🧠 AutoGen-Compatible Multi-Agent Research POC with Ollama + BraveSearch

🧠 AutoGen-Compatible Multi-Agent Research POC with Ollama + BraveSearch

This project is a proof of concept for a local-first multi-agent system using:

• 🤖 Local LLMs via Ollama
• 🧩 Tool-call detection using <tool_call>... syntax
• 🔍 Web search via Brave Search API or Brave MCP plugin server
• 🧠 Two collaborating agents: Searcher and Synthesizer

---

📁 Folder Structure

MultiResearchPOC/
├── main.py                   # Entry point
├── agents/
│   ├── searcher.py           # Ollama-powered research agent
│   └── synthesizer.py        # Summarizer 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

---

🚀 Getting Started

1. Clone the project

git clone <your-repo-url>
cd MultiResearchPOC

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 at: https://developer.brave.com/api-search/

4. Run Ollama locally

ollama run llama3:8b

If using Docker: make sure to reference the host as http://host.docker.internal:11434

5. Run the program

python main.py

You should see:

• A response from the Searcher agent
• A tool call triggered
• Search results pulled from Brave
• A final summary from the Synthesizer agent

---

🔁 Switching Between API and MCP Plugin

Option 1: Brave Search API (default)

Used by default via:

"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 Steps & Improvements

Feature	Description
🧠 Add Planner Agent	Dynamically decide which agent/tool to call
🧩 Add More Tools	CrunchbaseSearch, TwitterTrends, YouTubeSearch, etc.
📄 Markdown Output	Save session logs for review or integration with Obsidian
🖼️ Add UI	Use Chainlit, FastAPI, or Discord bot for interaction
🌐 Wrap as API	Convert to a local API for web or CLI usage

---

📜 Sample Output

🤖 Searcher Response:
<tool_call>BraveSearch({"query": "African AI startups 2024 promising not mainstream"})</tool_call>

🌐 Tool Output:
• AI 100: ...
• Five African AI startups to watch in 2023 ...

🧠 Final Summary:
- CB Insights lists top private AI companies...
- 5 African startups solving problems in healthcare, marketing...

---

🙌 Credits

• Ollama
• Brave Search API
• Semantic Kernel
• Inspired by AutoGen

---

For questions or ideas, open an issue or start a discussion!

🏗️ Built for the Microsoft AI Agents Hackathon This project was created as part of the Microsoft AI Agents Hackathon — a challenge focused on building intelligent, tool-using, autonomous agents powered by open-source and Microsoft technologies.

The goal of this project is to showcase a local-first, multi-agent system that can:

Generate dynamic tool calls

Perform live web research using Brave Search

Collaborate between agents to synthesize useful insights

🔗 Submission: [TBD]

local multi-agent AI research bot | Ollama + Brave + AutoGen | Built for Microsoft AI Agents Hackathon

Disclaimer

This project is a personal proof-of-concept developed entirely outside of my employment, using personal time and tools. It is unrelated to any current or anticipated business activities of my employer and contains no proprietary or confidential information.