How do you create a human-in-the-loop tool that pauses the agent and waits for user approval before executing?

Q: How do you create a human-in-the-loop tool that pauses the agent and waits for user approval before executing?

langchain provides built-in `humanInTheLoopMiddleware` that intercepts tool calls before execution and waits for human approval[citation:4]. This middleware can be configured to require approval for specific tools only, while allowing others to run automatically. The human operator can approve (execute with original arguments), edit (modify arguments), or reject (provide a manual response) the tool call. This pattern is essential for sensitive operations like writing to a database, sending emails, or making payments where you want human oversight before execution. ```python from langchain import create_agent from langchain.agents.middleware import humanInTheLoopMiddleware # Configure which tools require approval hitl_middleware = humanInTheLoopMiddleware( interrupt_on={ "send_email": { "allowedDecisions": ["approve", "edit", "reject"], "description": "⚠️ Sending an email requires your approval" }, "delete_record": { "allowedDecisions": ["approve", "reject"], # No edits allowed "description": "⚠️ Deleting database records requires confirmation" }, "search_web": False # Auto-approve (no interruption) } ) agent = create_agent( model="openai:gpt-4o", tools=[send_email, delete_record, search_web], middleware=[hitl_middleware] ) # When the agent calls send_email, execution pauses # The system sends an approval request to the user # Only resumes after user responds (approve/edit/reject) # Requires a checkpointer to persist interrupted state from langgraph.checkpoint.memory import MemorySaver agent = create_agent( model="openai:gpt-4o", tools=[send_email], middleware=[hitl_middleware], checkpointer=MemorySaver() ) ```

Difficulty: 5/10

human-in-the-loop, LangChain callbacks, agent control flow

Create a human-in-the-loop tool using langchain's built-in `humanInTheLoopMiddleware`, which intercepts tool calls, sends approval requests, and resumes execution only after receiving user input (approve, edit, or reject).

langchain provides built-in humanInTheLoopMiddleware that intercepts tool calls before execution and waits for human approval[citation:4]. This middleware can be configured to require approval for specific tools only, while allowing others to run automatically. The human operator can approve (execute with original arguments), edit (modify arguments), or reject (provide a manual response) the tool call. This pattern is essential for sensitive operations like writing to a database, sending emails, or making payments where you want human oversight before execution.

Implementing Human-in-the-Loop Middleware

How do you create a human-in-the-loop tool that pauses the agent and waits for user approval before executing?

Difficulty: 5/10

human-in-the-loop, LangChain callbacks, agent control flow

Create a human-in-the-loop tool using langchain's built-in `humanInTheLoopMiddleware`, which intercepts tool calls, sends approval requests, and resumes execution only after receiving user input (approve, edit, or reject).

Implementing Human-in-the-Loop Middleware

How do you create a human-in-the-loop tool that pauses the agent and waits for user approval before executing?

Create a human-in-the-loop tool using langchain's built-in `humanInTheLoopMiddleware`, which intercepts tool calls, sends approval requests, and resumes execution only after receiving user input (approve, edit, or reject).

How do you create a human-in-the-loop tool that pauses the agent and waits for user approval before executing?

Create a human-in-the-loop tool using langchain's built-in `humanInTheLoopMiddleware`, which intercepts tool calls, sends approval requests, and resumes execution only after receiving user input (approve, edit, or reject).

Scenario Questions

Follow-up Questions

How do you create a human-in-the-loop tool that pauses the agent and waits for user approval before executing?

Create a human-in-the-loop tool using langchain's built-in humanInTheLoopMiddleware, which intercepts tool calls, sends approval requests, and resumes execution only after receiving user input (approve, edit, or reject).

How do you create a human-in-the-loop tool that pauses the agent and waits for user approval before executing?

Create a human-in-the-loop tool using langchain's built-in humanInTheLoopMiddleware, which intercepts tool calls, sends approval requests, and resumes execution only after receiving user input (approve, edit, or reject).

Scenario Questions

Follow-up Questions

Create a human-in-the-loop tool using langchain's built-in `humanInTheLoopMiddleware`, which intercepts tool calls, sends approval requests, and resumes execution only after receiving user input (approve, edit, or reject).

Create a human-in-the-loop tool using langchain's built-in `humanInTheLoopMiddleware`, which intercepts tool calls, sends approval requests, and resumes execution only after receiving user input (approve, edit, or reject).