Skip to content
Rexaris AI

On-prem AI security: nothing about your prompts leaves your infrastructure

Regulated organisations cannot route employee prompts through a vendor's cloud to make them safe. Rexaris AI installs inside your perimeter, uses your provider keys and keeps policies, audit logs and masked values in your own database.

Most AI security products are SaaS: your traffic goes to them first. For banks, payment companies, government bodies and critical infrastructure that is often a non-starter, whatever the vendor's certifications say.

Rexaris AI is designed as a self-hosted control plane. It runs on Docker or Kubernetes you operate, stores data in your PostgreSQL, calls OpenAI and Anthropic with your credentials, and can serve only internal models when the environment is isolated.

Request a Private Pilot

What on-prem means here

  • 100% inside your infrastructure

    Application, database, policies and audit logs run where you decide. Updates are images you pull, not a service you depend on.

  • Bring your own keys

    Provider credentials are configured by you and never shared with us.

  • No vendor data path

    Customer prompts and responses do not pass through our cloud. There is no telemetry of prompt content.

  • Air-gapped ready

    The architecture supports restricted networks: internal LLMs only, no outbound calls, offline licensing.

Deployment in three steps

  • Deploy

    Install with Docker or Kubernetes inside your network.

    Learn more

  • Connect

    Add OpenAI, Anthropic and internal LLM endpoints such as Qwen, Llama, vLLM or Ollama.

    Learn more

  • Enforce

    Define policies and route corporate AI traffic through the gateway, rule by rule.

    Learn more

  • Keep sensitive work internal

    Confidential data classes can be answered by internal models only.

    Learn more

Frequently asked questions

Does Rexaris AI need internet access?
Only to reach the external providers you choose to allow. In an isolated environment it can run with internal models exclusively.
What infrastructure is required?
A container runtime (Docker or Kubernetes), PostgreSQL and network access from clients to the gateway. Sizing depends on request volume; a single node handles a pilot.
How are updates delivered?
As versioned container images that you pull and roll out on your schedule, including into air-gapped registries.
Do you offer a cloud version?
The product is built on-prem first. Private cloud or VPC deployments in your own account follow the same model: your infrastructure, your keys, your data.

Related solutions

From the blog

Run the pilot inside your own perimeter

A small team, your providers, your database. Validate policies on real workflows before anything scales.

Request a Private Pilot