LangGraph

LangGraph is a framework for building stateful, multi-step agent workflows as directed graphs where each node is an LLM call, tool invocation, or decision point. Built by the LangChain team, it handles the complex control flow that simple chains cannot express.

Key use cases include:

• Building multi-agent systems with coordinated handoffs between specialized agents
• Workflows with loops and cycles — agents that iterate until a condition is met
• Human-in-the-loop patterns where humans approve or redirect agent actions
• Long-running workflows with persistent state that survives restarts
• Complex branching logic based on LLM outputs or tool results
• Building custom ReAct agents, planners, and reflective agents

LangGraph is used by teams building production-grade agentic applications that need reliability, observability, and complex control flow beyond what simple prompt chains offer. It integrates natively with LangChain components and LangSmith tracing.

LangGraph Cloud provides deployment, scaling, and a built-in studio UI for debugging agent workflows visually.

1. Install LangGraph:

   pip install langgraph langchain-openai

2. Set your LLM provider key:

   export OPENAI_API_KEY='sk-...'

3. Build a simple agent graph:

   from langgraph.prebuilt import create_react_agent
   from langchain_openai import ChatOpenAI
   
   llm = ChatOpenAI(model='gpt-4o')
   agent = create_react_agent(llm, tools=[])
   result = agent.invoke(
       {'messages': [{'role': 'user', 'content': 'Hello!'}]}
   )

4. Read the LangGraph tutorials to learn about state management, checkpointing, and multi-agent patterns.

Pricing: LangGraph is free and open source. LangGraph Cloud (managed deployment) is available through LangSmith paid plans. You pay only for the LLM API calls your agents make.