Previous Card
Diverse And Private Synthetic Datasets Generation for RAG evaluation: A multi-agent framework
Anemoi is a novel semi-centralized multi-agent system (MAS) designed to overcome limitations of traditional centralized MAS, such as strong planner dependency and restricted inter-agent communication. It leverages an Agent-to-Agent (A2A) communication MCP server from Coral Protocol to enable direct, structured collaboration among agents. This approach enhances scalability, reduces costs, and allows for adaptive plan updates, demonstrating superior performance on the GAIA benchmark with a smaller LLM as the planner. ✨
Article Points:
1
Anemoi is a semi-centralized MAS built on A2A communication MCP server.
2
Reduces reliance on a single planner, performing well even with small LLMs.
3
Enables direct, structured inter-agent collaboration and real-time monitoring.
4
Supports adaptive plan updates and minimizes redundant context passing.
5
Achieved 52.73% accuracy on GAIA with GPT-4.1-mini planner.
6
Outperformed the OWL baseline by +9.09% under identical LLM settings.
Problem Addressed
Strong planner dependency
Limited inter-agent communication
Costly context passing
Solution Overview
A2A communication MCP server
Direct agent collaboration
Semi-centralized paradigm
Architecture Components
Planner Agent
Critique Agent
Answer-Finding Agent
Worker Agents
MCP Server Operations
Key Advantages
Reduces planner reliance
Adaptive plan refinement
Minimizes context redundancy
Scalable & cost-efficient
Performance on GAIA
52.73% accuracy
GPT-4.1-mini planner
Outperforms OWL by +9.09%
Remaining Challenges