> ## Documentation Index
> Fetch the complete documentation index at: https://docs.run-agent.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# LangGraph

> Deploy LangGraph agents with RunAgent

<Info>
  **Prerequisites**: Basic understanding of LangGraph and completed [Deploy Your First Agent](/tutorials/deploy-your-first-agent) tutorial
</Info>

## Overview

LangGraph is a powerful framework for building stateful, multi-step AI agents with complex workflows. RunAgent makes it easy to deploy LangGraph agents and access them from any programming language.

## Quick Start

### 1. Create a LangGraph Agent

```bash theme={null}
runagent init my-langgraph-agent --framework langgraph
cd my-langgraph-agent
```

### 2. Install Dependencies

```bash theme={null}
pip install -r requirements.txt
```

### 3. Configure Your Agent

The generated `runagent.config.json` will be pre-configured for LangGraph:

```json theme={null}
{
  "agent_name": "my-langgraph-agent",
  "description": "LangGraph agent with stateful workflows",
  "framework": "langgraph",
  "agent_architecture": {
    "entrypoints": [
      {
        "file": "agents.py",
        "module": "langgraph_agent",
        "tag": "main"
      }
    ]
  }
}
```

## Basic LangGraph Agent

Here's a simple LangGraph agent that demonstrates the core concepts:

```python agents.py theme={null}
from typing import Dict, Any, List
from langgraph import StateGraph, END
from langgraph.graph import StateGraph
from langchain_core.messages import HumanMessage, AIMessage
from langchain_openai import ChatOpenAI

# Define the state structure
class AgentState:
    messages: List[Any]
    user_input: str
    response: str
    step_count: int

# Initialize the LLM
llm = ChatOpenAI(model="gpt-4", temperature=0.7)

def process_input(state: AgentState) -> AgentState:
    """Process the user input and prepare for reasoning"""
    state.step_count = 0
    state.messages = [HumanMessage(content=state.user_input)]
    return state

def reason_step(state: AgentState) -> AgentState:
    """Perform reasoning step"""
    state.step_count += 1
    
    # Add reasoning message
    reasoning = f"Step {state.step_count}: Analyzing the user's request..."
    state.messages.append(AIMessage(content=reasoning))
    
    return state

def generate_response(state: AgentState) -> AgentState:
    """Generate the final response"""
    # Get response from LLM
    response = llm.invoke(state.messages)
    state.response = response.content
    state.messages.append(response)
    
    return state

def should_continue(state: AgentState) -> str:
    """Decide whether to continue reasoning or finish"""
    if state.step_count >= 3:  # Max 3 reasoning steps
        return "finish"
    elif "complex" in state.user_input.lower():
        return "reason"
    else:
        return "finish"

# Build the graph
workflow = StateGraph(AgentState)

# Add nodes
workflow.add_node("process_input", process_input)
workflow.add_node("reason", reason_step)
workflow.add_node("generate_response", generate_response)

# Add edges
workflow.add_edge("process_input", "reason")
workflow.add_conditional_edges(
    "reason",
    should_continue,
    {
        "reason": "reason",
        "finish": "generate_response"
    }
)
workflow.add_edge("generate_response", END)

# Compile the graph
app = workflow.compile()

def langgraph_agent(user_input: str) -> Dict[str, Any]:
    """Main entrypoint for the LangGraph agent"""
    try:
        # Initialize state
        initial_state = AgentState(
            messages=[],
            user_input=user_input,
            response="",
            step_count=0
        )
        
        # Run the workflow
        result = app.invoke(initial_state)
        
        return {
            "response": result.response,
            "step_count": result.step_count,
            "messages": [msg.content for msg in result.messages],
            "status": "success"
        }
        
    except Exception as e:
        return {
            "response": f"Error: {str(e)}",
            "step_count": 0,
            "messages": [],
            "status": "error"
        }
```

## Advanced LangGraph Patterns

### 1. Multi-Agent Workflows

```python multi_agent.py theme={null}
from typing import Dict, Any, List
from langgraph import StateGraph, END
from langchain_core.messages import HumanMessage, AIMessage, SystemMessage

class MultiAgentState:
    messages: List[Any]
    user_input: str
    researcher_output: str
    writer_output: str
    final_response: str

def researcher_agent(state: MultiAgentState) -> MultiAgentState:
    """Research agent that gathers information"""
    research_prompt = f"""
    Research the following topic: {state.user_input}
    Provide key facts and insights.
    """
    
    # Simulate research (in real app, use web search, databases, etc.)
    state.researcher_output = f"Research findings for: {state.user_input}"
    return state

def writer_agent(state: MultiAgentState) -> MultiAgentState:
    """Writer agent that creates content"""
    writer_prompt = f"""
    Based on this research: {state.researcher_output}
    Create a comprehensive response to: {state.user_input}
    """
    
    # Simulate writing (in real app, use LLM)
    state.writer_output = f"Written content based on: {state.researcher_output}"
    return state

def coordinator_agent(state: MultiAgentState) -> MultiAgentState:
    """Coordinator that combines outputs"""
    state.final_response = f"""
    Research: {state.researcher_output}
    
    Content: {state.writer_output}
    
    Final Response: Based on the research and writing, here's the complete answer.
    """
    return state

# Build multi-agent workflow
multi_workflow = StateGraph(MultiAgentState)

multi_workflow.add_node("researcher", researcher_agent)
multi_workflow.add_node("writer", writer_agent)
multi_workflow.add_node("coordinator", coordinator_agent)

multi_workflow.add_edge("researcher", "writer")
multi_workflow.add_edge("writer", "coordinator")
multi_workflow.add_edge("coordinator", END)

multi_app = multi_workflow.compile()

def multi_agent_workflow(user_input: str) -> Dict[str, Any]:
    """Multi-agent workflow entrypoint"""
    initial_state = MultiAgentState(
        messages=[],
        user_input=user_input,
        researcher_output="",
        writer_output="",
        final_response=""
    )
    
    result = multi_app.invoke(initial_state)
    
    return {
        "response": result.final_response,
        "research": result.researcher_output,
        "writing": result.writer_output,
        "status": "success"
    }
```

### 2. Conditional Workflows

```python conditional_workflow.py theme={null}
from typing import Dict, Any, List
from langgraph import StateGraph, END

class ConditionalState:
    user_input: str
    intent: str
    response: str
    confidence: float

def classify_intent(state: ConditionalState) -> ConditionalState:
    """Classify user intent"""
    input_lower = state.user_input.lower()
    
    if any(word in input_lower for word in ["question", "what", "how", "why"]):
        state.intent = "question"
        state.confidence = 0.9
    elif any(word in input_lower for word in ["task", "do", "help", "assist"]):
        state.intent = "task"
        state.confidence = 0.8
    else:
        state.intent = "general"
        state.confidence = 0.5
    
    return state

def handle_question(state: ConditionalState) -> ConditionalState:
    """Handle question intent"""
    state.response = f"Answering your question: {state.user_input}"
    return state

def handle_task(state: ConditionalState) -> ConditionalState:
    """Handle task intent"""
    state.response = f"Helping you with this task: {state.user_input}"
    return state

def handle_general(state: ConditionalState) -> ConditionalState:
    """Handle general intent"""
    state.response = f"General response to: {state.user_input}"
    return state

def route_intent(state: ConditionalState) -> str:
    """Route based on intent"""
    return state.intent

# Build conditional workflow
conditional_workflow = StateGraph(ConditionalState)

conditional_workflow.add_node("classify", classify_intent)
conditional_workflow.add_node("question_handler", handle_question)
conditional_workflow.add_node("task_handler", handle_task)
conditional_workflow.add_node("general_handler", handle_general)

conditional_workflow.add_edge("classify", "route")
conditional_workflow.add_conditional_edges(
    "classify",
    route_intent,
    {
        "question": "question_handler",
        "task": "task_handler",
        "general": "general_handler"
    }
)
conditional_workflow.add_edge("question_handler", END)
conditional_workflow.add_edge("task_handler", END)
conditional_workflow.add_edge("general_handler", END)

conditional_app = conditional_workflow.compile()

def conditional_agent(user_input: str) -> Dict[str, Any]:
    """Conditional workflow entrypoint"""
    initial_state = ConditionalState(
        user_input=user_input,
        intent="",
        response="",
        confidence=0.0
    )
    
    result = conditional_app.invoke(initial_state)
    
    return {
        "response": result.response,
        "intent": result.intent,
        "confidence": result.confidence,
        "status": "success"
    }
```

## Streaming with LangGraph

LangGraph agents can also provide streaming responses:

```python streaming_agent.py theme={null}
from typing import Iterator, Dict, Any
from langgraph import StateGraph, END

def streaming_langgraph_agent(user_input: str) -> Iterator[str]:
    """Streaming LangGraph agent"""
    yield f"🤖 Starting LangGraph workflow for: {user_input}\n\n"
    
    # Simulate workflow steps
    steps = [
        "📝 Processing input...",
        "🧠 Analyzing context...",
        "🔍 Gathering information...",
        "💭 Reasoning through solution...",
        "✍️ Generating response...",
        "✅ Finalizing output..."
    ]
    
    for i, step in enumerate(steps):
        yield f"Step {i+1}: {step}\n"
        # Simulate processing time
        import time
        time.sleep(0.5)
    
    yield f"\n🎉 Workflow complete! Response: {user_input} processed successfully."
```

## Configuration for Multiple Entrypoints

Update your `runagent.config.json` to include multiple LangGraph workflows:

```json theme={null}
{
  "agent_name": "advanced-langgraph-agent",
  "description": "Advanced LangGraph agent with multiple workflows",
  "framework": "langgraph",
  "agent_architecture": {
    "entrypoints": [
      {
        "file": "agents.py",
        "module": "langgraph_agent",
        "tag": "basic"
      },
      {
        "file": "multi_agent.py",
        "module": "multi_agent_workflow",
        "tag": "multi_agent"
      },
      {
        "file": "conditional_workflow.py",
        "module": "conditional_agent",
        "tag": "conditional"
      },
      {
        "file": "streaming_agent.py",
        "module": "streaming_langgraph_agent",
        "tag": "streaming"
      }
    ]
  }
}
```

## Testing Your LangGraph Agent

### Python Client

```python test_langgraph.py theme={null}
from runagent import RunAgentClient

# Connect to your LangGraph agent
client = RunAgentClient(
    agent_id="your_agent_id_here",
    entrypoint_tag="basic",
    local=True
)

# Test basic workflow
result = client.run(user_input="What is the capital of France?")
print(f"Response: {result['response']}")
print(f"Steps: {result['step_count']}")

# Test multi-agent workflow
multi_client = RunAgentClient(
    agent_id="your_agent_id_here",
    entrypoint_tag="multi_agent",
    local=True
)

multi_result = multi_client.run(user_input="Explain quantum computing")
print(f"Multi-agent response: {multi_result['response']}")

# Test streaming
stream_client = RunAgentClient(
    agent_id="your_agent_id_here",
    entrypoint_tag="streaming",
    local=True
)

print("Streaming response:")
for chunk in stream_client.run(user_input="Process this data"):
    print(chunk, end="", flush=True)
```

### JavaScript Client

```javascript test_langgraph.js theme={null}
import { RunAgentClient } from 'runagent';

const client = new RunAgentClient({
    agentId: 'your_agent_id_here',
    entrypointTag: 'basic',
    local: true
});

await client.initialize();

const result = await client.run({
    user_input: 'What is the capital of France?'
});

console.log('Response:', result.response);
console.log('Steps:', result.step_count);
```

## Best Practices

### 1. **State Management**

* Keep state objects simple and focused
* Use clear naming conventions
* Avoid deep nesting in state

### 2. **Error Handling**

* Wrap workflow execution in try-catch blocks
* Provide meaningful error messages
* Log errors for debugging

### 3. **Performance Optimization**

* Use conditional edges to avoid unnecessary steps
* Implement early termination when possible
* Cache expensive operations

### 4. **Testing**

* Test each node independently
* Test the complete workflow
* Use mock data for testing

## Common Patterns

<AccordionGroup>
  <Accordion title="Research and Writing Workflow">
    Use LangGraph to create agents that research topics and then write about them.
  </Accordion>

  <Accordion title="Multi-Step Problem Solving">
    Break complex problems into smaller steps with conditional logic.
  </Accordion>

  <Accordion title="Agent Collaboration">
    Create multiple specialized agents that work together.
  </Accordion>

  <Accordion title="Human-in-the-Loop">
    Add human approval steps for critical decisions.
  </Accordion>
</AccordionGroup>

## Troubleshooting

### Common Issues

1. **State Serialization Errors**
   * Ensure all state fields are serializable
   * Use simple data types when possible

2. **Graph Compilation Errors**
   * Check that all nodes are properly defined
   * Verify edge connections are correct

3. **Memory Issues**
   * Limit the number of messages in state
   * Implement state cleanup for long conversations

### Debug Tips

```python theme={null}
# Add debugging to your workflows
def debug_node(state: AgentState) -> AgentState:
    print(f"Debug: Current state = {state}")
    return state

# Add debug node to your workflow
workflow.add_node("debug", debug_node)
```

## Next Steps

<CardGroup cols={2}>
  <Card title="Advanced Patterns" icon="cog" href="/how-to/advanced-tasks">
    Learn advanced LangGraph patterns and techniques
  </Card>

  <Card title="Production Deployment" icon="cloud" href="/runagent-cloud/cloud-deployment">
    Deploy your LangGraph agent to production
  </Card>

  <Card title="Multi-Language Access" icon="globe" href="/tutorials/multi-language-wrapper">
    Access your LangGraph agent from different languages
  </Card>

  <Card title="Performance Tuning" icon="gauge" href="/explanation/production-considerations">
    Optimize your LangGraph workflows for production
  </Card>
</CardGroup>

<Note>
  **🎉 Great job!** You've learned how to deploy LangGraph agents with RunAgent. LangGraph's powerful workflow capabilities combined with RunAgent's multi-language access make for a powerful combination!
</Note>

<Card title="Still have a question?" icon="circle-question" href="/components/columns">
  * Join our [Discord Community](https://discord.gg/Q9P9AdHVHz)
  * Email us: [hi@run-agent.ai](mailto:hi@run-agent.ai)
  * Follow us on [X](https://x.com/run_agent)
  * New here? [Sign up](https://run-agent.ai/dashboard)
</Card>