a.teamsChefBot

Microsoft Teams Conversational Bot with AI: Teams Chef

This is a conversational streaming bot for Microsoft Teams that thinks it's a Chef to help you cook apps using the Teams AI Library. The bot uses the gpt-4o model to chat with Teams users and respond in a polite and respectful manner, staying within the scope of the conversation.

• Microsoft Teams Conversational Bot with AI: Teams Chef
  • Summary
  • Setting up the sample
  • Testing the sample
    • Using Teams Toolkit for Visual Studio Code
    • Using Teams App test tool

Summary

This sample illustrates how to use Retrieval Augmented Generation (RAG) to easily inject contextual relevant information into the prompt sent to the model. This results in better and more accurate replies from the bot.

The sample uses a local Vector Database, called Vectra, and Semantic Search to find the most relevant information to include in the prompt for the users input. The index can be found in ./index/teams-ai and includes all of the projects Getting Started docs and the source code for the Teams AI Library. This means you can ask the Teams Chef Bot anything about the library and it can answer it. You can even ask it to write sample code for you!

In addition, the sample illustrates our streaming feature. This is enabled via using our DefaultAugmentation class, and by setting stream: true in the OpenAIModel declaration. The informative message can be set in the ActionPlanner declaration via the startStreamingMessage config. Custom logic is available through the StreamingResponse class.

Setting up the sample

1. Clone the repository

git clone https://github.com/Microsoft/teams-ai.git

Important

To prevent issues when installing dependencies after cloning the repo, copy or move the sample directory to it's own location first. If you opened this sample from the Sample Gallery in Teams Toolkit, you can skip to step 2.

cd path/to/04.ai-apps/a.teamsChefBot
code .

2. Add your OpenAI key to the SECRET_OPENAI_KEY variable in the ./env/.env.local.user file and comment out lines 90-93 in ./src/index.ts file.

If you are using Azure OpenAI then follow these steps:

• Comment the SECRET_OPENAI_KEY variable in the ./env/.env.local.user file.
• Add your Azure OpenAI key and endpoint values to the SECRET_AZURE_OPENAI_KEY and SECRET_AZURE_OPENAI_ENDPOINT variables.
• Open the teamsapp.local.yml file and modify the last step to use Azure OpenAI variables instead:

- uses: file/createOrUpdateEnvironmentFile
    with:
      target: ./.env
      envs:
        BOT_ID: ${{BOT_ID}}
        BOT_PASSWORD: ${{SECRET_BOT_PASSWORD}}
        #OPENAI_KEY: ${{SECRET_OPENAI_KEY}}
        AZURE_OPENAI_KEY: ${{SECRET_AZURE_OPENAI_KEY}}
        AZURE_OPENAI_ENDPOINT: ${{SECRET_AZURE_OPENAI_ENDPOINT}}

• Open ./infra/azure.bicep and comment out lines 72-75 and uncomment lines 76-83.
• Open ./infra/azure.parameters.json and replace lines 20-22 with:

      "azureOpenAIKey": {
        "value": "${{SECRET_AZURE_OPENAI_KEY}}"
      },
      "azureOpenAIEndpoint": {
        "value": "${{SECRET_AZURE_OPENAI_ENDPOINT}}"
      }

Please note: If you are using Azure OpenAI, you will need both a GPT model and embedding model deployment to get this sample working. See OpenAIModel in index.ts and OpenAIEmbeddings in VectraDataSource.ts.

3. Update ./src/prompts/chat/config.json and ./src/index.ts with your model deployment name.

Testing the sample

The easiest and fastest way to get up and running is with Teams Toolkit as your development guide. To use Teams Toolkit to automate setup and debugging, please continue below.

Otherwise, if you only want to run the bot locally and build manually, please jump to the BotFramework Emulator section. For different ways to test a sample see: Multiple ways to test

Using Teams Toolkit for Visual Studio Code

The simplest way to run this sample in Teams is to use Teams Toolkit for Visual Studio Code.

1. Ensure you have downloaded and installed Visual Studio Code
2. Install the Teams Toolkit extension
3. Copy this sample into a new folder outside of teams-ai
4. Select File > Open Folder in VS Code and choose this sample's directory
5. Using the extension, sign in with your Microsoft 365 account where you have permissions to upload custom apps
6. Ensure that you have set up the sample from the previous step.
7. Select Debug > Start Debugging or F5 to run the app in a Teams web client.
8. In the browser that launches, select the Add button to install the app to Teams.

If you do not have permission to upload custom apps (sideloading), Teams Toolkit will recommend creating and using a Microsoft 365 Developer Program account - a free program to get your own dev environment sandbox that includes Teams.

Using Teams App test tool

If you are using Azure OpenAI then follow these steps:

• Comment the SECRET_OPENAI_KEY variable in the ./env/.env.testtool file.
• Add your Azure OpenAI key and endpoint values to the SECRET_AZURE_OPENAI_KEY and SECRET_AZURE_OPENAI_ENDPOINT variables
• Open the teamsapp.testtool.yml file and modify the last step to use Azure OpenAI variables instead:

- uses: file/createOrUpdateEnvironmentFile
  with:
    target: ./.localConfigs.testTool
    envs:
      TEAMSFX_NOTIFICATION_STORE_FILENAME: ${{TEAMSFX_NOTIFICATION_STORE_FILENAME}}
      # OPENAI_KEY: ${{SECRET_OPENAI_KEY}}
      AZURE_OPENAI_KEY: ${{SECRET_AZURE_OPENAI_KEY}}
      AZURE_OPENAI_ENDPOINT: ${{SECRET_AZURE_OPENAI_ENDPOINT}}

1. Ensure you have downloaded and installed Visual Studio Code
2. Install the Teams Toolkit extension
3. Copy this sample into a new folder outside of teams-ai
4. Select File > Open Folder in VS Code and choose this sample's directory
5. From the left pane, select Run and Debug(Ctrl+Shift+D) and select Debug in Test Tool in dropdown list.
6. Select Debug > Start Debugging or F5 to run the app.
7. The browser will pop up to open Teams App Test Tool.