Support / Document Azure OpenAI via OpenAI-compatible /openai/v1 endpoint

[Aranastou]

Summary

Strix works correctly with Azure OpenAI when using Azure’s OpenAI-compatible API (/openai/v1), but this is currently undocumented and non-obvious for users.

This issue proposes documenting the correct configuration so users can run Strix against Azure OpenAI without any code changes.

---

Background

Strix uses LiteLLM and communicates with LLMs using the OpenAI-compatible API.

Azure OpenAI exposes a fully compatible endpoint at: https://<RESOURCE-NAME>.cognitiveservices.azure.com/openai/v1/

When this endpoint is used as LLM_API_BASE, Strix works out of the box.

However, many Azure users attempt to use legacy Azure endpoints such as:

/openai/deployments//chat/completions?api-version=...

These legacy endpoints are not OpenAI-compatible and will not work with Strix.

Verified Working Configuration

No Strix code changes are required.

Environment Variables

# Model / deployment name
export STRIX_LLM="openai/gpt-4o"
# or: openai/<azure-deployment-name>

# Azure OpenAI API key
export LLM_API_KEY="AZURE_OPENAI_API_KEY"

# Azure OpenAI OpenAI-compatible base URL
export LLM_API_BASE="https://<RESOURCE-NAME>.cognitiveservices.azure.com/openai/v1/"

Important notes:
STRIX_LLM must match the Azure deployment name
LLM_API_BASE must include /openai/v1/
No api-version query parameter is required