Skip to content

Latest commit

 

History

History
136 lines (107 loc) · 8.11 KB

README.md

File metadata and controls

136 lines (107 loc) · 8.11 KB
Raw

Copilot Chat Sample Application

This sample is for educational purposes only and is not recommended for production deployments.

About Copilot Chat

This sample allows you to build your own integrated large language model chat copilot. This is an enriched intelligence app, with multiple dynamic components including command messages, user intent, and memories.

The chat prompt and response will evolve as the conversation between the user and the application proceeds. This chat experience is orchestrated with Semantic Kernel and a Copilot Chat skill containing numerous functions that work together to construct each response.

UI Sample

Configure your environment

Before you get started, make sure you have the following requirements in place:

Start the WebApi Backend Server

The sample uses two applications, a front-end web UI, and a back-end API server. First, let’s set up and verify the back-end API server is running.

  1. Generate and trust a localhost developer certificate. Open a terminal and run:

    • For Windows and Mac run dotnet dev-certs https --trust and select Yes when asked if you want to install this certificate.
    • For Linux run dotnet dev-certs https

    Note: It is recommended you close all instances of your web browser after installing the developer certificates.

  2. Navigate to samples/apps/copilot-chat-app/webapi and open appsettings.json

    • Update the Completion and Embedding configuration sections:

      • Update AIService to the AI service you will be using (i.e., AzureOpenAI or OpenAI).
      • If your are using Azure OpenAI, update Endpoint to your Azure OpenAI resource Endpoint address (e.g.,
        http://contoso.openai.azure.com).

        If you are using OpenAI, this property will be ignored.

      • Set your Azure OpenAI key by opening a terminal in the webapi project directory and using dotnet user-secrets 
        cd semantic-kernel/samples/apps/copilot-chat-app/webapi
dotnet user-secrets set "Completion:Key" "MY_AZUREOPENAI_OR_OPENAI_KEY"
dotnet user-secrets set "Embedding:Key" "MY_AZUREOPENAI_OR_OPENAI_KEY"
      • Update DeploymentOrModelID to the Azure OpenAI deployment or OpenAI models you want to use.
        • For Completion, CopilotChat is optimized for Chat completion models, such as gpt-3.5-turbo and gpt-4

          Important: gpt-3.5-turbo is normally labelled as "gpt-35-turbo" (no period) in Azure OpenAI and "gpt-3.5-turbo" (with a period) in OpenAI.

        • For Embedding, text-embedding-ada-002 is sufficient and cost-effect for generating embeddings.

    • (Optional) To enable speech-to-text for chat input, update the AzureSpeech configuration section:

      If you have not already, you will need to create an Azure Speech resource (see ./webapi/appsettings.json for more details).

      • Update Region to whichever region is appropriate for your speech sdk instance.
      • Set your Azure speech key by opening a terminal in the webapi project directory and setting a dotnet user-secrets value for AzureSpeech:Key 
        dotnet user-secrets set "AzureSpeech:Key" "MY_AZURE_SPEECH_KEY"

  3. Build and run the back-end API server

    1. Open a terminal and navigate to samples/apps/copilot-chat-app/webapi

    2. Run dotnet build to build the project.

    3. Run dotnet run to start the server.

    4. Verify the back-end server is responding, open a web browser and navigate to https://localhost:40443/probe

      The first time accessing the probe you may get a warning saying that there is a problem with website's certificate. Select the option to accept/continue - this is expected when running a service on localhost It is important to do this, as your browser may need to accept the certificate before allowing the frontend to communicate with the backend.

    You may also need to acknowledge the Windows Defender Firewall, and allow the app to communicate over private or public networks as appropriate.

  4. Build and start the front-end application

    1. You will also need an Azure Active Directory (AAD) application registration.

      For more details on creating an application registration, go here.

      • Select Single-page application (SPA) as platform type, and set the Web redirect URI to https://localhost:3000
      • Select Accounts in any organizational directory and personal Microsoft Accounts as supported account types for this sample.
      • Make a note of the Application (client) ID from the Azure Portal, we will use of it later.

    2. Open a terminal and navigate to samples/apps/copilot-chat-app/webapp Copy env.example into a new file .env and update the REACT_APP_AAD_CLIENT_ID with the AAD application (Client) ID created above. For example:

      REACT_APP_BACKEND_URI=https://localhost:40443/
REACT_APP_AAD_CLIENT_ID=00000000-0000-0000-0000-000000000000
REACT_APP_AAD_AUTHORITY=https://login.microsoftonline.com/common

    3. To build and run the front-end application

      yarn install
yarn start

    4. With the back end and front end running, your web browser should automatically launch and navigate to http://localhost:3000

      The first time running the front-end application may take a minute or so to start.

    5. Sign in with a Microsoft personal account or a "Work or School" account.

    6. Consent permission for the application to read your profile information (i.e., your name).

    If you you experience any errors or issues, consult the troubleshooting section below.

  5. Have fun!

    Note: Each chat interaction will call Azure OpenAI/OpenAI which will use tokens that you may be billed for.

Troubleshooting

1. Localhost SSL certificate errors

If you are stopped at an error message similar to the one above, your browser may be blocking the front-end access to the back end while waiting for your permission to connect. To resolve this, try the following:

  1. Confirm the backend service is running by opening a web browser, and navigating to https://localhost:40443/probe
    • You should see a confirmation message: Semantic Kernel service is up and running
  2. If your browser asks you to acknowledge the risks of visiting an insecurewebsite, you must acknowledge the message before the front end will be allowed to connect to the back-end server.
    • Acknowledge, continue, and navigate until you see the message Semantic Kernel service is up and running
  3. Navigate to https://localhost:3000 or refresh the page to use the Copilot Chat application.

2. Configuration issues after updating your repo/fork.

As of PR #470, we have updated some of the top-level configuration keys. Most notably,

  • CompletionConfig is now Completion
  • EmbeddingConfig is now Embedding

You may also need to update the keys used for any secrets set with dotnet user-secrets set

3. Issues using text completion models, such as text-davinci-003

As of PR #499, CopilotChat now focuses support on chat completion models, such as gpt-3.5-* and gpt-4-* See OpenAI's model compatiblity for the complete list of current models supporting chat completions.

Additional resources

  1. Import Document Application: Import a document to the memory store.