# AgentHub — Production-Grade Multi-Agent Runtime Platform

A production-focused multi-agent runtime platform, built on FastAPI and LangGraph, that dispatches user requests through a supervisor agent to specialized subagents with session isolation, long-term memory and dynamic model switching.

AgentHub positions itself as an agent runtime platform rather than another orchestration framework, aimed at the last-mile problem of moving a multi-agent prototype into production. It adopts a supervisor pattern: a top-level supervisor agent identifies user intent and dispatches the request to the appropriate specialized subagent to complete the task.

It supports high concurrency with FastAPI's async request handling, token-level streaming over server-sent events, and automatic fallback and retry between model providers through a routing layer; multi-user isolation through independent session spaces and LangGraph checkpointer-based state persistence with complete data separation between users; runtime switching between model providers such as OpenAI, Anthropic, Groq or Ollama without restarting a session; and long-term memory built on a LangGraph store with PGVector-based semantic retrieval that persists a user's preferences across sessions.

It deploys with a three-container Docker Compose setup covering the backend, frontend and supporting services, and is aimed at individual developers learning LangGraph, startups validating a multi-agent product, and enterprises building internal assistants and knowledge bases.

Website: https://github.com/realyinchen/AgentHub