opik-trace

name	opik-trace
description	Add LLM tracing and observability to your code. Use when instrumenting functions, integrating frameworks (LangChain, OpenAI, etc.), or adding custom spans.

opik-trace

Quick Reference

Decorator:  @opik.track(name="my_func")
OpenAI:     client = track_openai(OpenAI())
LangChain:  config={"callbacks": [OpikTracer()]}
LlamaIndex: set_global_handler("opik")

Quick Health Check

Python: Run opik healthcheck TypeScript: Verify config exists at ~/.opik.config or env vars are set

✅ "Connection successful" / config exists → Continue below ❌ "Connection failed" / no config → Run /opik-setup first, then return here

Note: OPIK_PROJECT_NAME is optional - traces go to "default" project if unset.

Framework Detection

Check your project dependencies to find the right integration pattern:

Python: Look in requirements.txt or pyproject.toml TypeScript/Node: Look in package.json

If you see...	Use this pattern
openai	Client Wrapper
anthropic	Client Wrapper
langchain	Callback Handler
llama-index or llama_index	Global Handler
None of the above	Decorator Pattern

Integration Patterns

Client Wrapper

For OpenAI and Anthropic SDKs.

Python:

from opik.integrations.openai import track_openai
from openai import OpenAI

client = track_openai(OpenAI())
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello"}]
)
# Automatically traced!

TypeScript:

import { trackOpenAI } from "opik-openai";
import OpenAI from "openai";

const client = trackOpenAI(new OpenAI());
const response = await client.chat.completions.create({
  model: "gpt-4",
  messages: [{ role: "user", content: "Hello" }]
});
// Automatically traced!

Note: TypeScript requires npm install opik opik-openai

Callback Handler

For LangChain.

Python:

from opik.integrations.langchain import OpikTracer

tracer = OpikTracer()
chain.invoke(input, config={"callbacks": [tracer]})

Global Handler

For LlamaIndex.

Python:

from llama_index.core import set_global_handler

set_global_handler("opik")
# All LlamaIndex operations now traced

Decorator Pattern

For custom functions.

Python:

import opik

@opik.track(name="process_query")
def process_query(query: str) -> str:
    result = f"Processed: {query}"  # Your logic here
    return result

TypeScript:

import { track } from "opik";

const processQuery = track(async (query: string) => {
  const result = `Processed: ${query}`;  // Your logic here
  return result;
}, { name: "process_query" });

Next Steps

See FRAMEWORKS.md for detailed integration patterns for each framework.

Common Options

Option	Description
name	Custom name for the span
project_name	Override default project
tags	List of tags for filtering
metadata	Additional key-value data