Skip to content

AZURE_README.md

Latest commit

 

History

History
153 lines (106 loc) · 4.49 KB

AZURE_README.md

File metadata and controls

153 lines (106 loc) · 4.49 KB

Azure OpenAI Integration for Presenton

This repository contains Check Point's fork of Presenton with Azure OpenAI support.

🚀 What's New

This fork adds enterprise-grade Azure OpenAI integration to Presenton:

✅ Azure OpenAI Native Support - Works with Azure OpenAI Legacy API endpoints
✅ GPT-5 Reasoning Models - Advanced custom template creation with GPT-5
✅ Dual Model Strategy - Fast gpt-4o-mini for basic, GPT-5 for advanced features
✅ Production Ready - Fully tested with Check Point Azure infrastructure

📋 Quick Start

Prerequisites

  • Docker
  • Azure OpenAI endpoint access
  • (Optional) Pexels API key for images

Run with Azure OpenAI

docker build -t presenton-azure:latest .

docker run -d --name presenton -p 5000:80 \
  -e LLM="custom" \
  -e CUSTOM_LLM_URL="https://your-azure-endpoint.azure-api.net/path" \
  -e CUSTOM_LLM_API_KEY="your-azure-api-key" \
  -e CUSTOM_MODEL="gpt-4o-mini-2024-07-18" \
  -e OPENAI_API_KEY="your-azure-api-key" \
  -e GPT5_MODEL="gpt-5-2025-08-07" \
  -e IMAGE_PROVIDER="pexels" \
  -e PEXELS_API_KEY="your-pexels-key" \
  -e GOOGLE_API_KEY="optional" \
  -v "./app_data:/app_data" \
  presenton-azure:latest

Access at: http://localhost:5000

⚙️ Configuration

Required Environment Variables

Variable Description Example
CUSTOM_LLM_URL Azure endpoint URL https://your-endpoint.azure-api.net/path
CUSTOM_LLM_API_KEY Azure API key your-api-key
CUSTOM_MODEL Fast model for basic presentations gpt-4o-mini-2024-07-18
OPENAI_API_KEY Same as CUSTOM_LLM_API_KEY (for frontend) your-api-key

Optional Environment Variables

Variable Description Default
GPT5_MODEL Model for custom templates gpt-5-2025-08-07
IMAGE_PROVIDER Image service pexels
PEXELS_API_KEY Pexels API key -
GOOGLE_API_KEY Google API key -

Model Strategy

  • CUSTOM_MODEL (gpt-4o-mini): Fast basic presentations (~5-10 seconds)
  • GPT5_MODEL (gpt-5): Advanced custom templates (~30-60 seconds per slide)

🔧 Azure OpenAI Modifications

Technical Changes

  1. Azure Legacy API Client (servers/fastapi/services/llm_client.py)

    • Auto-detects Azure URLs
    • Uses AsyncAzureOpenAI client
    • Handles Azure authentication format

  2. Model Validation Skip (servers/fastapi/utils/)

    • Graceful handling when Azure endpoint lacks /models endpoint
    • Proceeds with configured model without validation

  3. GPT-5 Support (servers/fastapi/api/v1/ppt/endpoints/slide_to_html.py)

    • Converted Responses API → Chat Completions API
    • Azure-compatible parameters:
      • max_completion_tokens instead of max_tokens
      • No temperature parameter (GPT-5 default only)
      • 16000 token limit for reasoning
    • Separate GPT5_MODEL configuration

  4. Docker Dependencies (Dockerfile)

    • Added zstd for Ollama

📚 Usage

Basic Presentations (Fast)

  1. Navigate to http://localhost:5000
  2. Enter topic and click "Generate"
  3. Uses gpt-4o-mini for speed

Custom Templates (Advanced)

  1. Go to http://localhost:5000/custom-template
  2. Upload .pptx file
  3. Click "Process File"
  4. Uses GPT-5 reasoning for quality
  5. Expect 30-60s per slide (reasoning tokens)

API Documentation

FastAPI docs: http://localhost:5000/docs

🔍 Troubleshooting

Authentication errors (401)

  • Verify CUSTOM_LLM_API_KEY is correct
  • Check API key has model access

Slow custom templates

  • Expected: GPT-5 reasoning takes time
  • Normal: 30-60 seconds per slide

Empty GPT-5 responses

  • Fork includes fix: 16000 token limit
  • GPT-5 needs tokens for reasoning + content

🌳 Branch Structure

  • main - Sync with upstream Presenton
  • azure-openai-legacy-api - Azure integration (active)

📝 Key Commits

  • fc13b2e - Initial Azure OpenAI support
  • 2d230e7 - GPT-5 Chat Completions API

🔗 Links

📄 License

Apache License 2.0 (same as upstream)

🤝 Support

Note: No sensitive data (API keys, endpoints) are committed to this repository. All configuration is via environment variables.