Azure OpenAI Proxy (Azure Functions)

This repo contains a lightweight Azure Function that exposes an OpenAI-compatible API (/v1/chat/completions and /v1/responses) backed by Azure OpenAI, using:

• Azure Functions as a proxy

• Managed Identity for Azure authentication (no API keys in code)

• Function / Host keys as a shared access token

• Compatibility with:

  • OpenAI Python SDK
  • DSPy (via chat completions)

This setup is ideal for university / research tenants where Entra ID user management is restricted.

---

Quickstart

See Quickstart.ipynb

---

Architecture (high level)

Client (OpenAI SDK / DSPy)
  └─ x-functions-key (shared token)
       ↓
Azure Function (this repo)
  └─ Managed Identity (RBAC)
       ↓
Azure OpenAI

---

Public Endpoints

Once deployed, the Function exposes:

• POST /api/v1/chat/completions
• POST /api/v1/responses

Base URL format:

https://<function-app>.azurewebsites.net/api/v1

---

Repository Structure

.
├─ function_app.py     # Azure Function (HTTP proxy)
├─ requirements.txt    # Python dependencies
├─ host.json           # Functions host config
└─ README.md

---

Required Azure Configuration

1. Enable Managed Identity

On the Function App:

• Settings → Identity → System-assigned = On

2. Grant RBAC

On the Azure OpenAI resource:

• Assign the Function’s managed identity the role:

  • Cognitive Services OpenAI User
  • (or Cognitive Services User if required by tenant)

3. Application Settings (Environment Variables)

Set these in Function App → Configuration → Application settings:

AZURE_OPENAI_BASE=https://<your-openai-resource>.cognitiveservices.azure.com
AZURE_OPENAI_DEPLOYMENT=gpt-4o-mini
AZURE_OPENAI_API_VERSION=2024-05-01-preview

Restart the Function App after saving.

---

Authentication (Shared Access)

Access is controlled via Azure Functions keys.

• Use a Host key (recommended) to authorize all endpoints
• Send it as an HTTP header:

x-functions-key: <HOST_KEY>

⚠️ Do not share the _master key.

---

Local Development

Prerequisites

• Python 3.11+
• Azure CLI (az login)
• Azure Functions Core Tools (optional)

Run locally

pip install -r requirements.txt
func start

Local settings can be placed in local.settings.json (not committed).

---

Example: OpenAI SDK (Responses API)

from openai import OpenAI
import os

client = OpenAI(
    base_url="https://<function-app>.azurewebsites.net/api/v1",
    api_key="unused",
    default_headers={"x-functions-key": os.environ["FUNCTION_HOST_KEY"]},
)

resp = client.responses.create(
    model="gpt-4o-mini",
    input="Tell me a one line story.",
    temperature=0.7,
)

print(resp.output_text)

---

Example: DSPy

import dspy
import os

lm = dspy.LM(
    "openai/gpt-4o-mini",
    api_base="https://<function-app>.azurewebsites.net/api/v1",
    api_key="unused",
    headers={"x-functions-key": os.environ["FUNCTION_HOST_KEY"]},
)

dspy.configure(lm=lm)

pred = dspy.Predict("question -> answer")(question="Tell me a one line story.")
print(pred.answer)

---

Notes & Limitations

• This proxy is pass-through: all OpenAI parameters (e.g. temperature) are forwarded as-is.
• Streaming requires additional handling (not included here).
• For rate limiting, analytics, or multiple keys, consider placing this Function behind Azure API Management (APIM).