Skip to main content
Most observability tools sit above the model client. Keep your tracing setup the same, then point the underlying OpenAI-compatible or Anthropic-compatible client at AI Stats.

​
Langfuse with OpenAI SDK

pip install langfuse openai

import os
from langfuse.openai import openai

client = openai.OpenAI(
    api_key=os.environ["AI_STATS_API_KEY"],
    base_url="https://api.phaseo.app/v1",
)

response = client.responses.create(
    model="openai/gpt-5-nano",
    input="Reply with only: ok",
)

​
LangSmith with LangChain

Use the LangChain setup and enable LangSmith through the usual LangChain environment variables: 
export LANGSMITH_TRACING=true
export LANGSMITH_API_KEY="your_langsmith_key"
export AI_STATS_API_KEY="your_ai_stats_key"

​
OpenTelemetry

If your framework emits OpenTelemetry spans, AI Stats does not require a special exporter. Add useful request context yourself:
  • model id
  • endpoint, such as responses or chat.completions
  • AI Stats request id from response metadata or logs
  • selected provider when response metadata is enabled

​
Notes

  • Do not send AI Stats API keys to observability tools as attributes.
  • If the observability library wraps the OpenAI SDK, configure the wrapped client with base_url: "https://api.phaseo.app/v1".
  • For LangChain traces, pair the tracing callback with the AI Stats ChatOpenAI configuration.
Last modified on June 11, 2026