LangGraph

Type Annotations

Langgraph SDK highly involves following Python type annotations:

• typed dictionary
• union
• optional
• any
• lambda function

Langgraph Elements

• State: shared data structure stores information
• Nodes: individual processing units
• Graph: represents the workflow
• Edges: determine which node to execute next
  • Conditional edge: based on specific conditions
• Start
• End
• Tools
  • ToolNode: the main job is to run a tool
• StateGraph
• Runnable: can be executed
• Messages
  • Human Message
  • AI Message
  • System Message
  • Tool Message
  • Function Message

Python Tricks

Annotated

Annotated can provide additional context without affecting the type system.

email = Annotated[str, "a valid email address"]
print(email.__metadata__)

Docstring

In Langgraph, tool functions must have docstrings if description not provided.