Introduction
- Overview
- StateGraph - Components
- State : is the shared memory or context that persists and updates as the graph runs
- Node : are the units of work (like agents, tools, or logic steps).
- Edge : define how data or control flows between nodes
- Functional Mechanism
- Reducer : helps control how the state is updated when multiple nodes modify it (default: overwrite)
- Command : a special return value that a node can produce to tell the graph how to continue execution
- Subgraph : lets you group logic into reusable workflows (can act like nodes themselves)
- Checkpoint : Mechanism for saving and restoring state during long-running or interruptible workflows
- Streaming : Langgraph supports real-time streaming of intermediate outputs from LLMs
- Concurrency(Async, Parallel) : Some graphs can run multiple nodes in parallel if their dependencies don’t overlap
State
- State : is the shared memory or context that persists and updates as the graph runs
- Implementation: TypedDict or Pydantic BaseModel
from typing import Any, TypedDict
class State(TypedDict):
messages : dict[str]
Node
- Node : are the units of work (like agents, tools, or logic steps).
- Implementation : python function
def add_message(state: State) → dict[str, Any]
# return the fields to update
return {"messages": "Message Added"}
Edge
- Edge : define how data or control flows between nodes
- Types
- Start Edge
- Normal Edge
- Conditional Edge
- End edge
# Define Graph by connecting edges
graph = StateGraph(state)
# Start Edge
graph.add_edge(START, first_node)
# Conditional Edge
graph.add_conditional_edges(
"decision_node",
routing_funtion,
{"option_a": "node_a", "option_b", "node_b"}
)
# Normal Edge
graph.add_edge("node_a", "node_b")
# End Edg
graph.add_edge("node_b", END)
References
