Skip to main content
Version: Next

Multi-Framework Runtime

Agent Kernel supports running agents from multiple frameworks within a single runtime environment. This allows you to leverage the unique strengths of different frameworks in one application while maintaining a unified interface for interaction.

Overview​

The multi-framework capability enables you to:

  • Mix and Match: Use OpenAI Agents SDK for production workflows alongside CrewAI for role-based tasks
  • Unified CLI: Interact with all agents through a single command-line interface
  • Session Flexibility: Switch between agents from different frameworks within the same session
  • Specialized Agents: Deploy framework-specific agents for their optimal use cases
info

While agents from different frameworks can run in the same runtime and be accessed in the same session, they cannot directly communicate with each other. Each framework maintains its own agent lifecycle and communication protocols.

Supported Framework Combinations​

You can combine any of the supported frameworks:

Basic Example​

Here's a complete example showing OpenAI Agents SDK and CrewAI agents running together:

from agents import Agent as OpenAIAgent
from crewai import Agent as CrewAIAgent

from agentkernel.cli import CLI
from agentkernel.crewai import CrewAIModule
from agentkernel.openai import OpenAIModule

# OpenAI Agents SDK agents
general_agent = OpenAIAgent(
    name="general",
    handoff_description="Agent for general questions",
    instructions="You provide assistance with general queries. Give short and clear answers",
)

math_agent = OpenAIAgent(
    name="math",
    handoff_description="Specialist agent for math questions", 
    instructions="You provide help with math problems. Explain your reasoning at each step.",
)

# CrewAI agent
history_agent = CrewAIAgent(
    role="history",
    goal="Specialist agent for history questions",
    backstory="You provide assistance with history queries with important details and context.",
    verbose=False,
)

# Register agents with their respective modules
OpenAIModule([general_agent, math_agent])
CrewAIModule([history_agent])

if __name__ == "__main__":
    CLI.main()

Installation​

To use multiple frameworks, install Agent Kernel with the required framework extensions:

pip install agentkernel[cli,crewai,openai]

For all supported frameworks:

pip install agentkernel[cli,crewai,openai,langgraph,google-adk]

Project Structure​

A typical multi-framework project structure:

my-multi-agent-app/
├── pyproject.toml          # Dependencies for all frameworks
├── demo.py                 # Main application entry point
├── agents/
│   ├── openai_agents.py    # OpenAI Agents SDK definitions
│   ├── crewai_agents.py    # CrewAI agent definitions
│   └── langgraph_agents.py # LangGraph workflow definitions
└── README.md

Usage Patterns​

Specialized Agent Roles​

Each framework excels in different scenarios:

# OpenAI Agents SDK - Great for production workflows
customer_service = OpenAIAgent(
    name="customer_service",
    instructions="Handle customer inquiries professionally with access to tools.",
)

# CrewAI - Excellent for role-based collaboration  
research_agent = CrewAIAgent(
    role="researcher",
    goal="Conduct thorough research on topics",
    backstory="Expert researcher with attention to detail",
)

# LangGraph - Perfect for complex multi-step workflows
workflow_agent = create_langgraph_workflow()

OpenAIModule([customer_service])
CrewAIModule([research_agent]) 
LangGraphModule([workflow_agent])

Configuration​

Dependencies​

Add all required frameworks to your pyproject.toml:

[project]
dependencies = [
    "agentkernel[cli,crewai,openai,langgraph]>=0.1.0",
]

Environment Variables​

Each framework may require specific environment variables:

# For OpenAI Agents SDK
OPENAI_API_KEY=your_openai_key

# For Google ADK  
GOOGLE_APPLICATION_CREDENTIALS=path/to/credentials.json

# Framework-specific configurations
CREWAI_CONFIG_PATH=./config/crewai.yaml

Best Practices​

Framework Selection​

Choose frameworks based on their strengths:

Use CaseRecommended FrameworkReason
Production APIsOpenAI Agents SDKRobust, well-tested, OpenAI optimized
Role-based teamsCrewAINatural role definitions, collaboration focus
Complex workflowsLangGraphAdvanced state management, conditional logic
Google ecosystemGoogle ADKNative Google service integration

Agent Organization​

# Group related agents by framework
class AgentRegistry:
    def __init__(self):
        # OpenAI agents for core business logic
        self.openai_agents = [
            create_customer_service_agent(),
            create_sales_agent(),
        ]
        
        # CrewAI agents for specialized roles
        self.crewai_agents = [
            create_research_agent(),
            create_writer_agent(),
        ]
        
        # LangGraph for complex workflows
        self.langgraph_workflows = [
            create_approval_workflow(),
        ]
    
    def register_all(self):
        OpenAIModule(self.openai_agents)
        CrewAIModule(self.crewai_agents)
        LangGraphModule(self.langgraph_workflows)

Error Handling​

try:
    OpenAIModule(openai_agents)
    print("✓ OpenAI agents registered")
except Exception as e:
    print(f"✗ Failed to register OpenAI agents: {e}")

try:
    CrewAIModule(crewai_agents)
    print("✓ CrewAI agents registered")
except Exception as e:
    print(f"✗ Failed to register CrewAI agents: {e}")

Limitations​

  • No Inter-Framework Communication: Agents from different frameworks cannot directly communicate
  • Separate Memory: Each framework maintains its own memory and session state
  • Framework Dependencies: All required frameworks must be installed and properly configured
  • Resource Overhead: Running multiple frameworks increases memory and processing requirements

Example Projects​

Check out the complete working example:

This example demonstrates OpenAI Agents SDK and CrewAI agents running together with CLI interaction.

Next Steps​