langchain-middleware

Human-in-the-loop approval, custom middleware, and structured output patterns for LangChain agents. Requirements: Checkpointer + thread_id config for all HITL workflows. @tool def send_email(to: str, subject: str, body: str) -> str: """Send an email.""" return f"Email sent to {to}" agent = create_agent( model="gpt-4.1", tools=[send_email], checkpointer=MemorySaver(), # Required for HITL middleware=[ HumanInTheLoopMiddleware( interrupt_on={ "send_email": {"allowed_decisions": ["approve", "edit", "reject"]}, } ) ], ) config = {"configurable": {"thread_id": "session-1"}} result1 = agent.invoke({ "messages": [{"role": "user", "content": "Send email to john@example.com "}] }, config=config) if " interrupt " in result1: print(f"Waiting for approval: {result1[' interrupt ']}") result2 = agent.invoke( Command(resume={"decisions": [{"type": "approve"}]}), config=config ) Six decorator hooks are available. Two patterns: agent = create_agent( model="gpt-4.1", tools=[send_email], checkpointer=MemorySaver(), # Required middleware=[HumanInTheLoopMiddleware({...})] ) agent.invoke(input, config={"configurable": {"thread_id": "user-123"}}) // CORRECT import { Command } from "@langchain/langgraph"; await agent.invoke(new Command({ resume: { decisions: [{ type: "approve" }] } }), config); - HumanInTheLoopMiddleware pauses execution before dangerous tool calls, allowing humans to approve, edit...