Prisma® AIRS™ AI Gateway sits between your users and applications and every large language model (LLM) provider they call. Without a gateway, AI traffic flows directly to providers with no visibility, no access controls, and no way to enforce consistent security policy. AI Gateway solves this by acting as a single proxy through which all LLM requests pass, giving you a complete record of what was asked, who asked it, what the model returned, and what it cost.

AI Gateway connects to more than 1,600 LLMs across 50 or more providers — including OpenAI, Anthropic, Google, AWS Bedrock, Azure OpenAI, Mistral, Cohere, and self-hosted models — through a single API endpoint. Your developers send requests to AI Gateway using the standard OpenAI API format, and the gateway routes each request to the correct provider. Switching providers or adding fallback models requires no application code changes.

The gateway is built on workspaces, which are logical containers that let you manage LLM access, budgets, rate limits, and guardrail policies per team or application. Each workspace issues a virtual key — a scoped credential that controls which LLMs the workspace can reach and how much it can spend. Guardrails run inline on every request and response, blocking prompt injection attempts, detecting sensitive data, and enforcing topic restrictions before traffic reaches the provider or your users.

AI Gateway is available in two deployment models: SaaS, where Palo Alto Networks hosts the gateway data plane on your behalf, and Hybrid, where you host the data plane in your own Kubernetes cluster so that request and response payloads never leave your environment. Both models are managed from Strata Cloud Manager. For details on choosing a model, see AI Gateway Deployment Models.