Portkey is an AI gateway: a proxy that sits between an application and the providers of large language models (LLMs) and passes every model request through one API. The platform also covers observability (logs and traces of each request), guardrails (checks on model input and output), governance and prompt management. Palo Alto Networks has completed the acquisition of Portkey, and the product is now named PRISMA AIRS AI Gateway.
An application sends requests in an OpenAI-compatible format to the gateway, which routes them to more than 1,600 language, vision, audio and image models. A configuration can retry a failed request up to 5 times, fall back to another provider or model on chosen errors, balance load across models and keys, and cache answers, either exact repeats (simple caching) or similar questions (semantic caching), to reduce cost and latency.
The gateway itself is open source under the MIT license and runs locally with Node.js, in Docker or on Cloudflare. Portkey also sells a hosted service by subscription: a free Developer plan, a paid Production plan and custom Enterprise pricing.