Unable to configure Azure OpenAI custom endpoint in Stakpak due to missing API version option; requests fail with "error decoding response body

[SURAJTHEGREAT]

Description: When configuring a custom OpenAI-compatible endpoint hosted in Azure with Stakpak, there is no option to specify the required API version. Referencing the documentation https://stakpak.gitbook.io/docs/get-started/configure-stakpak#stakpak-open-source , I attempted to append the version as a query parameter to the endpoint URL, but requests fail. Without a placeholder for API version in the configuration flow, it’s unclear how to correctly set up Azure OpenAI endpoints.

Attempted endpoint: https://rsg-dummy.openai.azure.com?api-version=2024-12-01-preview/
Outcome: Requests fail with [Error] Bad Request: ApiError("error decoding response body")

Request:

Please provide guidance on how to correctly configure an Azure OpenAI-compatible endpoint in Stakpak, specifically how to set the api_version required by Azure.
If Stakpak supports Azure OpenAI, is there a configuration key (e.g., profiles.default.openai.api_version or similar) or a different endpoint format expected?
If appending api-version in the URL is unsupported, can support be added, or can you document the correct approach?

Reproduction steps (Docker):

1. Run container: docker run -it --entrypoint stakpak ghcr.io/stakpak/agent:latest

2. Select: Use my own Model/API Key

3. Select: Bring your own model

4. Enter OpenAI-compatible API endpoint: https://rsg-dummy.openai.azure.com?api-version=2024-12-01-preview/
   Note: Endpoint masked as rsg-dummy

5. Enter API key: myownapikey

6. Set models:
   Smart model name: o3-mini
   Eco model name:o3-mini

7. Configuration preview printed by Stakpak

[profiles.default]
provider="local"
smart_model="o3-mini"
eco_model="o3-mini"

[profiles.default.openai]
api_endpoint="https://rsg-dummy.openai.azure.com?api-version=2024-12-01-preview"
api_key="***"

8. Make a request (e.g., “Dockerize my app”) → failure: [Error] Bad Request: ApiError("error decoding response body")

Observed behavior:

The endpoint and credentials work in the same environment when using the Azure OpenAI SDK with explicit api_version. This suggests the failure is due to how Stakpak constructs requests or handles the api-version parameter.

Working python code

import os
from openai import AzureOpenAI

endpoint = "https://rsg-dummy.openai.azure.com/"
model_name = "o3-mini"
deployment = "o3-mini"

subscription_key = "myownapikey"
api_version = "2024-12-01-preview"

client = AzureOpenAI(
    api_version=api_version,
    azure_endpoint=endpoint,
    api_key=subscription_key,
)

response = client.chat.completions.create(
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "I am going to Paris, what should I see?"},
    ],
    max_completion_tokens=100000,
    model=deployment,
)

print(response.choices[0].message.content)

[SURAJTHEGREAT]

Tagging @kajogo777 , @mstfash for visibility

[SURAJTHEGREAT]

Hi Team,
Any update on this ?