Connecting AoAi Chat completion to internet search

Hi Mohamed Hussein,

Thanks for posting your question on Microsoft Q&A, In addition to the response of AshokPeddakotla-MSFT, Please check if this is helpful aswell.

Yes , it’s possible to connect the Chat Completions API with Bing Search without using the Assistants module with out of box ways. This can be achieved by programmatically integrating the two services. Here’s how you can do it:

Steps to Connect Chat Completions API with Bing Search

1. Set Up Bing Search API

Obtain an API key for the Azure Bing Search API.

• Use the API endpoint to fetch search results based on the user's query.

Leverage the Chat Completions API

• Use the Chat Completions API to handle user input and process it as a query.
  • Determine if the input requires external search results, like fetching live information.

Programmatic Integration

• Fetch search results from Bing Search API.
  • Feed these results into the Chat Completions API to form a natural and structured response.
  • This allows Bing's data to enhance the Chat Completion's conversational flow.

Example Use Case

• User Query: "What’s the latest update on AI technology?"
  • Workflow:
  • Chat Completions API processes the query and identifies it as needing a web search.
  • Bing Search API retrieves relevant articles or news.
• Chat Completions API integrates the results into its response for the user.

Example code:

import openai
import requests

# API Keys and Endpoints
OPENAI_API_KEY = "your-openai-key"
BING_API_KEY = "your-bing-key"
BING_ENDPOINT = "https://api.bing.microsoft.com/v7.0/search"

# User Input
user_query = "What’s the latest update on AI technology?"

# Bing Search API Function
def fetch_bing_results(query):
    headers = {"Ocp-Apim-Subscription-Key": BING_API_KEY}
    params = {"q": query, "count": 3}
    response = requests.get(BING_ENDPOINT, headers=headers, params=params)
    return response.json().get("webPages", {}).get("value", [])

# Get Bing Results
search_results = fetch_bing_results(user_query)
formatted_results = "\n".join([f"{res['name']}: {res['url']}" for res in search_results])

# Chat Completions API
openai.api_key = OPENAI_API_KEY
response = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": user_query},
        {"role": "assistant", "content": f"Here are the latest updates:\n{formatted_results}"}
    ]
)

# Display Response
print(response["choices"][0]["message"]["content"])

Consider the below as combining Bing Search and the Chat Completions API introduces additional processing time, as the application needs to handle two separate API calls—one for retrieving search results and another for generating responses. To ensure meaningful outputs, post-process Bing Search results to align them with user intent, using filtering. Additionally, both APIs operate on a usage-based billing model, with costs incurred per query for Bing Search and token usage for Chat Completions. Monitoring usage and optimizing API calls can help manage latency and minimize operational expenses effectively.

If this resolves your query, please mark the answer as accepted. If you have further questions, feel free to ask, I'd be happy to help.