Grounding with Bing Search

Grounding with Bing Search allows your Azure AI Agents to incorporate real-time public web data when generating responses. You need to create a Grounding with Bing Search resource, and then connect this resource to your Azure AI Agents. When a user sends a query, Azure AI Agents decide if Grounding with Bing Search should be leveraged or not. If so, it will leverage Bing to search over public web data and return relevant chunks. Lastly, Azure AI Agents will use returned chunks to generate a response.

You can ask questions such as "what is the top news today" or "what is the recent update in the retail industry in the US?", which require real-time public data.

Developers and end users don't have access to raw content returned from Grounding with Bing Search. The model response, however, includes citations with links to the websites used to generate the response, and a link to the Bing query used for the search. You can retrieve the model response by accessing the data in the thread that was created. These two references must be retained and displayed in the exact form provided by Microsoft, as per Grounding with Bing Search's Use and Display Requirements. See the how to display Grounding with Bing Search results section for details.

How Grounding with Bing Search works

The user query is the message that an end user sends to an agent, such as "should I take an umbrella with me today? I'm in Seattle." Instructions are the system message a developer can provide to share context and provide instructions to the AI model on how to use various tools or behave.

When a user sends a query, the customer's AI model deployment first processes it (using the provided instructions) to later perform a Bing search query (which is visible to developers). Grounding with Bing returns relevant search results to the customer's model deployment, which then generates the final output.

The authorization will happen between Grounding with Bing Search service and Azure AI Foundry Agent Service. Any Bing search query that is generated and sent to Bing for the purposes of grounding is transferred, along with the resource key, outside of the Azure compliance boundary to the Grounding with Bing Search service. Grounding with Bing Search is subject to Bing's terms and do not have the same compliance standards and certifications as the Azure AI Foundry Agent Service, as described in the Grounding with Bing Search Terms of Use. It is your responsibility to assess whether the use of Grounding with Bing Search in your agent meets your needs and requirements.

Supported capabilities and known issues

• Grounding with Bing Search tool is designed to retrieve real-time information from web, NOT specific web domains.
• NOT Recommended to summarize an entire web page.
• Within one run, the AI model will evaluate the tool outputs and may decide to invoke the tool again for more information and context. AI model may also decide which piece(s) of tool outputs are used to generate the response.
• Azure AI Agent service will return AI model generated response as output so end-to-end latency will be impacted pre-/post-processing of LLMs.
• Grounding with Bing Search tool does NOT return the tool output to developers and end users.

Usage support

Setup

1. Create an Azure AI Agent by following the steps in the quickstart.

2. Create a Grounding with Bing Search resource. You need to have owner or contributor role in your subscription or resource group to create it.

   1. You can create one in the Azure portal, and select the different fields in the creation form. Make sure you create this Grounding with Bing Search resource in the same resource group as your Azure AI Agent, AI Project, and other resources.

   

   1. You can also create one through code-first experience. If so, you need to manually register Bing Search as an Azure resource provider. You must have permission to perform the /register/action operation for the resource provider. The permission is included in the Contributor and Owner roles.

       az provider register --namespace 'Microsoft.Bing'
   

3. After you have created a Grounding with Bing Search resource, you can find it in Azure portal. Navigate to the resource group you've created the resource in, search for the Grounding with Bing Search resource you have created.

   

Optional parameters

When you add the Grounding with Bing Search tool to your agent, you can pass the following parameters. These parameters will impact the Grounding with Bing Search tool output, and the AI model might not fully use all of the outputs. See the code examples for information on API version support and how to pass these parameters.

How to display Grounding with Bing Search results

According to Grounding with Bing's terms of use and use and display requirements, you need to display both website URLs and Bing search query URLs in your custom interface. You can find website URLs through annotations parameter in API response and Bing search query URLs through runstep details. To render the webpage, we recommend you replace the endpoint of Bing search query URLs with www.bing.com and your Bing search query URL would look like "https://www.bing.com/search?q={search query}"

run_steps = project_client.agents.list_run_steps(run_id=run.id, thread_id=thread.id)
run_steps_data = run_steps['data']
print(f"Last run step detail: {run_steps_data}")

Next steps

See code samples for using the Grounding with Bing tool programmatically.