LangGraph Development

Source: https://github.com/langchain-ai/docs (src/oss/langgraph/)

LangGraph is the low-level orchestration framework for building stateful, long-running agents. It models workflows as graphs with nodes (functions) and edges (transitions). Use LangGraph when you need fine-grained control over agent behavior; use LangChain's create_agent for simpler use cases.

Core Concepts

• State: Shared data structure representing the current snapshot
• Nodes: Functions that process state and return updates
• Edges: Determine which node runs next (conditional or fixed)

"Nodes do the work, edges tell what to do next."

Quick Start

from langgraph.graph import StateGraph, MessagesState, START, END

def call_model(state: MessagesState):
    # Your LLM logic here
    return {"messages": [{"role": "ai", "content": "Hello!"}]}

graph = StateGraph(MessagesState)
graph.add_node("model", call_model)
graph.add_edge(START, "model")
graph.add_edge("model", END)

app = graph.compile()
result = app.invoke({"messages": [{"role": "user", "content": "Hi!"}]})

State Definition

Using TypedDict (Recommended)

from typing import TypedDict, Annotated
from langgraph.graph import StateGraph
from operator import add

class State(TypedDict):
    messages: Annotated[list, add]  # Reducer: append to list
    count: int                       # Overwrite on update

graph = StateGraph(State)

Using MessagesState (Convenience)

from langgraph.graph import StateGraph, MessagesState

# MessagesState has a pre-defined "messages" key with append reducer
graph = StateGraph(MessagesState)

Adding Nodes

def node_a(state: State) -> dict:
    """Nodes receive state and return updates."""
    return {"count": state["count"] + 1}

def node_b(state: State) -> dict:
    return {"messages": [{"role": "ai", "content": f"Count is {state['count']}"}]}

graph.add_node("a", node_a)
graph.add_node("b", node_b)

Adding Edges

Fixed Edges

from langgraph.graph import START, END

graph.add_edge(START, "a")
graph.add_edge("a", "b")
graph.add_edge("b", END)

Conditional Edges

def router(state: State) -> str:
    """Return the name of the next node."""
    if state["count"] > 5:
        return "end_node"
    return "continue_node"

graph.add_conditional_edges(
    "a",
    router,
    {"end_node": END, "continue_node": "b"}
)

Persistence (Checkpointing)

Enable durable execution with checkpointers:

from langgraph.checkpoint.memory import MemorySaver

checkpointer = MemorySaver()
app = graph.compile(checkpointer=checkpointer)

# Each thread_id maintains separate state
config = {"configurable": {"thread_id": "user-123"}}
result = app.invoke({"messages": [...]}, config)

# Resume from checkpoint
result = app.invoke({"messages": [...]}, config)

Production Checkpointers

# PostgreSQL
from langgraph.checkpoint.postgres import PostgresSaver
checkpointer = PostgresSaver.from_conn_string("postgresql://...")

# SQLite
from langgraph.checkpoint.sqlite import SqliteSaver
checkpointer = SqliteSaver.from_conn_string("sqlite:///checkpoints.db")

Streaming

app = graph.compile()

# Stream node outputs
for chunk in app.stream({"messages": [...]}):
    print(chunk)

# Stream specific events
for event in app.stream({"messages": [...]}, stream_mode="updates"):
    print(event)

# Stream tokens from LLM
for event in app.stream({"messages": [...]}, stream_mode="messages"):
    print(event)

Human-in-the-Loop

Interrupts

app = graph.compile(
    checkpointer=checkpointer,
    interrupt_before=["sensitive_node"],  # Pause before this node
)

config = {"configurable": {"thread_id": "123"}}

# Run until interrupt
result = app.invoke({"messages": [...]}, config)

# Review state, then continue
result = app.invoke(None, config)  # Resume from checkpoint

Dynamic Breakpoints

from langgraph.errors import NodeInterrupt

def my_node(state: State):
    if needs_approval(state):
        raise NodeInterrupt("Approval required")
    return {"result": "processed"}

Subgraphs

Compose complex workflows from smaller graphs:

# Define inner graph
inner_graph = StateGraph(State)
inner_graph.add_node("inner_node", inner_fn)
inner_app = inner_graph.compile()

# Use as node in outer graph
outer_graph = StateGraph(State)
outer_graph.add_node("subgraph", inner_app)
outer_graph.add_edge(START, "subgraph")

ReAct Agent Pattern

from langgraph.graph import StateGraph, MessagesState, START, END
from langgraph.prebuilt import ToolNode

def call_model(state: MessagesState):
    response = model.invoke(state["messages"])
    return {"messages": [response]}

def should_continue(state: MessagesState) -> str:
    last_message = state["messages"][-1]
    if last_message.tool_calls:
        return "tools"
    return END

graph = StateGraph(MessagesState)
graph.add_node("model", call_model)
graph.add_node("tools", ToolNode(tools))

graph.add_edge(START, "model")
graph.add_conditional_edges("model", should_continue, {"tools": "tools", END: END})
graph.add_edge("tools", "model")

app = graph.compile()

Best Practices

1. Use TypedDict for state - Clear typing, good performance
2. Define reducers explicitly - Control how state updates are merged
3. Use checkpointers in production - Enable durability and recovery
4. Keep nodes focused - Single responsibility, easier testing
5. Use conditional edges for branching - Clean separation of routing logic
6. Stream for long operations - Better UX, visible progress

When to Use LangGraph vs LangChain

Documentation Index

Syncing Documentation

cd skills/langgraph
bun run scripts/sync-docs.ts