conditional-routing-in-langgraph

Conditional Routing in LangGraph

Purpose

Conditional routing enables LangGraph workflows to make dynamic decisions about execution paths based on state values. This transforms static pipelines into adaptive agentic systems that respond intelligently to runtime conditions.

When to Use This Skill

Use this skill when workflows need to:

• Branch execution based on confidence scores or quality metrics
• Route to different nodes based on classification results
• Implement retry logic with conditional loops
• Handle success/failure paths differently
• Create adaptive multi-path workflows
• Combine state updates with routing decisions (Command API)
• Implement map-reduce parallel workflows (Send API)

Required Imports

from typing import Literal, TypedDict, Annotated, Sequence, Callable, Any
from collections.abc import Hashable  # For type annotations in path functions
from langgraph.graph import StateGraph, START, END
from langgraph.types import Send, Command
import operator  # For reducer functions in parallel execution

Note: The Command API was released December 2024. The Send API has been available since LangGraph 0.2.0+. LangGraph v1.0 was released October 2025 as a stability-focused release. All examples are tested with LangGraph 1.0.x and remain compatible with future 1.x releases. Check PyPI for the latest version.

Convention: Throughout this skill, examples use workflow as the variable name for the StateGraph instance. Create it with: workflow = StateGraph(YourStateClass).

Core Concepts

Conditional Edges

Unlike direct edges (add_edge), conditional edges use functions to determine the next node:

from typing import Literal

def router(state: State) -> Literal["high_confidence_path", "low_confidence_path"]:
    """Returns name of next node based on state."""
    if state["confidence"] > 0.8:
        return "high_confidence_path"
    return "low_confidence_path"

workflow.add_conditional_edges(
    "classify",
    router,
    {
        "high_confidence_path": "respond",
        "low_confidence_path": "clarify"
    }
)