LangGraph

Langgraph is a framework based on LangChain to build agentic workflows where you define the flow and the possible routes, while the LLM takes the decision on how to follow through!

Lang-Graph is not as restrictive as Langchain with it’s router system as the LLM can decide on what tools and flows it can take, they provide memory and a persistence layer as well.

This framework has three main components

• State - The state of the application is stored in a Pydantic BaseModel or a TypedDict this state is shared between all the application nodes
• Nodes - These encapsulate the application logic, they take in the state and modify the state and return it, or they use the state and do something else
• Edges - These are the guiding system, they give information on what node should be executed next

There are different types of edges

• Normal edges - if you know that after a certain step you always call node 2, then you use the normal edge
• Conditional Edge - You call a function to determine which nodes to move to next
• Entry Point - the node to call when the user input arrives
• Conditional Entry point - call a function to determine which nodes to call first when user input arrives.

In short: _nodes do the work. edges tell what to do next _

refer to The Open Source LLM Agent Handbook How to Automate Complex Tasks with LangGraph and CrewAI > What Is LangGraph?