Bing Search Agent (Azure AI Projects + Bing Grounding)

This repository contains a minimal Python example showing how to use the Azure AI Projects SDK with a Bing grounding tool to build an AI agent that can answer questions using live web search (via a Bing connection).

The example script, agent.py, creates:

• Azure AI agent: Powered by a deployed model in your Azure AI Project.
• Bing grounding tool: Configured via an Azure AI connection to Bing.
• Conversation thread: Sends a user message like “what's the weather in London now?”.
• Agent run: Executes the agent with the Bing tool and returns a grounded answer.
• Console output:
  • Assistant response: Textual answer from the agent.
  • URL citations: Source links returned by Bing used to ground the answer.

---

Prerequisites

• Python: 3.9+ (recommended)
• Azure:
  • Azure subscription
  • Azure AI Project set up
  • A deployed model in your Azure AI Project (e.g., GPT-style deployment)
  • A Bing connection configured in the same project
• Credentials:
  • Access to your Azure subscription (e.g., via az login or environment-based credentials)

---

Installation

1. Clone this repo

git clone <your-repo-url>.git
cd Bing_Search_Agent

2. Create and activate a virtual environment (recommended)

python -m venv .venv
.\.venv\Scripts\activate  # On Windows
# source .venv/bin/activate  # On macOS / Linux

3. Install dependencies

pip install -r requirements.txt

---

Configuration

This project uses python-dotenv to load environment variables from a .env file.

Create a .env file in the project root with:

PROJECT_CONNECTION_STRING="your-azure-ai-project-connection-string"
MODEL_DEPLOYMENT_NAME="your-model-deployment-name"
BING_CONNECTION_NAME="your-bing-connection-name"

• PROJECT_CONNECTION_STRING: Found in your Azure AI Project (e.g., in Azure AI Studio).
• MODEL_DEPLOYMENT_NAME: The name of the model deployment you want to use.
• BING_CONNECTION_NAME: The name of the Bing connection you created in the project.

The script uses DefaultAzureCredential, so ensure your local environment is authenticated with Azure (for example with az login, managed identity, or environment variables).

---

How It Works

agent.py performs the following steps:

1. Load configuration from environment variables using dotenv.
2. Create an AIProjectClient with DefaultAzureCredential and PROJECT_CONNECTION_STRING.
3. Look up the Bing connection by BING_CONNECTION_NAME and get its id.
4. Create a BingGroundingTool using that connection ID.
5. Create an agent with:
   • model from MODEL_DEPLOYMENT_NAME
   • Name: bing-assistant
   • Instructions: "You are a helpful assistant"
   • tools: bing.definitions
   • Headers: {"x-ms-enable-preview": "true"}
6. Create a thread and send a message with user content:
   • Example: "whats the weather in London now?"
7. Create and process a run for that thread and agent.
8. Log results to the console:
   • Agent and thread IDs
   • Run status (and error if any)
   • The assistant’s text answer
   • The URL citations extracted from the message annotations.

---

Usage

Run the script:

python agent.py

You should see:

• Agent ID and Thread ID printed
• Run status (e.g., completed)
• The assistant’s answer to your question
• A list of URL citations used as sources

---

Customization

• Change the user question
  Edit the content in the create_message call within agent.py:

  content="whats the weather in London now?"

  Replace this with any query where web search is useful.

• Change agent behavior
  Modify the instructions parameter in create_agent to adjust the agent’s tone, role, or behavior.

• Extend tools
  If your Azure AI Project has more tools configured, you can add them to the tools list when creating the agent.

---

Notes

• This repository is a minimal sample for learning how to:
  • Wire Azure AI Projects with a Bing grounding tool.
  • Create simple agents and threads.
  • Use citations from web search results.
• For production scenarios, consider:
  • Structured logging and error handling
  • Persisting agents/threads/messages
  • Adding an API or UI layer on top of this script.

---

License

Specify your preferred license here (for example, MIT or Apache-2.0).