> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.lakera.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.lakera.ai/_mcp/server.

# Self-hosting AI Guardrails

Self-hosting Check Point AI Guardrails allows organizations to keep all data within
their own infrastructure. AI Guardrails can be deployed on-premises or in a private
cloud, giving you full control over data residency and network access.

## Deployment Options

AI Guardrails is available as a fully managed
[SaaS solution or as a self-hosted deployment](/guard#deployment-options). For
self-hosted customers, we support flexible deployment models to fit your infrastructure:

| Deployment Method     | Description                                                                                                                                                 |
| --------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Kubernetes (Helm)** | Deploy to any Kubernetes cluster using our official Helm chart from Docker Hub. Includes support for autoscaling, TLS, health probes, and GPU acceleration. |
| **Docker**            | Run the AI Guardrails container directly with Docker or any OCI-compatible runtime for simpler setups.                                                      |
| **Air-gapped**        | Full offline deployment support for the most restrictive environments. Containers can be exported and loaded without internet access.                       |

## What's Included

The self-hosted deployment includes the following components:

* **AI Guardrails** - The core security screening service for text-based threat
  detection
* **API Gateway** - Request routing and prompt chunking for high-throughput workloads
* **Triton Inference Server** - GPU-accelerated inference for text classifiers
* **Triton TensorRT-LLM** - GPU-accelerated inference for Audio Guard

## Key Capabilities

* **GPU Support** - Leverage NVIDIA GPUs for low-latency inference.
  [Contact us](https://www.lakera.ai/contact) for current hardware recommendations for
  your deployment
* **Horizontal Scaling** - Stateless containers that scale horizontally
* **Policy Management** - Configure guardrails via policy files
* **Audio Defense** - Screen audio inputs for audio-based attacks
* **TLS/SSL** - Encrypted communication between components
* **Container Versioning** - Stable, nightly, and version-pinned releases

## Getting Started

To deploy AI Guardrails in your environment, you will need:

* A valid [AI Guardrails Enterprise license](https://platform.lakera.ai/pricing)
* Container registry credentials (provided by Check Point)
* A Kubernetes cluster or Docker-compatible host

> **Note**
>
> Comprehensive deployment documentation — including resource requirements, sizing, latency benchmarks, step-by-step Helm chart installation, GPU configuration, autoscaling, security hardening, Audio Guard setup, and troubleshooting — is available at [self-hosted.docs.lakera.ai](https://self-hosted.docs.lakera.ai).
>
> Access to the self-hosted documentation portal is provided to AI Guardrails customers.
> If you are an existing customer, please contact
> [support@lakera.ai](mailto:support@lakera.ai) for access. If you are interested in
> self-hosting AI Guardrails, please [contact us](https://www.lakera.ai/contact).