Autogen

Multi-Agent Frameworks

Why Multi-Agent?
Single agents struggle with complex tasks. Multi-agent systems break problems down by having specialized agents collaborate, each with distinct roles and capabilities.

Microsoft AutoGen

Overview
Framework for building multi-agent conversational systems with flexible agent orchestration.

Key Concepts
| Concept | Description |
|---------|-------------|
| ConversableAgent | Base agent that can send/receive messages |
| AssistantAgent | LLM-powered agent for reasoning |
| UserProxyAgent | Represents human, can execute code |
| GroupChat | Manages multi-agent conversations |

Example

from autogen import AssistantAgent, UserProxyAgent

# Create agents
assistant = AssistantAgent(
    name="assistant",
    llm_config={"model": "gpt-4o"}
)

user_proxy = UserProxyAgent(
    name="user_proxy",
    code_execution_config={"work_dir": "coding"}
)

# Start conversation
user_proxy.initiate_chat(assistant, message="Analyze sales data")

CrewAI

Overview
Framework for orchestrating role-playing AI agents working together as a "crew" on complex tasks.

Key Components
| Component | Description |
|-----------|-------------|
| Agent | Entity with role, goal, backstory |
| Task | Specific work item for an agent |
| Crew | Group of agents + their tasks |
| Tools | Capabilities agents can use |

Example

from crewai import Agent, Task, Crew

researcher = Agent(
    role="Research Analyst",
    goal="Find accurate information",
    backstory="Expert researcher with attention to detail"
)

writer = Agent(
    role="Content Writer",
    goal="Create engaging content",
    backstory="Experienced writer"
)

research_task = Task(description="Research AI trends", agent=researcher)
writing_task = Task(description="Write article based on research", agent=writer)

crew = Crew(agents=[researcher, writer], tasks=[research_task, writing_task])
result = crew.kickoff()

Comparison
| Feature | AutoGen | CrewAI |
|---------|---------|--------|
| Agent style | Conversational | Role-playing |
| Orchestration | Flexible | Sequential/hierarchical |
| Code execution | Built-in | Via tools |
| Use case | General | Creative workflows |

Multi-Agent Patterns
- Hierarchical: Manager delegates to workers
- Sequential: Agents work in order
- Collaborative: Agents discuss and iterate
- Competitive: Agents propose, vote on solutions