Building a Multi-Agent System

Build a team of agents that work together - each specializing in one task for better results.

What We’ll Build

A content creation team with 3 agents:

Researcher

Finds information

Writer

Creates content

Editor

Polishes the result

Step-by-Step Guide

Create the Agents

from praisonaiagents import Agent, AgentTeam

# Agent 1: Researcher
researcher = Agent(
    name="Researcher",
    instructions="Research topics thoroughly and provide key facts",
    web=True  # Can search the web
)

# Agent 2: Writer
writer = Agent(
    name="Writer",
    instructions="Write engaging content based on research"
)

# Agent 3: Editor
editor = Agent(
    name="Editor",
    instructions="Polish content for clarity and grammar"
)

Create the Team

# Combine agents into a team
team = AgentTeam(
    agents=[researcher, writer, editor],
    process="sequential"  # One after another
)

Run the Team

result = team.start("Create a blog post about the benefits of exercise")
print(result)

Complete Example

from praisonaiagents import Agent, AgentTeam

# Create specialized agents
researcher = Agent(
    name="Researcher",
    instructions="""You research topics thoroughly.
    - Find key facts and statistics
    - Identify main points
    - Organize information clearly""",
    web=True
)

writer = Agent(
    name="Writer",
    instructions="""You write engaging content.
    - Use research provided
    - Write clear, readable text
    - Include introduction and conclusion"""
)

editor = Agent(
    name="Editor",
    instructions="""You polish and improve content.
    - Fix grammar and spelling
    - Improve clarity and flow
    - Ensure consistent tone"""
)

# Create the team
team = AgentTeam(
    agents=[researcher, writer, editor],
    process="sequential"
)

# Run the team
result = team.start("Create a blog post about renewable energy")
print(result)

How It Works

Step	Agent	What Happens
1	Researcher	Gathers information on the topic
2	Writer	Creates content from research
3	Editor	Polishes the final result

Different Team Patterns

Parallel Team

All agents work at the same time when each Task sets async_execution=True inside a workflow process and the team is started with await team.astart() — the sync team.start() runs them sequentially and warns:

import asyncio
from praisonaiagents import Agent, Task, AgentTeam

team = AgentTeam(
    agents=[analyst1, analyst2, analyst3],
    tasks=[
        Task(description="Analysis A", agent=analyst1, async_execution=True),
        Task(description="Analysis B", agent=analyst2, async_execution=True),
        Task(description="Analysis C", agent=analyst3, async_execution=True),
    ],
    process="workflow"  # parallel fan-out via async_execution on tasks
)

async def main():
    await team.astart()   # parallel fan-out — sync .start() would run sequentially and warn

asyncio.run(main())

async_execution=True only runs in parallel under await team.astart() / await team.arun(). The sync team.start() / team.run() runs those tasks sequentially and logs a warning — see PraisonAI PR #5227.

Hierarchical Team

Manager coordinates workers:

manager = Agent(instructions="Coordinate the team")
worker1 = Agent(instructions="Handle research")
worker2 = Agent(instructions="Handle writing")

team = AgentTeam(
    agents=[manager, worker1, worker2],
    process="hierarchical"
)

Best Practices

Clear Roles

Each agent should have one specific job

Simple Instructions

Keep agent instructions focused and clear

Start with 2-3 Agents

Don’t overcomplicate - add more only when needed

Test Each Agent First

Make sure each agent works alone before combining

Next: Building Conversational Agents

Learn how to create agents that remember conversations.

​What We’ll Build

Researcher

Writer

Editor

​Step-by-Step Guide

​Complete Example

​How It Works

​Different Team Patterns

​Parallel Team

​Hierarchical Team

​Best Practices

Next: Building Conversational Agents

What We’ll Build

Step-by-Step Guide

Complete Example

How It Works

Different Team Patterns

Parallel Team

Hierarchical Team

Best Practices