Langtrace adds support for Cleanlab TLM

Karthik Kalyanaraman

⸱

Cofounder and CTO

Mar 5, 2025

Introduction

We are thrilled to announce our latest integration update — Langtrace now supports Cleanlab's TLM natively! This integration enables you to automatically capture traces that includes the trustworthiness score, explanation and other metadata returned by Cleanlab's APIs. With this seamless integration, you can now gain deeper insights into your language model interactions while leveraging Cleanlab's robust capabilities.

Setup

Follow these steps to integrate Langtrace with Cleanlab:

Create Your Langtrace Project
Sign up on Langtrace, create a new project, and obtain your Langtrace API key.
Install the Langtrace Python SDK
pip install langtrace-python-sdk
Configure Your Environment
Set up your environment variables by creating a .env file and adding your Langtrace API key:
export LANGTRACE_API_KEY=<YOUR_LANGTRACE_API_KEY>
Similarly, ensure you have your Cleanlab API key available as an environment variable
export CLEANLAB_API_KEY=<YOUR_CLEANLAB_API_KEY>

Initialize and set up Langtrace in your code. See code sample below:

import os

from dotenv import find_dotenv, load_dotenv
from langtrace_python_sdk import langtrace
from langtrace_python_sdk.utils.with_root_span import with_langtrace_root_span

_ = load_dotenv(find_dotenv())

langtrace.init()

from cleanlab_tlm import TLM
from openai import OpenAI

openai_client = OpenAI()

tlm = TLM(
    api_key=os.getenv("TLM_API_KEY"),
    options={"log": ["explanation"], "model": "gpt-4o-mini"},
)


def inference(prompt: str):
    response = openai_client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role": "user", "content": prompt},
        ],
        stream=False,
    )
    response_text = response.choices[0].message.content
    return response_text


@with_langtrace_root_span("Get Trustworthiness Score")
def inference_get_trustworthiness_score(prompt: str):
    response = inference(prompt)
    return tlm.get_trustworthiness_score(prompt, response)


print(inference_get_trustworthiness_score("How many r's are in strawberry?"))

Review Your Traces in Langtrace
Once you run your code, Langtrace will automatically capture and log all relevant interactions. Visit your Langtrace dashboard to see the traces.

Additional Resources