Skip to main content
Wrap an OpenAI or AsyncOpenAI client with wrap_llm() to auto-trace every chat.completions.create() call and tool use block in the response.
Using OpenAI Agents SDK? See the dedicated OpenAI Agents integration — lifecycle hooks cover the full agent runner.

​
Installation

pip install langsight openai

​
Quick start

import langsight
from openai import OpenAI

ls = langsight.init()

raw_client = OpenAI()
client = ls.wrap_llm(raw_client, agent_name="my-agent", session_id="sess-001")

response = client.chat.completions.create(
    model="gpt-4o",
    tools=[{"type": "function", "function": {"name": "get_weather", ...}}],
    messages=[{"role": "user", "content": "What's the weather in Berlin?"}],
)
# response is unchanged — same OpenAI response object
# LangSight auto-traces: LLM generation span + tool_use spans
The wrapper intercepts client.chat.completions.create() (sync) and client.chat.completions.acreate() (async).

​
What gets traced

SpanFields
LLM generationserver_name="openai", tool_name="generate/gpt-4o", span_type="agent", tokens, model_id
Tool call (per tool)tool_name from response, span_type="tool_call", parent_span_id, input_args
Session: sess-001
├── openai/generate/gpt-4o          1200ms  success  (512 in / 128 out tokens)
│   ├── get_weather                                   success
│   └── get_forecast                                  success

​
Parameters

client = ls.wrap_llm(
    OpenAI(),                  # required — openai.OpenAI or openai.AsyncOpenAI
    agent_name="my-agent",     # optional — shown in dashboard
    session_id="sess-001",     # optional — groups spans into a session
    trace_id="trace-abc",      # optional — links across multi-agent tasks
)