URL Fetching Chat Application with Semantic Kernel

This demo project integrates URL fetching capabilities into a chat application using Semantic Kernel.

Overview

In this demo, we use Semantic Kernel to integrate a custom Python plugin. This plugin extracts web content from URLs specified within the chat prompt.

Repository Contents

• static/ - Contains webclient HTML
• .env - Environment variable configurations
• .env.example - Example environment configuration file
• Dockerfile - Docker configuration
• api-key.txt - Placeholder for API key storage
• app.py - Main application script
• docker-compose.yml - Docker Compose configuration
• fetchurl.py - Script for fetching URLs
• requirements.txt - Python dependencies

Prerequisites

1. Azure account (step-by-step guide)
2. GPT4o deployed in Microsoft AI Studio (step-by-step guide steps:#1 to #3)
3. Basic Python knowledge

Instructions to Run the Demo

1. Clone the Repository:

git clone https://github.com/KostaMalsev/fetchurl-chat-semantic-kernel-demo
cd fetchurl-chat-semantic-kernel-demo

2. Configuration:

• Update your AZURE_OPENAI_ENDPOINT in docker-compose.yml
• Update the api-key.txt with your AZURE_OPENAI_API_KEY

3. Run the Application:

You can run the application using Docker or directly with Python.

Using Docker:

1. Install Docker Compose (step-by-step guide)
2. Run the following command:

docker-compose up

3. Test the demo in your browser at http://localhost:8000/static/webclient.html

Note: You can find the Docker image of the service on DockerHub: chatserver

Using Python directly:

python app.py

Key Components

Semantic Kernel Service (app.py):

• Uses semantic_kernel for model invocation with FetchPlugin
• Implements FastAPI endpoint at localhost:8000/demoprompt for handling chat prompts
• Uses fetchurl for web content retrieval

Custom FetchPlugin:

class FetchPlugin:
    @kernel_function(name="get_content_from_url", description="Get the content from url")
    def get_content_from_url(self, url: str) -> str:
        return fetch_text_content_from_url(url)

Main Flow:

1. On startup: Sets up kernel with AI service and FetchPlugin
2. On request: Invokes model via kernel with the FetchPlugin
3. Returns model response to web client

Limitations and Considerations

1. This demo uses a single plugin, but Semantic Kernel supports multiple custom plugins
2. The fetch method is limited and may not work on all websites
3. Be aware of AI API usage costs. Use Azure Cost Management for budgeting and monitoring

Resources

Semantic Kernel repo