TLDR: Co-TAP is a three-layer agent interaction protocol (HAI, UAP, MEK) designed to overcome challenges in multi-agent systems. HAI standardizes human-agent interaction for real-time, reliable communication. UAP enables seamless interoperability and communication between diverse agents through a unified registry and protocol conversion gateway. MEK facilitates collective intelligence by standardizing memory management, extracting valuable insights, and sharing knowledge across agents. Together, these protocols form a synergistic framework for building scalable and intelligent multi-agent applications.
Multi-agent systems, where multiple AI agents work together to solve complex problems, are becoming increasingly important with the rise of Large Language Models (LLMs). However, these systems often face significant hurdles, such as agents struggling to communicate with each other, inefficient interactions with human users, and difficulties in sharing learned experiences and knowledge. These issues lead to what researchers call ‘information silos’ and hinder the development of true collective intelligence.
To tackle these challenges, the Co-TAP Team has introduced Co-TAP (Triple Agent Protocol), a comprehensive three-layer agent interaction protocol. This framework aims to provide a robust foundation for building the next generation of efficient, scalable, and intelligent multi-agent applications. You can read the full technical report here: Co-TAP: Three-Layer Agent Interaction Protocol.
The Three Pillars of Co-TAP
Co-TAP is built upon three core protocols, each addressing a specific dimension of multi-agent system challenges:
1. HAI: Human-Agent Interaction Protocol
The HAI protocol operates at the interaction layer, focusing on how humans and agents communicate. It standardizes the flow of information between users, interfaces, and agents, ensuring real-time, reliable, and synergistic interactions. Imagine a seamless conversation where an agent can respond instantly, and you can intervene or control its tasks at any point. HAI achieves this through an event-driven communication paradigm, using a unified JSON event stream. This allows for features like token-by-token real-time streaming, dynamic updates to user interfaces based on agent actions, and the ability for agents to use external tools while keeping the user in the loop. It essentially makes human-agent collaboration more intuitive and trustworthy.
2. UAP: Unified Agent Protocol
At the infrastructure layer, the UAP protocol is designed to break down communication barriers between diverse agents. In today’s multi-agent landscape, agents are often built using different frameworks and communication protocols, leading to high adaptation costs and fragmented ecosystems. UAP acts as a universal translator and connector, creating an ‘Internet of Agents.’ It uses a unified service registry and a smart gateway that can convert between various agent protocols (like A2A, ACP, and MCP). This modular design allows agents to discover each other autonomously, communicate seamlessly regardless of their underlying technology, and collaborate on complex tasks without redundant information exchange. UAP ensures that agents can ‘plug-and-play’ and work together efficiently.
3. MEK: Memory-Extraction-Knowledge Protocol
The MEK protocol operates at the cognitive layer, focusing on how agents learn, accumulate experience, and share knowledge. Individual agents can learn from their interactions, but sharing this learning across a collective has been a major hurdle. MEK establishes a standardized ‘Memory (M) – Extraction (E) – Knowledge (K)’ cognitive chain. The Memory module efficiently stores and organizes multimodal experiences. The Extraction stage intelligently identifies and pulls out valuable information from these memories. Finally, the Knowledge stage distills this personalized information into reusable, shareable knowledge that can be absorbed by other agents. This process allows agents to transform isolated experiences into systematic cognitive abilities, fostering true collective intelligence where the entire system continuously learns and improves.
Also Read:
- Unlocking Realistic Social Dynamics: A Modular Approach to Agent Simulation
- The Science of LLM Teamwork: Measuring Emergent Coordination
A Collaborative Loop for Collective Intelligence
The true power of Co-TAP lies in the synergy of these three protocols. They work together in a continuous loop: HAI provides the interface for human interaction and feedback, UAP enables the underlying communication and collaboration between heterogeneous agents, and MEK ensures that the experiences gained during these interactions are transformed into shareable knowledge, which then enhances the capabilities of all agents in the system. This integrated approach allows multi-agent systems to move beyond simple aggregation of computational power to achieve human-like collaboration and emergent behaviors, leading to a new paradigm of problem-solving.
The Co-TAP framework, developed by Shunyu An, Miao Wang, Yongchao Li, Dong Wan, Lina Wang, Ling Qin, Liqin Gao, Congyao Fan, Zhiyong Mao, Jiange Pu, Wenji Xia, Dong Zhao, Rui Hu, Ji Lu, Guiyue Zhou, Baoyu Tang, Yanqin Gao, Yongsheng Du, Daigang Xu, Lingjun Huang, Baoli Wang, Xiwen Zhang, Luyao Wang, and Shilong Liu, offers a solid engineering foundation and theoretical guidance for building the next generation of intelligent multi-agent applications.
