haive.core.tools.interrupt_tool_wrapper

from typing import Any Human-in-the-Loop Tool Wrapper for LangGraph Agents.

This module defines a utility function add_human_in_the_loop that allows LangChain tools to be wrapped with interrupt-based human review via LangGraph. This enables human approval, editing, or feedback substitution before a tool is executed.

Typical usage:

from this_module import add_human_in_the_loop

@tool def search_docs(query: str) -> str:

return f”Results for: {query}”

safe_tool = add_human_in_the_loop(search_docs) result = safe_tool.invoke({“query”: “pydantic base models”})

Functions

add_human_in_the_loop(tool, *[, interrupt_config])

Wrap a LangChain tool with human-in-the-loop interrupt logic for approval, editing, or feedback.

Module Contents

haive.core.tools.interrupt_tool_wrapper.add_human_in_the_loop(tool, *, interrupt_config=None)

Wrap a LangChain tool with human-in-the-loop interrupt logic for approval, editing, or feedback.

This function wraps an existing LangChain tool (or plain callable) with LangGraph’s interrupt system, allowing a human to review each call before it is executed.

Parameters:

• tool (Callable | BaseTool) – The LangChain tool (or callable function) to wrap with human-in-the-loop support. If a plain callable is passed, it will be converted into a LangChain BaseTool.

• interrupt_config (HumanInterruptConfig, optional) –

  Configuration dict defining which types of human input are allowed. If not provided, the default enables all three options:

  {
      "allow_accept": True,
      "allow_edit": True,
      "allow_respond": True,
  }
  

Returns:

A LangChain-compatible tool that prompts a human to approve, edit, or respond to each call before invoking the original tool logic.

Return type:

BaseTool

Raises:

ValueError – If the human interrupt returns an unsupported response type.

Example

>>> @tool
... def get_user_profile(user_id: str) -> str:
...     return f"Profile for {user_id}"

>>> reviewed_tool = add_human_in_the_loop(get_user_profile)
>>> reviewed_tool.invoke({"user_id": "123"})
# Will prompt human before invoking