​
Integrating with Existing AI Frameworks

Kubiya seamlessly integrates with popular AI frameworks, allowing you to enhance your existing agent systems with production-grade orchestration, execution, and scaling capabilities. Instead of replacing your current setup, Kubiya acts as an execution layer that brings deterministic workflows and enterprise features to your AI agents.

​
Why Integrate with Kubiya?

​
Supported Frameworks

LangGraph

Multi-agent workflows with state management

  • Graph-based agent orchestration
  • Stateful conversations
  • Human-in-the-loop patterns
  • Complex routing logic

LangChain

Chain-based agent pipelines

  • Sequential processing chains
  • RAG and retrieval patterns
  • Tool integration
  • Memory management

CrewAI

Multi-agent collaboration

  • Role-based agent teams
  • Collaborative task solving
  • Agent coordination
  • Hierarchical workflows

Custom Frameworks

Your own AI framework

  • Any Python-based framework
  • Custom agent implementations
  • Proprietary AI systems
  • Legacy integrations

​
Integration Patterns

There are three main patterns for integrating existing AI frameworks with Kubiya:

​
1. Workflow Wrapping Pattern

Wrap your existing agent code within Kubiya workflows for execution orchestration:

from kubiya_workflow_sdk import workflow, step

@workflow
def langgraph_agent_workflow():
    # Your existing LangGraph code
    result = step("langgraph_execution").python(
        code="""
import langgraph
from your_existing_code import create_agent_graph

# Your existing LangGraph setup
graph = create_agent_graph()
result = graph.invoke({"messages": [{"role": "user", "content": "${INPUT}"}]})
return result
        """,
        requirements=["langgraph", "langchain"]
    )
    
    return result

​
2. Agent Server Integration

Deploy your framework as a custom agent server provider:

from kubiya_workflow_sdk.providers import BaseProvider

class LangGraphProvider(BaseProvider):
    def __init__(self):
        self.graph = self.create_langgraph_agent()
    
    async def compose(self, task: str, mode: str = "act"):
        # Your existing LangGraph logic
        result = await self.graph.ainvoke({
            "messages": [{"role": "user", "content": task}]
        })
        return result
    
    def create_langgraph_agent(self):
        # Your existing agent creation code
        from langgraph import StateGraph
        # ... your graph setup
        return compiled_graph

​
3. Hybrid Architecture

Combine Kubiya’s orchestration with your framework’s agent logic:

from kubiya_workflow_sdk import workflow, step

@workflow
def hybrid_ai_pipeline():
    # Use Kubiya's ADK for planning
    plan = step("planning").adk_compose(
        task="Create a plan for: ${INPUT}",
        mode="plan"
    )
    
    # Use your existing framework for execution
    execution = step("crewai_execution").python(
        code="""
from your_existing_code import execute_plan_with_crewai
result = execute_plan_with_crewai("${plan.strategy}")
return result
        """,
        requirements=["crewai"]
    ).depends("planning")
    
    # Use Kubiya for monitoring and reporting
    report = step("reporting").adk_compose(
        task="Generate report for execution: ${execution}",
        mode="act"
    ).depends("crewai_execution")

​
Key Integration Benefits

Kubernetes-Native Scaling

Your existing AI agents automatically gain:

  • Horizontal pod autoscaling
  • Resource management and limits
  • Load balancing across replicas
  • Multi-zone deployments
# Your LangGraph agents now scale automatically
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: langgraph-agent-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: langgraph-agent
  minReplicas: 2
  maxReplicas: 20
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 70

​
Migration Strategies

​
Gradual Migration

1

Start with Workflow Wrapping

Begin by wrapping your existing agents in Kubiya workflows without changing core logic

2

Add Orchestration Features

Gradually add streaming, monitoring, and error handling capabilities

3

Optimize for Production

Refactor critical paths to use native Kubiya features for better performance

4

Full Integration

Move to custom providers or hybrid architectures for maximum benefit

​
Parallel Development

1

Deploy Side by Side

Run your existing system alongside Kubiya-integrated versions

2

A/B Testing

Compare performance and reliability between implementations

3

Gradual Traffic Shift

Slowly move traffic from legacy to Kubiya-integrated agents

4

Sunset Legacy

Retire old implementation once migration is complete

​
Common Integration Scenarios

​
Getting Started

Choose your framework and follow the specific integration guide:

LangGraph Integration

Connect stateful multi-agent graphs

LangChain Integration

Enhance chain-based workflows

CrewAI Integration

Scale multi-agent teams

​
Support & Community

GitHub Examples

Browse real-world integration examples and templates