25, the GDPR and PIPEDA.

  • Law 25

    Québec — personal information protection

  • GDPR

    European Union

  • PIPEDA

    Canada — private sector

On the CLOUD Act

Cortex is self-hosted and imposes no US provider: it can run entirely on local models. If you choose a US-hosted model, Cortex redacts and pseudonymizes before anything leaves.

Designed in Québec
Proud member of DSQ

07 · Offers

Start free. Go further soon.

Community

Free

MIT licence · Self-hosted

  • Fail-closed DLP and pseudonymization
  • Tamper-evident audit journal
  • OIDC and SAML SSO, MFA
  • Routing, failover and cache
  • Budgets and virtual keys
  • Knowledge base
Get started

Commercial

Coming soon

A commercial version of Cortex is in preparation for organizations that need more.

Keep me informed

Run it yourself. Starting today.

Install Cortex on your own infrastructure or tell us about your context.

Get startedTalk to the team