plano/demos/agent_orchestration/multi_agent_ag2/README.md

AG2 Multi-Agent Research Team — with Plano

What you'll see: A research assistant powered by AG2's multi-agent GroupChat — a team of specialized agents (researcher + analyst) collaborating behind a single Plano endpoint, with unified routing, orchestration, and observability.

The Problem

Building multi-agent systems today forces developers to:

• Pick one framework - can't mix AG2, CrewAI, LangChain, or custom agents easily
• Write plumbing code - authentication, request routing, error handling
• Rebuild for changes - want to swap frameworks? Start over
• Limited observability - no unified view across different agent frameworks

Most demos also show one agent per endpoint. But real-world tasks need multiple agents collaborating internally — and that internal collaboration should be transparent to the orchestration layer.

Plano + AG2 Solution

Plano acts as a framework-agnostic proxy and data plane that:

• Routes requests to the AG2 research team endpoint
• AG2 internally orchestrates a GroupChat (researcher → analyst)
• Provides unified tracing across the entire multi-agent flow
• Keeps internal agent collaboration transparent to the orchestration layer

How It's Different from the CrewAI/LangChain Demo

In the CrewAI/LangChain demo, each framework = one agent = one endpoint. Plano routes between them at the framework level.

Here, AG2 runs a team of agents behind a single endpoint. The researcher and analyst collaborate internally via AG2's GroupChat — Plano sees it as one agent, while the multi-agent work happens inside. This shows that Plano can orchestrate both simple agents and complex multi-agent teams equally well.

How To Run

Prerequisites

1. Install uv (Python package manager)

   curl -LsSf https://astral.sh/uv/install.sh | sh
   

2. Install Plano CLI

   uv tool install planoai
   

3. Install demo dependencies (includes [ag2] and FastAPI)

   cd demos/agent_orchestration/multi_agent_ag2
   uv sync
   

4. Set Environment Variables

   export OPENAI_API_KEY=your_key_here
   

Start the Demo

# From the demo directory
cd demos/agent_orchestration/multi_agent_ag2
./run_demo.sh

This installs dependencies, starts Plano natively, and runs the AG2 research agent as a local process:

• AG2 Research Agent (port 10530) — researcher + analyst GroupChat

Plano runs natively on the host (ports 12000, 8001).

Try It Out

1. Using curl

   curl -X POST http://localhost:8001/v1/chat/completions \
     -H "Content-Type: application/json" \
     -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Research the current state of AI agents in production"}]}'
   

2. More research queries

   "Analyze the trade-offs between RAG and fine-tuning for enterprise LLM applications"
   "What are the key trends in multi-agent AI systems in 2025?"
   "Research the pros and cons of different vector database solutions"
   

   Plano routes the request to the AG2 endpoint, where the researcher and analyst collaborate internally before returning a unified response.

Architecture

┌──────────────┐
│    Client    │
└──────┬───────┘
       │
       v
┌─────────────┐
│    Plano    │  (Routing & Observability)
└──────┬──────┘
       │
       v
┌──────────────────────┐
│  AG2 Research Agent  │  (Port 10530)
│  ┌────────────────┐  │
│  │   GroupChat    │  │
│  │  ┌──────────┐  │  │
│  │  │Researcher│  │  │
│  │  └────┬─────┘  │  │
│  │       │        │  │
│  │  ┌────▼─────┐  │  │
│  │  │ Analyst  │  │  │
│  │  └──────────┘  │  │
│  └────────────────┘  │
└──────────────────────┘
       │
       v
┌─────────────┐
│    Plano    │  (Proxy LLM calls)
└──────┬──────┘
       │
       v
┌─────────────┐
│   OpenAI    │
└─────────────┘

AG2 Research Team

Research Agent

• Framework: AG2 (formerly AutoGen)
• Architecture: Multi-agent GroupChat
• Agents:
  • Researcher — gathers detailed, factual information on any topic
  • Analyst — synthesizes findings into actionable insights and recommendations
• Key Feature: Internal multi-agent collaboration behind a single HTTP endpoint

Cleanup

./run_demo.sh down

Next Steps

• Add more agents — extend the GroupChat with a fact-checker, domain expert, or summarizer
• Add tools — register web search or database tools with @agent.register_for_llm() + @agent.register_for_execution()
• Mix frameworks — combine this AG2 team with CrewAI or LangChain agents in a single Plano config
• Production deployment — see Plano docs for scaling guidance

Learn More

• Plano Documentation
• AG2 Documentation