github.devotd.
NewLLM infrastructure

portkey-ai/gateway v1.15.2 — An AI Gateway Behind a Single API

Features of portkey-ai/gateway, the MIT-licensed AI gateway that connects more than 1,600 LLMs and over 50 AI guardrails behind a single API, along with the absence of changelog details for the v1.15.2 release.

Repositoryportkey-ai/gateway ↗

Features of portkey-ai/gateway, the MIT-licensed AI gateway that connects more than 1,600 LLMs and over 50 AI guardrails behind a single API, along with the absence of changelog details for the v1.15.2 release.

GitHub description: A blazing fast AI Gateway with integrated guardrails. Route to 1,600+ LLMs, 50+ AI Guardrails with 1 fast & friendly API.

stars
★ 13.2K
Forks
1.4K
Language
TypeScript
License
MIT
Latest release
v1.15.2
Read original source ↗View on GitHub ↗

portkey-ai/gateway is an open-source, MIT-licensed AI gateway that connects more than 1,600 LLMs and over 50 AI guardrails through a single API. The README describes the repository as a lightweight AI Gateway for fast, reliable connectivity to language, vision, audio, and image models. The repository was created in August 2023, and the latest release is v1.15.2.

The project's practical value lies in placing multiple model providers behind one API while applying automatic retries and fallback, load balancing, conditional routing, guardrail policies, and multimodal support. Fallback means handing a failed request over to another model path, and load balancing means spreading requests across several paths. The routing destination can be switched based on request conditions, and a separate MCP Gateway handles authentication, access control, and observability for an organization's internal MCP (Model Context Protocol) servers. The implementation language is TypeScript, and the repository has 13,235 GitHub stars.

That said, the provided release information contains no changes for v1.15.2. There is no way to tell what was newly added or what was fixed, and no performance measurements are included either. Of the evaluation scores, only momentum (4.1) is filled in. The breadth of features and the MIT license alone make it hard to expect production-grade performance and stability; judging that requires separate verification.