Traceloop — OpenTelemetry-first LLM Observability

Quick Start — One-line Init

Traceloop.init() wires up OpenTelemetry and auto-instruments every supported library it finds in the Python path. The @workflow and @task decorators create explicit spans — useful for naming the steps the way your team thinks about them, so traces read like operations not library internals.

# pip install traceloop-sdk
from traceloop.sdk import Traceloop
from traceloop.sdk.decorators import workflow, task

# Point at whatever OTEL backend you prefer
# (Traceloop Cloud, Jaeger, Grafana Tempo, Datadog, Honeycomb, ...)
Traceloop.init(
    app_name="tokrepo-demo",
    api_endpoint="https://api.traceloop.com",  # or your own OTEL collector URL
    api_key="tl-...",                            # or None for self-hosted
)

# Auto-instruments: OpenAI, Anthropic, LangChain, LlamaIndex, Pinecone, Chroma, ...
from openai import OpenAI
client = OpenAI()

@task(name="answer_question")
def answer(q: str) -> str:
    resp = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": q}],
    )
    return resp.choices[0].message.content

@workflow(name="qa_chain")
def qa_chain(user_q: str) -> str:
    # Nested spans: qa_chain → answer_question → OpenAI LLM call
    return answer(user_q)

print(qa_chain("Why does OpenTelemetry matter for LLMs?"))
# Backend now shows a trace tree with latency/cost/prompt at every level.

Key Features

Comparison

Use Cases

Pricing & License

Frequently Asked Questions