Beyond Rule-Based Workflows: An Information-Flow-Orchestrated Multi-Agents Paradigm via Agent-to-Agent Communication

Most existing LLM-based multi-agent systems (MAS) are built on predefined workflows, where engineers manually enumerate task states and specify routing rules in advance. Such workflow-driven designs essentially form rule-based decision trees, which are brittle, labor-intensive to design, and incapable of exhaustively covering complex real-world task states.

This repository introduces an Information-Flow-Orchestrated Multi-Agent Paradigm based on Agent-to-Agent (A2A) communication from CORAL. Instead of relying on predefined workflows, a dedicated information flow orchestrator continuously monitors task progress and dynamically coordinates other agents via natural-language A2A communication.

Key Features

• 🚫 Workflow-free coordination — no predefined decision trees or routing rules
• 🔄 Dynamic task monitoring via a centralized information flow orchestrator
• 🧠 Natural-language A2A communication for flexible agent collaboration
• 🧩 Robust handling of edge cases in complex tasks

Evaluation

We evaluate the proposed paradigm on the GAIA benchmark, using the representative workflow-based MAS OWL as the baseline while controlling for agent roles and underlying models.

• Pass@1 Accuracy: 63.64% (vs. OWL’s 55.15%, +8.49%)
• Token Consumption: Nearly identical to OWL
• Qualitative Results: More flexible task monitoring and improved robustness in edge-case scenarios

🧪 Evaluation in GAIA

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 OPENAI_API_KEY="your_openai_api_key"
export SEARCH_ENGINE_ID="your_search_engine_id"
export CHUNKR_API_KEY="your_chunkr_api_key"
' >> ~/.bashrc && source ~/.bashrc

Create environment:

cd agent_defined_workflow
/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:

cd agent_defined_workflow
rm -rf venv/lib/python3.12/site-packages/camel
cp -r utils/camel venv/lib/python3.12/site-packages/

Run the experiment:

cd agent_defined_workflow
./run.sh