Anemoi

# Anemoi: A Semi-Centralized Multi-agent Systems Based on Agent-to-Agent Communication MCP server from Coral Protocol **Anemoi** is a semi-centralized multi-agent system (MAS) built on a **Agent-to-Agent (A2A) communication MCP server**. Unlike traditional **context-engineering + centralized** paradigms, Anemoi introduces **structured, direct inter-agent communication** — enabling agents to collaborate much like a real-world team. 🌀 *Like winds connecting distant lands, Anemoi enables agents to communicate directly in a semi-centralized network, achieving scalable coordination and seamless information flow.* <p align="center"> <img src="Anemoi/images/Anemoi_semi.png" alt="Anemoi Concept" width="70%"> </p> --- ## 🚀 Key Features * **Semi-Centralized Architecture** Reduces dependency on a single planner agent, supporting adaptive plan updates. * **Direct Agent-to-Agent Collaboration** Agents can monitor progress, assess results, identify bottlenecks, and propose refinements in real time. * **Efficient Context Management** Minimizes redundant prompt concatenation and information loss, improving scalability and cost-efficiency. * **Benchmark Performance** Achieved **52.73% accuracy** on the validation set of the GAIA benchmark, setting the **state-of-the-art** among small-LLM-based systems. > Surpasses OWL (43.63%) by **+9.09%** in the same worker agents and models configuration (gpt-4.1-mini as planner agent/ gpt-4o as worker agent). <p align="center"> <img src="Anemoi/images/Anemoi_workflow.png" alt="Anemoi Workflow" width="85%"> </p> --- ## 📄 Publication Our work has been released on arXiv: 👉 [Anemoi: A Semi-Centralized Multi-agent Systems Based on Agent-to-Agent Communication MCP server from Coral Protocol](https://arxiv.org/abs/2508.17068) If you find this project useful, please consider citing our paper: ``` @article{ren2025anemoi, title={Anemoi: A Semi-Centralized Multi-agent Systems Based on Agent-to-Agent Communication MCP server from Coral Protocol}, author={Ren, Xinxing and Forder, Caelum and Zang, Qianbo and Tahir, Ahsen and Georgio, Roman J. and Deb, Suman and Carroll, Peter and Gürcan, Önder and Guo, Zekun}, journal={arXiv preprint arXiv:2508.17068}, year={2025}, url={https://arxiv.org/abs/2508.17068} } ``` --- ## 🧪 Reproduction Set up environment variables: ``` echo ' export FIRECRAWL_API_KEY="your_firecrawl_api_key" export GOOGLE_API_KEY="your_google_api_key" export HF_HOME="your_hf_home_path" export OPENROUTER_API_KEY="your_openrouter_api_key" export SEARCH_ENGINE_ID="your_search_engine_id" export CHUNKR_API_KEY="your_chunkr_api_key" ' >> ~/.bashrc && source ~/.bashrc ``` Create environment: ``` cd Anemoi /usr/bin/python3.12 -m venv venv source venv/bin/activate pip install -r requirements.txt ``` We made some minor modifications to CAMEL 0.2.70 for our experiments: ``` rm -rf venv/lib/python3.12/site-packages/camel cp -r utils/camel venv/lib/python3.12/site-packages/ ``` Run the experiment: ``` cd .. ./gradlew run --console=plain ```