Azure OpenAI (tracing)

Trace calls made through the openai package’s AzureOpenAI client. The instrumentation differs by language:

• TypeScript uses the dedicated @respan/instrumentation-azure-openai instrumentor. It also supports the legacy @azure/openai OpenAIClient when that package is present.
• Python has no separate Azure package. The AzureOpenAI client ships inside the openai package, so the OpenAI instrumentation (respan-instrumentation-openai) captures Azure OpenAI calls.

This tracing setup uses your Azure OpenAI credentials directly. To route requests through Respan instead, use the Azure OpenAI gateway setup.

Setup

$
pip install respan-ai respan-instrumentation-openai openai

$
export AZURE_OPENAI_API_KEY="YOUR_AZURE_OPENAI_API_KEY"
$
export AZURE_OPENAI_ENDPOINT="https://YOUR_RESOURCE.openai.azure.com"
$
export OPENAI_API_VERSION="2024-10-21"
$
export RESPAN_API_KEY="YOUR_RESPAN_API_KEY"

1
import os
2
from openai import AzureOpenAI
3
from respan import Respan
4
from respan_instrumentation_openai import OpenAIInstrumentor
5
6
respan = Respan(instrumentations=[OpenAIInstrumentor()])
7
8
client = AzureOpenAI(
9
    api_key=os.environ["AZURE_OPENAI_API_KEY"],
10
    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
11
    api_version=os.environ["OPENAI_API_VERSION"],
12
    azure_deployment="gpt-4o-mini",
13
)
14
15
response = client.chat.completions.create(
16
    model="gpt-4o-mini",
17
    messages=[{"role": "user", "content": "Say hello in one sentence."}],
18
)
19
print(response.choices[0].message.content)

Configuration

The configuration, attributes, and captured-operations below describe the TypeScript AzureOpenAIInstrumentor. In Python, configuration lives on the OpenAI instrumentation — see the OpenAI SDK tracing page.

Attributes

With propagateAttributes

Override per-request attributes using a context scope.

1
import { AzureOpenAI } from "openai";
2
import * as OpenAI from "openai";
3
import { Respan } from "@respan/respan";
4
import { AzureOpenAIInstrumentor } from "@respan/instrumentation-azure-openai";
5
6
const respan = new Respan({
7
  instrumentations: [
8
    new AzureOpenAIInstrumentor({ openAIModule: OpenAI }),
9
  ],
10
});
11
await respan.initialize();
12
13
const client = new AzureOpenAI({
14
  apiKey: process.env.AZURE_OPENAI_API_KEY,
15
  endpoint: process.env.AZURE_OPENAI_ENDPOINT,
16
  apiVersion: process.env.OPENAI_API_VERSION,
17
  deployment: "gpt-4o-mini",
18
});
19
20
async function handleRequest(userId: string, question: string) {
21
  await respan.propagateAttributes(
22
    {
23
      customer_identifier: userId,
24
      thread_identifier: "conv_abc_123",
25
      metadata: { plan: "pro" },
26
    },
27
    async () => {
28
      const response = await client.chat.completions.create({
29
        model: "gpt-4o-mini",
30
        messages: [{ role: "user", content: question }],
31
      });
32
      console.log(response.choices[0]?.message?.content);
33
    }
34
  );
35
}

Captured operations

• AzureOpenAI.chat.completions.create()
• AzureOpenAI.completions.create()
• AzureOpenAI.embeddings.create()
• Streaming chat and text completion calls
• Legacy @azure/openai OpenAIClient chat, completion, embedding, and streaming methods