Module Agents Examples¶

This directory contains Dana module agent examples that work alongside the A2A agents in the parent directory.

Files¶

weather_expert.na - Simple weather analysis module agent
travel_planner.na - Simple travel planning module agent
demo_main.na - Complete demo showing both A2A and module agents working together

Quick Start¶

Start the A2A agents (run in background):

# From the parent directory - all in one terminal
cd examples/dana/08_a2a_multi_agents/
python start_weather_agent.py --port 5001 &
python start_planner_agent.py --port 5002 &
python start_ticket_agent.py --port 5003 &

# Wait for agents to start
sleep 3

Run the complete demo (from this directory):

# Method 1: Navigate to this directory (simple)
# After starting A2A agents, we are in examples/dana/08_a2a_multi_agents/
cd module_agents
uv run python -m opendxa.dana.exec.dana demo_main.na

# Method 2: Use DANA_PATH from anywhere
export DANA_PATH=/path/to/opendxa/examples/dana/08_a2a_multi_agents/module_agents
uv run python -m opendxa.dana.exec.dana examples/dana/08_a2a_multi_agents/module_agents/demo_main.na

Important Notes¶

Import Requirements¶

Dana module imports only work when running from the directory containing the .na files
The import weather_expert and import travel_planner statements in demo_main.na require running from this directory

Alternative: Using DANA_PATH¶

You can set the DANA_PATH environment variable to enable imports from any directory:

# From OpenDXA root directory
export DANA_PATH=$(pwd)/examples/dana/08_a2a_multi_agents/module_agents
uv run python -m opendxa.dana.exec.dana examples/dana/08_a2a_multi_agents/module_agents/demo_main.na

# Or inline
DANA_PATH=$(pwd)/examples/dana/08_a2a_multi_agents/module_agents uv run python -m opendxa.dana.exec.dana examples/dana/08_a2a_multi_agents/module_agents/demo_main.na

MCP Resource Adaptation¶

The examples use websearch.openai_websearch() but your MCP server may have different methods

Check your MCP server documentation and adapt the resource calls accordingly:

# Adapt to your MCP server's actual methods
# result = websearch.query(search_term)
# result = websearch.web_search(query)  
# result = websearch.search_web(text)

Module Agent Pattern¶

Each module agent follows this simple pattern:

# Required system variables
system:agent_name = "Agent Name"
system:agent_description = "What the agent does"

# Required solve function
def solve(task: str) -> str:
    return process_task(task)

Integration with A2A Agents¶

The demo shows how module agents and A2A agents work seamlessly together: - Mixed agent pools containing both types - Same solve() interface for both - Compatible with reason function - Support for resource filtering

Troubleshooting¶

Import Errors¶

Problem: ImportError: No module named 'weather_expert'
Solution 1: Make sure you're running from this module_agents/ directory
Solution 2: Set DANA_PATH environment variable: export DANA_PATH=$(pwd)/examples/dana/08_a2a_multi_agents/module_agents

MCP Resource Errors¶

Problem: AttributeError: 'MCPResource' object has no attribute 'search'
Solution: Update the resource method calls to match your MCP server's API

A2A Connection Errors¶

Problem: A2A agents not responding
Solution: Check agents are running with ps aux | grep python and restart if needed

This demonstrates Dana's unified multi-agent architecture!

Agent Pool Management¶

Managing multiple agents efficiently requires careful consideration of:

Pool Initialization: Setting up agent pools with proper resource allocation
Load Balancing: Distributing tasks across available agents
Health Monitoring: Checking agent status and handling failures
Dynamic Scaling: Adding or removing agents based on demand

Example agent pool configuration:

struct AgentPool:
    agents: list
    max_concurrent_tasks: int
    health_check_interval: int

def create_agent_pool(config: dict) -> AgentPool:
    pool = AgentPool(
        agents=[],
        max_concurrent_tasks=config.get("max_concurrent", 10),
        health_check_interval=config.get("health_interval", 30)
    )

    # Initialize agents based on configuration
    for agent_config in config.get("agents", []):
        agent = create_agent(agent_config)
        pool.agents.append(agent)

    return pool

Reason Function Integration¶

The reason() function provides seamless integration with multi-agent systems:

Agent Selection: Automatically choose the best agent for a task
Context Sharing: Share context between agents efficiently
Result Aggregation: Combine results from multiple agents
Error Handling: Graceful fallback when agents fail

Example reason function usage with agents:

def solve_complex_task(task: str, agents: list) -> str:
    # Use reason function to coordinate agents
    result = reason(
        f"Solve this task using the available agents: {task}",
        context=[task, agents],
        agents=agents,
        strategy="best_match"
    )

    return result