Portkey AI Gateway

Over 1,600 models from providers including OpenAI, Anthropic, Google, Bedrock, and local deployments sit behind one unified API in Portkey's gateway, which the company open sourced under MIT after running it in production for its hosted platform. The TypeScript service starts locally with a single npx command or deploys via Docker and edge runtimes, and it layers production plumbing over raw provider APIs: automatic retries and fallbacks, load balancing across keys and providers, conditional routing, request caching, and more than 50 integrated guardrails that can validate or block inputs and outputs in the request path. Configs are declarative JSON, so switching a workload between providers means changing a route rather than code. The hosted Portkey platform adds observability, prompt management, and virtual keys, with a free developer tier that records up to 10,000 logs per month and paid plans above it. With roughly 12,600 GitHub stars, the gateway is a common first piece of LLM infrastructure for teams needing reliability across vendors.