JS Integration Litellm Proxy
This is a Jupyter notebook

Cookbook: LiteLLM (Proxy) + Langfuse OpenAI Integration (JS/TS)

This notebook demonstrates how to use the following stack to experiment with 100+ LLMs from different providers without changing code:

Let's get started!

Install dependencies

Note: This cookbook uses Deno.js, which requires different syntax for importing packages and setting environment variables.

import { OpenAI } from "npm:openai@^4.0.0";
import { observeOpenAI } from "npm:langfuse@^3.6.0";

Setup environment

// Set env variables, Deno-specific syntax
Deno.env.set("OPENAI_API_KEY", "");
Deno.env.set("LANGFUSE_PUBLIC_KEY", "");
Deno.env.set("LANGFUSE_SECRET_KEY", "");
Deno.env.set("LANGFUSE_HOST", ""); // 🇪🇺 EU region
// Deno.env.set("LANGFUSE_HOST", ""); // 🇺🇸 US region

Setup Lite LLM Proxy

In this example, we'll use GPT-3.5-turbo directly from OpenAI, and llama3 and mistral via the Ollama on our local machine.


  1. Create a litellm_config.yaml to configure which models are available (docs (opens in a new tab)). We'll use gpt-3.5-turbo, and llama3 and mistral via Ollama in this example. Make sure to replace <openai_key> with your OpenAI API key.
      - model_name: gpt-3.5-turbo
          model: gpt-3.5-turbo
          api_key: <openai_key>
      - model_name: ollama/llama3
          model: ollama/llama3
      - model_name: ollama/mistral
          model: ollama/mistral
  2. Ensure that you installed Ollama and have pulled the llama3 (8b) and mistral (7b) models: ollama pull llama3 && ollama pull mistral
  3. Run the following cli command to start the proxy: litellm --config litellm_config.yaml

The Lite LLM Proxy should be now running on (opens in a new tab)

To verify the connection you can run litellm --test

Log single LLM Call via Langfuse OpenAI Wrapper

The Langfuse SDK offers a wrapper function around the OpenAI SDK, automatically logging all OpenAI calls as generations to Langfuse.

We wrap the client for each call separately in order to be able to pass a name.

For more details, please refer to our documentation (opens in a new tab).

const PROXY_URL = "";
const client = observeOpenAI(new OpenAI({baseURL: PROXY_URL}));
const systemPrompt = "You are a very accurate calculator. You output only the result of the calculation.";
const gptCompletion = await{
  model: "gpt-3.5-turbo", 
  messages: [
    {role: "system", content: systemPrompt},
    {role: "user", content: "1 + 1 = "}
const llamaCompletion = await{
  model: "ollama/llama3",
  messages: [
    {role: "system", content: systemPrompt},
    {role: "user", content: "3 + 3 = "}
// notebook only: await events being flushed to Langfuse
await client.flushAsync();

Public trace links for the following examples:

Trace nested LLM Calls using Langfuse JS SDK

To capture nested LLM calls, use langfuse.trace to create a parent trace and pass it to observeOpenAI. This allows you to group multiple generations into a single trace, providing a comprehensive view of the interactions. You can also add rich metadata to the trace, such as custom names, tags, and userid. For more details, refer to the Langfuse JS/TS SDK documentation (opens in a new tab).

We'll use the trace to log a rap battle between GPT-3.5-turbo, llama3, and mistral.

import { Langfuse } from "npm:langfuse";
const langfuse = new Langfuse();
async function rapBattle(topic: string) {
  const trace = langfuse.trace({name: "Rap Battle", input: topic});
  let messages = [
    {role: "system", content: "You are a rap artist. Drop a fresh line."},
    {role: "user", content: `Kick it off, today's topic is ${topic}, here's the mic...`}
  const gptCompletion = await observeOpenAI(new OpenAI({baseURL: PROXY_URL}), {
      parent: trace, generationName: "rap-gpt-3.5-turbo"
    model: "gpt-3.5-turbo",
  const firstRap = gptCompletion.choices[0].message.content;
  messages.push({role: "assistant", content: firstRap});
  console.log("Rap 1:", firstRap);
  const llamaCompletion = await observeOpenAI(new OpenAI({baseURL: PROXY_URL}), {
      parent: trace, generationName: "rap-llama3"
    model: "ollama/llama3", 
  const secondRap = llamaCompletion.choices[0].message.content;
  messages.push({role: "assistant", content: secondRap});
  console.log("Rap 2:", secondRap);
  const mistralCompletion = await observeOpenAI(new OpenAI({baseURL: PROXY_URL}), {
      parent: trace, generationName: "rap-mistral"
    model: "ollama/mistral",
  const thirdRap = mistralCompletion.choices[0].message.content;
  messages.push({role: "assistant", content: thirdRap});
  console.log("Rap 3:", thirdRap);
  trace.update({output: messages})
  return messages;
await rapBattle("typography");
await langfuse.flushAsync();

Example Trace (public link (opens in a new tab))

Public Trace

Learn more

Check out the docs to learn more about all components of this stack:

If you do not want to capture traces via the OpenAI SDK Wrapper, you can also directly log requests from the LiteLLM Proxy to Langfuse. For more details, refer to the LiteLLM Docs (opens in a new tab).

Was this page useful?

Questions? We're here to help

Subscribe to updates