LLM Tracekit

2.8.1 verified Fri May 01 auth: no python

Meta-package for LLM Tracekit - OpenTelemetry instrumentations for LLM providers. Current version: 2.8.1. Release cadence: frequent, follows semantic versioning.

pip install llm-tracekit

Common errors

error ModuleNotFoundError: No module named 'llm_tracekit' ↓

cause Package not installed; typo in package name llm-tracekit vs llm_tracekit

fix

Install: pip install llm-tracekit. Import uses underscores: import llm_tracekit

error ImportError: cannot import name 'Tracekit' from 'llm_tracekit' ↓

cause Using old API from v1.x (from llm_tracekit.core import Tracekit) with v2.x

fix

Use: from llm_tracekit import Tracekit

error TypeError: Tracekit() missing 1 required positional argument: 'service_name' ↓

cause service_name is required since v2.5.0

fix

Provide service_name or set LLM_TRACEKIT_SERVICE_NAME environment variable

Warnings

gotcha If you install llm-tracekit without opentelemetry-sdk, console export may fail silently. ↓

fix Install opentelemetry-sdk: pip install opentelemetry-sdk

breaking Tracekit class moved from llm_tracekit.core to llm_tracekit in v2.0. Old import breaks. ↓

fix Change 'from llm_tracekit.core import Tracekit' to 'from llm_tracekit import Tracekit'

gotcha Setting enable_console=True may cause high log volume in production. ↓

fix Use only for debugging; set via environment variable OTEL_TRACES_EXPORTER=console for consistency.

Imports

Tracekit

wrong

from llm_tracekit import LLMTracekit

correct

from llm_tracekit import Tracekit

Class is named Tracekit, not LLMTracekit

enable_instrumentations

wrong

from llm_tracekit.instrument import enable

correct

from llm_tracekit import enable_instrumentations

Function is at top-level, not nested

Quickstart

Initialize Tracekit with basic OpenTelemetry configuration. Ensure OTEL_EXPORTER_OTLP_ENDPOINT is set for remote export.

import os
from llm_tracekit import Tracekit

tk = Tracekit(
    service_name="my-llm-app",
    otlp_endpoint=os.environ.get('OTEL_EXPORTER_OTLP_ENDPOINT', 'http://localhost:4317'),
    enable_console=True
)
tk.start()

# Use with OpenAI, Anthropic, etc.