vLLM Agentic API

vLLM Agentic API is a Rust-based gateway that adds stateful agentic capabilities to vLLM inference. It handles conversation state, tool execution, and multi-turn orchestration server-side, letting developers run agent workloads on open models with their own GPUs. The system supports OpenAI-compatible Responses API, built-in web search, Model Context Protocol tools, and streaming via HTTP, SSE, and WebSockets.

Designed for developers building agent applications, it moves complexity from client code to the server. The maker indicates it was built with Claude Code. It integrates with vLLM, SGLang, and NVIDIA Dynamo inference engines, giving users full control over the inference stack.