LangChain

XeroML provides a native LangChain integration via a CallbackHandler that automatically captures all LangChain events as structured observations.

Python

Installation:

pip install xeroml langchain-openai

Usage:

from xeroml import CallbackHandler
from langchain_openai import ChatOpenAI
from langchain.schema import HumanMessage

handler = CallbackHandler()
llm = ChatOpenAI(model="gpt-4o", callbacks=[handler])

response = llm.invoke([HumanMessage(content="What is XeroML?")])

With chains:

from langchain.chains import LLMChain
from langchain.prompts import PromptTemplate

template = PromptTemplate(
    input_variables=["product"],
    template="Write a tagline for {product}."
)

chain = LLMChain(llm=llm, prompt=template)
result = chain.run("XeroML")

Adding user/session context:

from xeroml import CallbackHandler, propagate_attributes

def handle_request(message: str, user_id: str, session_id: str) -> str:
    handler = CallbackHandler()
    llm = ChatOpenAI(model="gpt-4o", callbacks=[handler])

    with propagate_attributes(user_id=user_id, session_id=session_id):
        return llm.invoke([HumanMessage(content=message)]).content

TypeScript

Installation:

npm install @xeroml/langchain @xeroml/otel @opentelemetry/sdk-node @langchain/openai

Setup:

import { NodeSDK } from "@opentelemetry/sdk-node";
import { XeroMLSpanProcessor } from "@xeroml/otel";

const sdk = new NodeSDK({
  spanProcessors: [new XeroMLSpanProcessor()],
});
sdk.start();

Usage:

import { XeroMLCallbackHandler } from "@xeroml/langchain";
import { ChatOpenAI } from "@langchain/openai";
import { HumanMessage } from "@langchain/core/messages";

const handler = new XeroMLCallbackHandler();
const llm = new ChatOpenAI({ model: "gpt-4o", callbacks: [handler] });

const response = await llm.invoke([new HumanMessage("What is XeroML?")]);

With propagation:

import { propagateAttributes } from "@xeroml/tracing";

async function handleRequest(message: string, userId: string, sessionId: string) {
  return await propagateAttributes({ userId, sessionId }, async () => {
    const handler = new XeroMLCallbackHandler();
    const llm = new ChatOpenAI({ callbacks: [handler] });
    return await llm.invoke([new HumanMessage(message)]);
  });
}

What Gets Captured

The LangChain integration captures:

• Chain start/end with inputs and outputs
• LLM call inputs (prompt), outputs (completion), and token usage
• Tool calls and their results
• Agent reasoning steps
• Retriever queries and retrieved documents
• Errors at any step