Portkey is an AI gateway and LLMOps platform that puts 1,600+ LLMs behind one unified API, adding observability, guardrails, caching and governance for production GenAI teams.
Integrating every model provider separately does not scale: teams rewrite code to switch models, lose visibility into spend and quality, and ship ungoverned prompts to production. Portkey solves this with a gateway pattern. You integrate once — three lines of code, with no changes to your existing stack — and route every request through Portkey, addressing models with a single signature like @openai/gpt-4.1. Underneath, the gateway fans out to more than 1,600 models across OpenAI, Anthropic, Google, AWS Bedrock, Azure OpenAI, Groq, Together AI and Mistral, with automatic fallbacks and load balancing when a provider degrades. In May 2026, Portkey was acquired by Palo Alto Networks and folded into the Prisma AIRS platform as its AI Gateway; the open-source gateway remains MIT-licensed and the hosted plans continue unchanged.
The gateway is only the entry point. Portkey layers a full production stack on top: real-time observability with cost tracking and anomaly alerts, 50+ guardrails covering PII redaction and jailbreak detection, semantic caching that can save thousands of dollars on repeated calls, prompt management without hard-coded strings, and RBAC-governed workspaces. The open-source gateway core is MIT-licensed for self-hosting, and since 2026 Portkey has operated as Palo Alto Networks’ PRISMA AIRS AI Gateway, positioning it for enterprises that need agent governance alongside network-grade security.
The Developer tier is free forever with 10,000 logged requests per month. Production costs $49/month for 100,000 logged requests plus $9 per additional 100,000, with longer retention and advanced guardrails. Enterprise is custom for 10M+ requests per month, VPC hosting and dedicated support. Token costs themselves are never marked up — you pay provider rates plus the gateway subscription.
Teams shipping AI products who need provider flexibility, cost visibility and governance in one place. Hobby developers with trivial volume can stay on the free Developer tier, but the observability caps make it a non-starter for real production workloads.