Abstract
The rapid advancement of large language models (LLMs) has paved the way forthe development of highly capable autonomous agents. However, existingmulti-agent frameworks often struggle with integrating diverse capablethird-party agents due to reliance on agents defined within their ownecosystems. They also face challenges in simulating distributed environments,as most frameworks are limited to single-device setups. Furthermore, theseframeworks often rely on hard-coded communication pipelines, limiting theiradaptability to dynamic task requirements. Inspired by the concept of theInternet, we propose the Internet of Agents (IoA), a novel framework thataddresses these limitations by providing a flexible and scalable platform forLLM-based multi-agent collaboration. IoA introduces an agent integrationprotocol, an instant-messaging-like architecture design, and dynamic mechanismsfor agent teaming and conversation flow control. Through extensive experimentson general assistant tasks, embodied AI tasks, and retrieval-augmentedgeneration benchmarks, we demonstrate that IoA consistently outperformsstate-of-the-art baselines, showcasing its ability to facilitate effectivecollaboration among heterogeneous agents. IoA represents a step towards linkingdiverse agents in an Internet-like environment, where agents can seamlesslycollaborate to achieve greater intelligence and capabilities. Our codebase hasbeen released at \url{https://github.com/OpenBMB/IoA}.