Skip to content

AutoSwarmBuilder: 3-Step Quickstart Guide

The AutoSwarmBuilder automatically designs and creates specialized multi-agent teams based on your task description. Simply describe what you need, and it will generate agents with distinct roles, expertise, personalities, and comprehensive system prompts - then orchestrate them using the most appropriate swarm architecture.

Overview

Feature Description
Automatic Agent Generation Creates agents with roles, personalities, and expertise based on task
Intelligent Architecture Selection Chooses optimal swarm type (Sequential, Concurrent, Hierarchical, etc.)
Comprehensive System Prompts Generates detailed prompts with decision-making frameworks
Flexible Execution Returns agents, swarm router config, or agent objects 
Your Task Description
        │
        ▼
   AutoSwarmBuilder
   (Boss System Prompt)
        │
        ▼
┌───────────────────────┐
│ Auto-Generated Team   │
│  - Agent Roles        │
│  - Personalities      │
│  - System Prompts     │
│  - Architecture Type  │
└───────────────────────┘
        │
        ▼
    Ready to Run

Step 1: Install and Import

pip install swarms

from swarms.structs.auto_swarm_builder import AutoSwarmBuilder

Step 2: Create AutoSwarmBuilder

# Initialize the builder
swarm_builder = AutoSwarmBuilder(
    name="Marketing-Team-Builder",
    description="Builds marketing teams automatically",
    model_name="gpt-4o",  # Boss agent model
    max_loops=1,
    execution_type="return-agents",  # or "return-swarm-router-config", "return-agents-objects"
    verbose=True
)

Step 3: Generate and Run

# Describe what you need
task = "Create a marketing team with 4 agents: market researcher, content strategist, copywriter, and social media specialist. They should collaborate on launching a new AI product."

# Auto-generate the team
result = swarm_builder.run(task=task)

# The builder creates:
# - 4 agents with specialized roles
# - Comprehensive system prompts for each
# - Appropriate swarm architecture
# - Ready-to-use configuration

print(result)

Complete Example

from swarms.structs.auto_swarm_builder import AutoSwarmBuilder
import json

# Create builder
swarm = AutoSwarmBuilder(
    name="Product-Development-Team",
    description="Auto-generates product development teams",
    model_name="gpt-4o",
    max_loops=1,
    execution_type="return-agents",
    verbose=True
)

# Define your need
task = """
Create a product development team with 5 specialized agents:
1. Product Manager - oversees strategy and roadmap
2. UX Designer - focuses on user experience
3. Backend Engineer - handles server-side development
4. Frontend Engineer - builds user interfaces
5. QA Engineer - ensures quality and testing

The team should work together to plan and build a new mobile app feature.
"""

# Generate the team
team_config = swarm.run(task=task)

# View the generated team
print(json.dumps(team_config, indent=2))

Execution Types

Type Returns Use Case
"return-agents" List of agent dictionaries Inspect and customize agents
"return-swarm-router-config" Complete SwarmRouter configuration Ready-to-use swarm
"return-agents-objects" List of Agent objects Direct execution

Example: Get Ready-to-Run Swarm

swarm = AutoSwarmBuilder(
    name="Research-Team",
    model_name="gpt-4o",
    execution_type="return-swarm-router-config",  # Get complete swarm
)

result = swarm.run(
    "Create a research team with data analyst, statistician, and research coordinator"
)

# Result is a complete SwarmRouter configuration
# Ready to use immediately

Configuration Options

Parameter Default Description
name Required Name of the builder
description Required Purpose of the builder
model_name "gpt-4o" Model for the boss agent that designs teams
max_loops 1 Loops for agent generation
execution_type "return-agents" What to return
verbose False Enable detailed logging

Use Cases

Scenario Team Description
Content Creation "Writers, editors, SEO specialists for blog content"
Software Development "Full-stack developers, QA engineers, DevOps for microservices"
Financial Analysis "Financial analysts, risk managers, compliance officers for investment portfolio"
Customer Support "Support agents, escalation specialists, quality reviewers for customer service"
Research "Researchers, data scientists, literature reviewers for scientific study"

Example: Financial Analysis Team

swarm = AutoSwarmBuilder(
    name="Financial-Team-Builder",
    model_name="gpt-4o",
    execution_type="return-agents",
)

team = swarm.run(
    """
    Create a financial analysis team with:
    - Equity Analyst: Analyzes stocks and market trends
    - Fixed Income Analyst: Evaluates bonds and debt instruments
    - Risk Manager: Assesses portfolio risk
    - Quantitative Analyst: Builds financial models

    Team should collaborate on portfolio management and investment recommendations.
    """
)

print(f"Generated {len(team)} specialized financial agents")

How It Works

  1. Task Analysis: Boss agent analyzes your requirements
  2. Agent Design: Creates agents with:
  3. Unique roles and purposes
  4. Distinct personalities
  5. Comprehensive system prompts
  6. Specific capabilities and limitations
  7. Architecture Selection: Chooses optimal swarm type
  8. Configuration Generation: Outputs ready-to-use configuration
  9. Return: Provides agents in requested format

Advanced Features

Custom Boss System Prompt

The boss agent uses a sophisticated system prompt that considers: - Task decomposition and analysis - Agent design excellence with personalities - Communication protocols and collaboration strategies - Multi-agent architecture selection - Quality assurance and governance

Supported Swarm Architectures

The boss can select from: - AgentRearrange - MixtureOfAgents - SpreadSheetSwarm - SequentialWorkflow - ConcurrentWorkflow - GroupChat - MultiAgentRouter - HierarchicalSwarm - MajorityVoting - And more...

Best Practices

  • Be Specific: Provide clear, detailed task descriptions
  • Define Roles: Specify the types of agents you need
  • State Objectives: Explain what the team should accomplish
  • Use Powerful Models: Use gpt-4o or claude-sonnet for best results
  • Review Output: Always review and potentially customize generated agents

Next Steps