OpenAI

Drop-in wrapper for the OpenAI SDK. Every chat or completion call becomes an llm step with full token usage and stop reason.

Setup

from openai import OpenAI
	from theta_observability import TraceClient
	from theta_observability.integrations import wrap_openai

client = TraceClient()
oai = wrap_openai(OpenAI(), client)

with client.trace(name="answer-question") as t:
    oai.chat.completions.create(
        model="gpt-4.1-mini",
        messages=[{"role": "user", "content": "Hi"}],
    )

What gets captured

Provider, model, and request parameters (temperature, top_p, max_tokens, tool definitions).
Full message history sent to the model.
Response text and tool calls.
Token usage (prompt, completion, total) and stop reason.
Latency, request ID, and any error.

Streaming

Streaming responses are captured incrementally — each chunk extends the step output. The step finalizes when the stream closes.

← cURL & raw HTTP

Anthropic →