Multi Agent Research POC

🚀 Multi-Agent Research Proof-of-Concept Project Compatible with AutoGen (Using Ollama and BraveSearch)

This project is a proof-of-concept for a locally prioritized multi-agent system, utilizing the following technologies:

• 🤖 Local LLM powered by Ollama
• 🧩 think... syntax for tool call detection
• 🔍 Web search via Brave Search API or Brave MCP Plugin Server
• 🧠 Two collaborative agents: Searcher and Synthesizer

📁 File Structure

Multi-Agent-Research-POC/
├── main.py                   # Entry point
├── agents/
│   ├── searcher.py           # Ollama-based research agent
│   └── synthesizer.py        # Summary generator agent
├── tools/
│   ├── tool_parser.py        # Tool call detection logic
│   └── tool_registry.py      # Tool dispatcher (API or MCP)
├── .env                      # Contains BRAVE_API_KEY
└── requirements.txt          # Python dependencies

🚀 Quick Start

1. Clone the project

git clone <your-repo-url>
cd Multi-Agent-Research-POC

2. Install dependencies

pip install -r requirements.txt

3. Set up your .env

echo "BRAVE_API_KEY=your_brave_api_key_here" > .env

Get your Brave API key: https://developer.brave.com/api-search/

4. Run Ollama locally

ollama run llama3:8b

If using Docker, make sure to set the host to http://host.docker.internal:11434

5. Run the program

python main.py

You should see:

• Responses from the Searcher agent
• Tool call triggers
• Search results fetched from Brave
• Final summaries generated by the Synthesizer agent

🔄 Switching between API and MCP Plugin

Option 1: Brave Search API (Default)

Use it as follows:

"BraveSearch": call_brave_api

Option 2: Brave MCP Plugin Server

1. Start the plugin server:

npx @modelcontextprotocol/server-brave-search

2. Update tools/tool_registry.py:

# "BraveSearch": call_brave_api,
"BraveSearch": call_brave_mcp_server

🚧 Next Improvements

💻 Example Output

# Search section
Search results: [...] Results related to the query [...]

# Summary generation section
Summary: [...] Key information extraction and organization [...]

Refer to the project documentation for more detailed output examples.

🙏 Acknowledgments

Thanks to the contributions of the following projects:

• Ollama - https://ollama.ai/
• AutoGen - https://github.com/HuggingFace/auto-gpt2
• Brave Search - https://brave.com/search/

📄 Project Statement

This project is for research purposes only and does not involve any commercial use. Please comply with applicable laws, regulations, and privacy policies regarding relevant data and results.