Python: Bug: Python: AzureRealtimeWebsocket works only with hardcoded api_key in the script.

[mkcod]

Describe the bug
A clear and concise description of what the bug is.

Looks like websocket connection is not using the same authentication method/variables same as azure ai settings. When I pass the value of api_key as hardcoded value in the script, then AzureRealtimeWebsocket works file. else, if i pass the api_key using environment variables, it is error out with websocket connection reject.

To Reproduce
Steps to reproduce the behavior:

1. Go to '...'
2. https://github.com/microsoft/semantic-kernel/blob/main/python/samples/concepts/realtime/realtime_agent_with_function_calling_websocket.py
3. Click on '....'
4. Find this code: realtime_agent = AzureRealtimeWebsocket()
5. Replace with: realtime_agent = AzureRealtimeWebsocket(
   api_version="2024-10-01-preview",
   api_key=key
   )
   where key is defined as below:
   load_dotenv()
   key=os.getenv("AZURE_OPENAI_API_KEY")

This gives the below error:
websockets.exceptions.InvalidStatus: server rejected WebSocket connection: HTTP 401

and as per logs:
DEBUG:websockets.client:> api-key: 9dxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx3j

where as the right key is:
5MxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxFk

when i pass the above key like below:

realtime_agent = AzureRealtimeWebsocket(
api_version="2024-10-01-preview",
api_key="5MxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxFk"
)

everything works fine,

this is not as per the api_key best practices as it exposes the api_key in the code.

6. Scroll down to '....'
7. See errorExpected behaviorA clear and concise description of what you expected to happen.

Screenshots
If applicable, add screenshots to help explain your problem.

Platform

• Language: [e.g. C#, Python] Python 3.12, Semantic Kernel 1.28.1
• Source: [e.g. NuGet package version 0.1.0, pip package version 0.1.0, main branch of repository] https://github.com/microsoft/semantic-kernel/blob/main/python/samples/concepts/realtime/realtime_agent_with_function_calling_websocket.py
• AI model: [e.g. OpenAI:GPT-4o-mini(2024-07-18)] gpt-4o-realtime-preview
• IDE: [e.g. Visual Studio, VS Code] vs code
• OS: [e.g. Windows, Mac] Windows

Additional context
Add any other context about the problem here.
Covered above everything.

[moonbox3]

Hi @mkcod, I am having a hard time repro'ing this issue. Do you have multiple env vars/.env files configured with different API keys? I have one .env file that VSCode turns into env vars on my behalf given that it's located at the root of my project. When we don't pass in any override kwargs to the AzureRealtimeWebsocket constructor, we're allowing Pydantic settings to grab the env vars/.env config as part of object creation.

The fact that it's finding a key, and that it is incorrect, could mean you have multiple keys defined? Per load_dotenv(): it looks for the first .env file by calling find_dotenv, starting in the current working directory and walking up the parent directories until it finds one. Only the first file found is read. If you have several .env files, only the upper‑most one is loaded unless you call load_dotenv again with an explicit path.

I'd test with something like:

import os, dotenv
dotenv.load_dotenv(override=True)
assert os.environ["AZURE_OPENAI_API_KEY"].startswith("5M"), "Wrong key in env"

[mkcod]

@moonbox3 - you were right, it was picking up wrong API key from the parent directory. apologies for the confusion.

[moonbox3]

@mkcod not a problem. Glad you got it figured out.