# OpenAI Provider

OPENAI_API_KEY='<add your OpenAI API key here>'

## Azure OpenAI in Microsoft Foundry

## (Azure OpenAI Service is now part of Microsoft Foundry: https://ai.azure.com?WT.mc_id=academic-105485-koreyst.

## Create/manage your resource and deployments from the Foundry portal - the env var names below are unchanged.)

AZURE_OPENAI_API_VERSION='2024-10-21' # Default is set! (current stable GA API version)

AZURE_OPENAI_API_KEY='<add your Foundry resource key here>'

AZURE_OPENAI_ENDPOINT='<add your Foundry resource endpoint here, e.g. https://<resource-name>.openai.azure.com>'

AZURE_OPENAI_DEPLOYMENT='<add your chat completion model deployment name here, e.g. gpt-5-mini>'

## Note: gpt-5-mini is a *reasoning* model - it does NOT support temperature/top_p, and uses

## max_output_tokens instead of max_tokens. To experiment with temperature, use the Llama model below.

AZURE_OPENAI_EMBEDDINGS_DEPLOYMENT='<add your embeddings model deployment name here, e.g. text-embedding-3-small>'

## Microsoft Foundry Models

## (Multi-provider model catalog - one endpoint/key for OpenAI, Meta, Mistral, Cohere, Microsoft, and more.

## Replaces GitHub Models, which retires end of July 2026. Get these from your Foundry project's "Overview" page.)

AZURE_INFERENCE_ENDPOINT='<add your Microsoft Foundry project endpoint here, e.g. https://<resource-name>.services.ai.azure.com/models>'

AZURE_INFERENCE_CREDENTIAL='<add your Microsoft Foundry Models API key here>'

## A non-reasoning model (e.g. a Llama model) that DOES support temperature/top_p - used by the

## temperature examples. Deploy one from the Foundry model catalog and put its deployment name here.

AZURE_INFERENCE_CHAT_MODEL='<add a temperature-capable model deployment name here, e.g. Llama-3.3-70B-Instruct>'

## Hugging Face

HUGGING_FACE_API_KEY='<add your HuggingFace API or token here>'