Multi-Agent Team

A team of specialized AI agents that automatically coordinate to handle complex requests. Demonstrates MUXI's automatic workflow decomposition and agent orchestration.

Difficulty: Advanced Time to setup: 10 minutes

What It Does

  • Multiple specialized agents (researcher, analyst, writer)
  • Automatic task decomposition
  • Parallel execution when possible
  • Agent-to-agent collaboration (A2A)
  • Produces comprehensive outputs

The Team

🔍 Researcher

  • Gathers information from web and documents
  • Identifies relevant sources
  • Summarizes findings

📊 Analyst

  • Processes data and findings
  • Identifies patterns and trends
  • Creates visualizations

✍️ Writer

  • Creates reports and summaries
  • Ensures consistent tone
  • Formats output professionally

🎯 Overlord

  • Coordinates the team
  • Assigns tasks to agents
  • Synthesizes final output

Features

  • ✅ Automatic Orchestration: No manual coordination needed
  • ✅ Parallel Execution: Independent tasks run simultaneously
  • ✅ Agent Specialization: Each agent has distinct expertise
  • ✅ Workflow Intelligence: Determines optimal execution order
  • ✅ Error Recovery: Failed tasks retry automatically

Prerequisites

  • MUXI Server running
  • OpenAI API key
  • Brave Search API key (for researcher)

Setup

# 1. Copy example
cp -r examples/05-multi-agent-team my-team
cd my-team

# 2. Install MCP servers
npm install -g @modelcontextprotocol/server-brave-search
npm install -g @modelcontextprotocol/server-filesystem

# 3. Set secrets
muxi secrets setup

# 4. Run
muxi dev

Test It

Simple Request (Single Agent)

curl -X POST http://localhost:8001/v1/chat \
  -d '{"message": "What is machine learning?"}'

# Overlord routes to researcher (simple factual question)

Complex Request (Multi-Agent Workflow)

curl -X POST http://localhost:8001/v1/chat \
  -d '{"message": "Research the top 3 cloud providers, analyze their pricing and features, and create a comparison report"}'

# Overlord creates workflow:
# 1. Researcher: Search for AWS, Azure, GCP (parallel)
# 2. Analyst: Compare pricing and features
# 3. Writer: Create formatted report

How It Works

graph TD
    U[User Request] --> O[Overlord]
    O --> C{Complexity<br/>Analysis}
    C -->|Simple<br/>Score: 4| R[Researcher Only]
    C -->|Complex<br/>Score: 9| W[Create Workflow]
    W --> T1[Task 1: Research AWS]
    W --> T2[Task 2: Research Azure]
    W --> T3[Task 3: Research GCP]
    T1 --> R1[Researcher]
    T2 --> R1
    T3 --> R1
    R1 --> T4[Task 4: Analyze]
    T4 --> A[Analyst]
    A --> T5[Task 5: Write Report]
    T5 --> WR[Writer]
    WR --> O
    R --> O
    O --> Response[Final Response]

Example Workflow

User Request:

"Research AI trends in 2025, analyze market growth, create an executive summary"

Overlord Analysis:

  • Complexity score: 9/10
  • Triggers workflow mode
  • Creates plan:

Workflow Execution:

[WORKFLOW] Created workflow for request req_abc123

Task 1: Research AI trends 2025
  Agent: researcher
  Tools: brave-search
  Status: ✓ Completed (3.2s)

Task 2: Research market data
  Agent: researcher
  Tools: brave-search
  Status: ✓ Completed (2.8s)

[Tasks 1-2 ran in parallel]

Task 3: Analyze trends and market data
  Agent: analyst
  Input: Results from Tasks 1-2
  Tools: data-analysis
  Status: ✓ Completed (4.1s)

Task 4: Create executive summary
  Agent: writer
  Input: Analysis from Task 3
  Tools: document-creation
  Status: ✓ Completed (5.3s)

[WORKFLOW] Completed in 15.4s total

Final Output:

# AI Trends 2025: Executive Summary

## Key Findings

Based on analysis of 12 recent reports and market data...

[Comprehensive 3-page executive summary with data, trends, and recommendations]

Configuration Highlights

Multiple Agents

Each agent is defined in agents/*.afs:

# agents/researcher.afs
schema: "1.0.0"
id: researcher
name: Researcher
description: Expert at gathering information

system_message: Expert at gathering information...

llm_models:
  - text: "openai/gpt-4-turbo"  # Faster for research
# agents/analyst.afs
schema: "1.0.0"
id: analyst
name: Analyst
description: Data analyst who identifies patterns

system_message: Data analyst who identifies patterns...

llm_models:
  - text: "openai/gpt-4o"  # Best reasoning
# agents/writer.afs
schema: "1.0.0"
id: writer
name: Writer
description: Professional writer

system_message: Professional writer who creates reports...

llm_models:
  - text: "openai/gpt-4"  # Best writing quality

Workflow Settings

overlord:
  workflow:
    auto_decomposition: true
    complexity_threshold: 7.0
    max_parallel_tasks: 5
    timeouts:
      task_timeout: 120
    retry:
      max_attempts: 3

Agent Selection

Overlord automatically chooses based on task:

overlord:
  agent_selection:
    - match: "research|find|search"
      agent: researcher

    - match: "analyze|compare|trends"
      agent: analyst

    - match: "write|create|summary|report"
      agent: writer

Use Cases

Business Intelligence

"Research our competitors' Q4 earnings, analyze trends, create investor report"

Workflow:
- Researcher: Finds earnings reports
- Analyst: Analyzes financial trends
- Writer: Creates investor-ready report

Content Creation

"Research sustainable energy, analyze adoption rates, write blog post"

Workflow:
- Researcher: Gathers facts and stats
- Analyst: Identifies key insights
- Writer: Creates engaging blog post

Market Research

"Research electric vehicle market in Europe, analyze growth, create executive summary"

Workflow:
- Researcher: Market data and reports
- Analyst: Growth analysis and forecasts
- Writer: Executive summary for C-suite

Customization

Add More Agents

Create additional agent files in agents/:

# agents/editor.afs
schema: "1.0.0"
id: editor
name: Editor
description: Proofreads and improves written content

system_message: Proofreads and improves written content.
# agents/translator.afs
schema: "1.0.0"
id: translator
name: Translator
description: Translates content to multiple languages

system_message: Translates content to multiple languages.

Adjust Complexity Threshold

overlord:
  complexity_threshold: 6.0  # More workflows (more aggressive)
  # or
  complexity_threshold: 8.0  # Fewer workflows (more selective)

Force Specific Workflow

sops:
  - name: market-research
    trigger: "market research"
    steps:
      - agent: researcher
        task: "Gather market data"
      - agent: analyst
        task: "Analyze data"
      - agent: writer
        task: "Create report"

Monitoring Workflows

Enable workflow logging:

overlord:
  workflow:
    log_workflows: true

See execution:

muxi logs my-team --follow

# Output shows:
[WORKFLOW] Task 1 started (researcher)
[WORKFLOW] Task 2 started (researcher)
[WORKFLOW] Task 1 completed in 3.2s
[WORKFLOW] Task 2 completed in 2.8s
[WORKFLOW] Task 3 started (analyst)
...

Performance Tuning

Reduce Latency

# formation.afs
overlord:
  workflow:
    max_parallel_tasks: 10  # More parallelization

With faster model in agents/researcher.afs:

llm_models:
  - text: "openai/gpt-4-turbo"  # Faster model

Reduce Costs

Use cheaper model in agents/researcher.afs:

llm_models:
  - text: "openai/gpt-3.5-turbo"  # Cheaper model
# formation.afs
overlord:
  workflow:
    max_parallel_tasks: 2  # Less parallel API calls

Common Issues

"All requests go to one agent"

Complexity threshold too high:

overlord:
  complexity_threshold: 7.0  # Lower to 6 or 5

"Tasks timeout"

Increase timeout in formation.afs:

overlord:
  workflow:
    timeouts:
      task_timeout: 180  # 3 minutes

"No parallel execution"

Check tasks have no dependencies in formation.afs:

overlord:
  workflow:
    min_tasks_for_parallel: 2  # Enable for 2+ tasks

Next Steps

Home Registry Documentation Star on GitHub
Get priority support