langchain-dev-utils 0.1.18

A practical utility library for LangChain and LangGraph development

langchain-dev-utils

A practical enhancement utility library for LangChain / LangGraph developers, empowering the construction of complex and maintainable large language model applications.

🚀 Installation

pip install -U langchain-dev-utils

📦 Core Features

1. Model Management

• Supports registering any chat model or embedding model provider
• Provides unified interfaces load_chat_model() / load_embeddings() to simplify model loading
• Fully compatible with LangChain’s official init_chat_model / init_embeddings, enabling seamless extension

from langchain_dev_utils import register_model_provider, load_chat_model
from langchain_qwq import ChatQwen

register_model_provider("dashscope", ChatQwen)
register_model_provider("openrouter", "openai", base_url="https://openrouter.ai/api/v1")

model = load_chat_model("dashscope:qwen-flash")
print(model.invoke("Hello!"))

---

2. Message Processing

• Automatically merges reasoning content (e.g., from DeepSeek models) into the content field
• Supports streaming and asynchronous streaming responses (stream / astream)
• Utility functions include:
  • merge_ai_message_chunk(): merges message chunks
  • has_tool_calling() / parse_tool_calling(): detects and parses tool calls
  • message_format(): formats messages or document lists (with numbering, separators, etc.)

from langchain_dev_utils import has_tool_calling, parse_tool_calling

response = model.invoke("What time is it now?")
if has_tool_calling(response):
    tool_calls = parse_tool_calling(response)
    print(tool_calls)

---

3. Tool Enhancement

• Easily extend existing tools with new capabilities
• Currently supports adding human-in-the-loop functionality to tools

from langchain_dev_utils import human_in_the_loop_async
from langchain_core.tools import tool
import asyncio
import datetime

@human_in_the_loop_async
@tool
async def async_get_current_time() -> str:
    """Asynchronously retrieve the current timestamp"""
    await asyncio.sleep(1)
    return str(datetime.datetime.now().timestamp())

---

4. Context Engineering

• Automatically generates essential context management tools:

  • create_write_plan_tool() / create_update_plan_tool()
  • create_write_note_tool() / create_query_note_tool() / create_ls_tool() / create_update_note_tool()

• Provides corresponding State classes—no need to reimplement them

from langchain_dev_utils import (
    create_write_plan_tool,
    create_update_plan_tool,
    create_write_note_tool,
    create_ls_tool,
    create_query_note_tool,
    create_update_note_tool,
)

plan_tools = [create_write_plan_tool(), create_update_plan_tool()]
note_tools = [create_write_note_tool(), create_ls_tool(), create_query_note_tool(), create_update_note_tool()]

---

5. Graph Orchestration

• Composes multiple StateGraphs in sequential or parallel fashion
• Supports complex multi-agent workflows:
  • sequential_pipeline(): executes subgraphs sequentially
  • parallel_pipeline(): executes subgraphs in parallel with dynamic branching (via the Send API)
• Allows specifying entry nodes and custom state/input/output schemas

from langchain_dev_utils import parallel_pipeline

graph = parallel_pipeline(
    sub_graphs=[graph1, graph2, graph3],
    state_schema=State,
    branches_fn=lambda state: [
        Send("graph1", arg={"a": state["a"]}),
        Send("graph2", arg={"a": state["a"]}),
    ]
)

---

📚 Documentation and Examples

For more information, please refer to the following documentation.