LangChain Python Basic

This example shows how to trace LangChain Python runs and wrap tools for enforcement.

Full code

import asyncio
import os

from langchain_openai import ChatOpenAI

from njiraai import NjiraAI
from njiraai_langchain import NjiraCallbackHandler, wrap_tool


async def main():
    njira = NjiraAI(
        api_key=os.environ["NJIRA_API_KEY"],
        project_id=os.environ["NJIRA_PROJECT_ID"],
        mode="active",
    )

    # Create callback handler for automatic tracing
    handler = NjiraCallbackHandler(njira)

    # Initialize LLM
    llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)

    # Example tool
    def calculator(expression: str) -> str:
        try:
            return str(eval(expression))
        except Exception:
            return "Error: invalid expression"

    # Wrap tool for enforcement
    safe_calculator = wrap_tool(
        calculator,
        njira,
        tool_name="calculator",
        description="Evaluates math expressions",
    )

    # Traced LLM call with callback
    response = await llm.ainvoke("What is 2 + 2?", config={"callbacks": [handler]})
    print(f"LLM response: {response.content}")

    # Enforced tool call
    result = await safe_calculator("2 + 2")
    print(f"Calculator result: {result}")

    # Flush traces
    await njira.flush()


if __name__ == "__main__":
    asyncio.run(main())

Run it

cd sdks/python/examples/langchain-basic
uv sync
NJIRA_API_KEY=your-key NJIRA_PROJECT_ID=your-project OPENAI_API_KEY=your-openai-key uv run python main.py

What's happening

1. NjiraCallbackHandler automatically captures spans for LLM calls, chains, and tool invocations
2. wrap_tool adds pre/post enforcement around tool execution
3. Traces are flushed at the end to ensure delivery