Implementation:Wandb Weave OpenAI Patcher

Knowledge Sources	Wandb Weave Weave Docs OpenAI SDK
Domains	Observability, LLM_Operations
Last Updated	2026-02-14 00:00 GMT

Overview

Wrapper documentation for the OpenAI SDK integration patcher provided by the Wandb Weave library.

Description

get_openai_patcher() returns a MultiPatcher containing 14 SymbolPatcher instances that wrap OpenAI SDK methods (sync and async variants) for chat completions, parse, moderations, embeddings, and responses. The patcher uses create_wrapper_sync and create_wrapper_async to build weave.op-decorated wrappers, and openai_accumulator to merge streaming ChatCompletionChunk objects.

This is a Wrapper Doc: it documents how Weave uses the external OpenAI SDK, not an API defined by Weave itself.

Usage

This patcher is invoked automatically by patch_openai() or implicit_patch(). No direct user interaction is needed.

Code Reference

Source Location

Repository: wandb/weave
File: weave/integrations/openai/openai_sdk.py
Lines: L732-896 (get_openai_patcher)
Lines: L389-476 (create_wrapper_sync/async)
Lines: L158-315 (openai_accumulator)

External Reference

OpenAI API Reference

Signature

def get_openai_patcher(
    settings: IntegrationSettings | None = None,
) -> MultiPatcher | NoOpPatcher:
    """Build a MultiPatcher for all OpenAI SDK methods.

    Patches 14 methods (sync + async) including:
    - chat.completions.create
    - chat.completions.parse
    - moderations.create
    - embeddings.create
    - responses.create
    - responses.parse
    """

Import

from weave.integrations.openai.openai_sdk import get_openai_patcher

I/O Contract

Inputs

Name	Type	Required	Description
settings	None	No	Configuration for the OpenAI integration

Outputs

Name	Type	Description
return	NoOpPatcher	Patcher containing 14 SymbolPatcher instances

Usage Examples

Transparent Tracing

import openai
import weave

weave.init("my-team/my-project")

# Normal OpenAI call - automatically traced
client = openai.OpenAI()
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}],
)
# The call, inputs, outputs, and token usage are all captured in Weave

Streaming

import openai
import weave

weave.init("my-team/my-project")

client = openai.OpenAI()
stream = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Tell me a story"}],
    stream=True,
)
# Streaming chunks are accumulated into a complete response
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Related Pages

Implements Principle

Principle:Wandb_Weave_LLM_Call_Interception

Requires Environment

Uses Heuristic

Heuristic:Wandb_Weave_Sentinel_Value_Handling

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment