SwarmRearrange Example

The SwarmRearrange is a multi-agent architecture that orchestrates multiple swarms (not individual agents) in a defined flow pattern. It allows you to chain swarms together sequentially or run them in parallel, creating complex multi-swarm workflows.

How It Works

1. Swarm Orchestration: Manages multiple swarms (like HierarchicalSwarm, MixtureOfAgents, etc.)
2. Flow Patterns: Uses arrow (->) for sequential and comma (,) for parallel execution
3. Flexible Execution: Supports complex workflows combining sequential and parallel patterns
4. Human-in-the-Loop: Optional human intervention points in the workflow

This architecture is perfect for: - Complex multi-stage workflows - Combining different swarm architectures - Orchestrating hierarchical processes - Multi-phase analysis pipelines

Installation

Install the swarms package using pip:

pip install -U swarms

Basic Setup

1. First, set up your environment variables:

WORKSPACE_DIR="agent_workspace"
OPENAI_API_KEY="your-api-key"

Step-by-Step Example

Step 1: Import Required Modules

from swarms import Agent, SwarmRearrange, SequentialWorkflow, MixtureOfAgents

Step 2: Create Individual Swarms

# Swarm 1: Research Swarm (using SequentialWorkflow)
researcher = Agent(
    agent_name="Researcher",
    system_prompt="You are a research specialist. Gather and analyze information.",
    model_name="gpt-4o-mini",
)

writer = Agent(
    agent_name="Writer",
    system_prompt="You are a technical writer. Write clear, structured documents.",
    model_name="gpt-4o-mini",
)

research_swarm = SequentialWorkflow(
    name="Research-Swarm",
    agents=[researcher, writer],
    max_loops=1,
)

# Swarm 2: Analysis Swarm (using MixtureOfAgents)
analyst1 = Agent(
    agent_name="Analyst-1",
    system_prompt="You are a data analyst. Analyze data and provide insights.",
    model_name="gpt-4o-mini",
)

analyst2 = Agent(
    agent_name="Analyst-2",
    system_prompt="You are a strategic analyst. Provide strategic recommendations.",
    model_name="gpt-4o-mini",
)

analysis_swarm = MixtureOfAgents(
    agents=[analyst1, analyst2],
    layers=1,
)

# Swarm 3: Review Swarm (single agent as a simple swarm)
reviewer = Agent(
    agent_name="Reviewer",
    system_prompt="You are a quality reviewer. Review and provide feedback.",
    model_name="gpt-4o-mini",
)

Step 3: Create SwarmRearrange with Flow Pattern

# Define flow: Research -> Analysis (parallel) -> Review
# This means: Research runs first, then Analysis swarms run in parallel, then Review
swarm_rearrange = SwarmRearrange(
    name="Document-Pipeline",
    description="A pipeline for document creation and review",
    swarms=[research_swarm, analysis_swarm, reviewer],
    flow="Research-Swarm -> Analysis-Swarm, Reviewer",
    max_loops=1,
    verbose=True,
)

Step 4: Run the SwarmRearrange

task = "Create a comprehensive report on AI trends in 2024"

result = swarm_rearrange.run(task=task)

print(result)

Flow Pattern Examples

Sequential Flow

# All swarms run one after another
flow = "Research-Swarm -> Analysis-Swarm -> Reviewer"

Parallel Flow

# Multiple swarms run at the same time
flow = "Research-Swarm, Analysis-Swarm -> Reviewer"

Complex Flow

# Mix of sequential and parallel
flow = "Research-Swarm -> Analysis-Swarm, Reviewer -> Final-Reviewer"

Advanced: Human-in-the-Loop

You can add human intervention points in your workflow:

def human_reviewer(input_text):
    """Custom human-in-the-loop function"""
    print(f"Review this output: {input_text}")
    feedback = input("Enter your feedback: ")
    return feedback

swarm_rearrange = SwarmRearrange(
    name="Human-Reviewed-Pipeline",
    swarms=[research_swarm, analysis_swarm, reviewer],
    flow="Research-Swarm -> H -> Analysis-Swarm -> Reviewer",
    human_in_the_loop=True,
    custom_human_in_the_loop=human_reviewer,
    max_loops=1,
)

Understanding the Flow

• -> means sequential: The swarm on the left completes before the one on the right starts
• , means parallel: All swarms separated by commas run simultaneously
• H means human-in-the-loop: Pauses for human input
• You can combine these patterns for complex workflows

Example: Multi-Phase Analysis

# Phase 1: Research (sequential)
# Phase 2: Analysis (parallel - multiple analysts)
# Phase 3: Synthesis (sequential)
# Phase 4: Review (parallel - multiple reviewers)

flow = "Research-Swarm -> Analyst-1, Analyst-2, Analyst-3 -> Synthesis-Swarm -> Reviewer-1, Reviewer-2"

Support and Community

If you're facing issues or want to learn more, check out the following resources:

Platform	Link	Description
📚 Documentation	docs.swarms.world	Official documentation and guides
💬 Discord	Join Discord	Live chat and community support
🐦 Twitter	@swarms_corp	Latest news and announcements