Prisma AIRS AI Gateway

What it does

Prisma AIRS AI Gateway, formerly Portkey, sits between an application and model providers. Requests go in as LLM calls, typically chat completions through an OpenAI-compatible API, and responses come back to the app. Each request is recorded with logs, traces, token counts, latency and cost. The gateway also routes MCP tool calls and agent-to-agent traffic.

Teams use it to put one control point in front of many models. Developers can configure fallbacks, load balancing, retries, caching and prompt templates. Platform and security teams can apply role-based access controls, budgets, rate limits and PII redaction. Integration is typically a change to the SDK base URL, with Python, Node.js and OpenAI SDK examples plus a cURL REST option. Palo Alto Networks completed its acquisition of Portkey on May 29, 2026, and the gateway is now sold as Prisma AIRS AI Gateway.

It suits engineering and AI platform teams shipping production LLM applications, and enterprises that need governance, SSO and compliance controls. Access is through a Portkey account and API key, and the open-source gateway can be self-hosted. Current plan details for the renamed offering should be confirmed directly with Palo Alto Networks, since older Portkey plan terms no longer describe it. Logs, retention and feature availability depend on the deployment and contract.

What it can do

• Unified API access to over 1,600 LLMs through one interface
• Routing with fallbacks, load balancing, retries and timeouts
• Caching, including simple and semantic caching
• Observability with logs, traces, feedback, metadata and alerts
• Prompt management with templates, playground, versioning and variables
• Governance with RBAC, budgets, rate limits and PII redaction

Works with

• OpenAI Python SDK
• OpenAI Node.js SDK
• Model Context Protocol (MCP) servers via MCP Gateway
• Okta SSO and custom OIDC SSO providers

Example workflows

Route a chat request with fallback

Input: an application's chat-completion request using the OpenAI SDK pointed at the gateway with a gateway API key and provider. Steps: the gateway forwards the request to the configured provider and applies retries, timeouts or fallbacks if configured. Output: a model response returned to the app, with the request logged for cost, tokens and latency.

Review usage and costs across teams

Input: requests from several projects or teams sent through the gateway with metadata attached. Steps: open the dashboard, filter logs by metadata, and review token counts, latency and cost; set budgets or rate limits where needed. Output: a per-team view of usage and spend, with alerts when limits are approached.

Before you choose

• Gateway adds latency per request compared with direct API calls; the docs cite an estimate of 20-40 ms.

Listed under

• Security
• AI and agent security
• AI gateway
• Developers
• Enterprise governance
• Guardrails
• LLM observability
• Prompt management

Confirm the current Prisma AIRS AI Gateway contract and deployment terms. Historical Portkey subscriptions are not presented as confirmed pricing for the renamed offering.

We couldn’t find published prices for Prisma AIRS AI Gateway. See their pricing page ↗