What additional advantages does Teams AI offer over Bot Framework if we don’t use the AI capabilities?

Question

What additional advantages does Teams AI offer over Bot Framework if we don’t use the AI capabilities?

Shravan Das | CM.com 0

I am exploring migrating from Bot Framework to Teams AI without using the AI layer. The documentation mentions that user authentication is integrated into the Teams AI library, making setup easier. Could anyone clarify which user authentication mechanism this refers to? https://learn.microsoft.com/en-us/microsoftteams/platform/bots/how-to/teams-conversational-ai/how-conversation-ai-get-started?tabs=dotnet1%2Cdotnet2%2Cdotnet3%2Cdotnet4#migrate-your-bot-to-use-teams-ai-library

Additionally, if we choose not to use Open AI (Ai layer) in the Teams AI library and migrate to it from bot framework, what benefits do we still get? Is this approach feasible, and is it recommended?

Based on this documentation https://github.com/microsoft/teams-ai/blob/b34bbd14e9d13aed140686e4f91dbb673982b1cf/getting-started/MIGRATION/02.DOTNET.md I noticed that we migrate from ActivityHandler to the Application object. Why is this change necessary? Does the Teams AI library no longer support ActivityHandler (class which IBot inherits and has OnMessageActivityAsync, onMemberadded etc)?

Looking forward to some insights!

Prasad-MSFT 8,846 Reputation points Microsoft External Staff

2025-04-02T04:59:23.3633333+00:00

Teams AI library simplifies user authentication by integrating directly with Microsoft Teams' authentication mechanisms. This includes support for Single Sign-On (SSO) with Azure Active Directory (AAD). The authentication mechanism referred to in the documentation is likely the Teams SSO, which allows users to authenticate seamlessly without requiring additional login prompts. This reduces friction for end users and simplifies the setup for developers.

The Bot Framework SDK, including ActivityHandler, is designed to support multiple channels (e.g., Slack, Facebook Messenger, etc.) and provides a generalized approach to handling activities. The Teams AI library, on the other hand, is specifically tailored for Microsoft Teams. By introducing the Application object, the library focuses on Teams-specific scenarios, such as handling tabs, messaging extensions, adaptive cards, and meeting lifecycle events, without the overhead of supporting other channels.
Shravan Das | CM.com 0 Reputation points

2025-04-02T08:52:42.98+00:00

Hey Prasad thanks for clarifying the doubt,
there's more to ask perhaps,

Actually, instead of integrating an LLM like OpenAI in the AI layer, we want to have our own agentic AI.

Instead of using the OpenAI model, can we configure an endpoint that triggers and retrieves responses from our agentic AI in real time? We want to leverage the Teams Gen AI library while removing the OpenAI model.

I'll give you a brief overview of the current flow:
In the bot configuration, we set up an endpoint for messaging(as shown in screenshot).

This endpoint will call our agentic AI and return responses accordingly,
so by replacing OpenAI's role in handling the prompts, answer and handling the actions accordingly, can this endpoint/agentic ai model be configured instead of this open ai layer?

The goal is to leverage the power of prompts while ensuring that our agentic AI is responsible for understanding the context and fetching responses from our Agentic AI model.

Additionally, could we incorporate actions as well? (One doubt.)
Prasad-MSFT 8,846 Reputation points Microsoft External Staff

2025-04-02T13:50:57.2933333+00:00

Yes, you can replace the OpenAI layer in the Teams AI library with your own agentic AI by configuring a custom endpoint. This approach allows you to leverage the Teams AI library's features (e.g., context, actions, prompts) while ensuring that your agentic AI handles the actual response generation.

This setup is feasible and recommended if you want to maintain control over the AI model while benefiting from the Teams AI library's capabilities.
Shravan Das | CM.com 0 Reputation points

2025-04-03T07:35:29.23+00:00
Just to be more specific about my previous message:

We have an endpoint configured that connects to our fine-tuned model. My question is, instead of adding OpenAI credentials to configure the model, can we use this endpoint directly?

Here is the code snippet for reference for configuring the open ai model in setup:
builder.Services.AddSingleton<OpenAIModel>(sp => new(

new AzureOpenAIModelOptions( config.Azure.OpenAIApiKey, "gpt-35-turbo", config.Azure.OpenAIEndpoint ) { LogRequests = true }, sp.GetService<ILoggerFactory>()

));

We have an endpoint suppose (/v1/message) that interacts with our fine-tuned model to generate responses.

My question is: Is it possible to use this endpoint configuration instead of OpenAI credentials(want to completely remove its credentials) to leverage the endpoint and get the response accordingly?

Additionally, is there any public documentation available that explains how to set up and configure this endpoint? I couldn't find a solution in the official Teams AI documentation. If you have any relevant documentation or resources that I can follow, that would be really helpful.

Looking forward to your response!
Prasad-MSFT 8,846 Reputation points Microsoft External Staff

2025-04-04T05:02:30.93+00:00

Unfortunately, there is no official Teams AI documentation specifically for integrating custom AI endpoints.
Shravan Das | CM.com 0 Reputation points

2025-04-04T07:07:24.97+00:00

Hey thanks for responding back,

so just for confirming, assuming that the use case described above is not supported? since there isn't no documentation to configure the custom endpoint instead of open ai
maybe in future ms teams ai library can support this usecase :)

Your answer

Prasad-MSFT 8,846 Reputation points Microsoft External Staff

2025-04-02T04:59:23.3633333+00:00

Teams AI library simplifies user authentication by integrating directly with Microsoft Teams' authentication mechanisms. This includes support for Single Sign-On (SSO) with Azure Active Directory (AAD). The authentication mechanism referred to in the documentation is likely the Teams SSO, which allows users to authenticate seamlessly without requiring additional login prompts. This reduces friction for end users and simplifies the setup for developers.

The Bot Framework SDK, including ActivityHandler, is designed to support multiple channels (e.g., Slack, Facebook Messenger, etc.) and provides a generalized approach to handling activities. The Teams AI library, on the other hand, is specifically tailored for Microsoft Teams. By introducing the Application object, the library focuses on Teams-specific scenarios, such as handling tabs, messaging extensions, adaptive cards, and meeting lifecycle events, without the overhead of supporting other channels.
Shravan Das | CM.com 0 Reputation points

2025-04-02T08:52:42.98+00:00

Hey Prasad thanks for clarifying the doubt,
there's more to ask perhaps,

Actually, instead of integrating an LLM like OpenAI in the AI layer, we want to have our own agentic AI.

Instead of using the OpenAI model, can we configure an endpoint that triggers and retrieves responses from our agentic AI in real time? We want to leverage the Teams Gen AI library while removing the OpenAI model.

I'll give you a brief overview of the current flow:
In the bot configuration, we set up an endpoint for messaging(as shown in screenshot).

This endpoint will call our agentic AI and return responses accordingly,
so by replacing OpenAI's role in handling the prompts, answer and handling the actions accordingly, can this endpoint/agentic ai model be configured instead of this open ai layer?

The goal is to leverage the power of prompts while ensuring that our agentic AI is responsible for understanding the context and fetching responses from our Agentic AI model.

Additionally, could we incorporate actions as well? (One doubt.)
Prasad-MSFT 8,846 Reputation points Microsoft External Staff

2025-04-02T13:50:57.2933333+00:00

Yes, you can replace the OpenAI layer in the Teams AI library with your own agentic AI by configuring a custom endpoint. This approach allows you to leverage the Teams AI library's features (e.g., context, actions, prompts) while ensuring that your agentic AI handles the actual response generation.

This setup is feasible and recommended if you want to maintain control over the AI model while benefiting from the Teams AI library's capabilities.
Shravan Das | CM.com 0 Reputation points

2025-04-03T07:35:29.23+00:00

Just to be more specific about my previous message:

We have an endpoint configured that connects to our fine-tuned model. My question is, instead of adding OpenAI credentials to configure the model, can we use this endpoint directly?

Here is the code snippet for reference for configuring the open ai model in setup:
builder.Services.AddSingleton<OpenAIModel>(sp => new(

new AzureOpenAIModelOptions( config.Azure.OpenAIApiKey, "gpt-35-turbo", config.Azure.OpenAIEndpoint ) { LogRequests = true }, sp.GetService<ILoggerFactory>()

));

We have an endpoint suppose (/v1/message) that interacts with our fine-tuned model to generate responses.

My question is: Is it possible to use this endpoint configuration instead of OpenAI credentials(want to completely remove its credentials) to leverage the endpoint and get the response accordingly?

Additionally, is there any public documentation available that explains how to set up and configure this endpoint? I couldn't find a solution in the official Teams AI documentation. If you have any relevant documentation or resources that I can follow, that would be really helpful.

Looking forward to your response!
Prasad-MSFT 8,846 Reputation points Microsoft External Staff

2025-04-04T05:02:30.93+00:00

Unfortunately, there is no official Teams AI documentation specifically for integrating custom AI endpoints.
Shravan Das | CM.com 0 Reputation points

2025-04-04T07:07:24.97+00:00

Hey thanks for responding back,

so just for confirming, assuming that the use case described above is not supported? since there isn't no documentation to configure the custom endpoint instead of open ai
maybe in future ms teams ai library can support this usecase :)

Share via

What additional advantages does Teams AI offer over Bot Framework if we don’t use the AI capabilities?

Your answer