/sample-app-aoai-chatGPT

[PREVIEW] Sample code for a simple web chat experience targeting chatGPT through AOAI.

Primary LanguagePythonMIT LicenseMIT

[Preview] Sample Chat App with AOAI

This repo contains sample code for a simple chat webapp that integrates with Azure OpenAI. Note: some portions of the app use preview APIs.

Prerequisites

  • An existing Azure OpenAI resource and model deployment of a chat model (e.g. gpt-35-turbo-16k, gpt-4)
  • To use Azure OpenAI on your data: one of the following data sources:
    • Azure AI Search Index
    • Azure CosmosDB Mongo vCore vector index
    • Elasticsearch index (preview)
    • Pinecone index (preview)
    • AzureML index (preview)

Deploy the app

Deploy with Azure Developer CLI

Please see README_azd.md for detailed instructions.

One click Azure deployment

Deploy to Azure

Click on the Deploy to Azure button and configure your settings in the Azure Portal as described in the Environment variables section.

Please see the section below for important information about adding authentication to your app.

Deploy from your local machine

Local Setup: Basic Chat Experience

  1. Copy .env.sample to a new file called .env and configure the settings as described in the Environment variables section.

    These variables are required:

    • AZURE_OPENAI_RESOURCE or AZURE_OPENAI_ENDPOINT
    • AZURE_OPENAI_MODEL
    • AZURE_OPENAI_KEY (optional if using Entra ID)

    These variables are optional:

    • AZURE_OPENAI_TEMPERATURE
    • AZURE_OPENAI_TOP_P
    • AZURE_OPENAI_MAX_TOKENS
    • AZURE_OPENAI_STOP_SEQUENCE
    • AZURE_OPENAI_SYSTEM_MESSAGE

    See the documentation for more information on these parameters.

  2. Start the app with start.cmd. This will build the frontend, install backend dependencies, and then start the app. Or, just run the backend in debug mode using the VSCode debug configuration in .vscode/launch.json.

  3. You can see the local running app at http://127.0.0.1:50505.

NOTE: You may find you need to set: MacOS: export NODE_OPTIONS="--max-old-space-size=8192" or Windows: set NODE_OPTIONS=--max-old-space-size=8192 to avoid running out of memory when building the frontend.

Local Setup: Chat with your data using Azure Cognitive Search

More information about Azure OpenAI on your data

  1. Update the AZURE_OPENAI_* environment variables as described above.

  2. To connect to your data, you need to specify an Azure Cognitive Search index to use. You can create this index yourself or use the Azure AI Studio to create the index for you.

    These variables are required when adding your data with Azure AI Search:

    • DATASOURCE_TYPE (should be set to AzureCognitiveSearch)
    • AZURE_SEARCH_SERVICE
    • AZURE_SEARCH_INDEX
    • AZURE_SEARCH_KEY (optional if using Entra ID)

    These variables are optional:

    • AZURE_SEARCH_USE_SEMANTIC_SEARCH
    • AZURE_SEARCH_SEMANTIC_SEARCH_CONFIG
    • AZURE_SEARCH_INDEX_TOP_K
    • AZURE_SEARCH_ENABLE_IN_DOMAIN
    • AZURE_SEARCH_CONTENT_COLUMNS
    • AZURE_SEARCH_FILENAME_COLUMN
    • AZURE_SEARCH_TITLE_COLUMN
    • AZURE_SEARCH_URL_COLUMN
    • AZURE_SEARCH_VECTOR_COLUMNS
    • AZURE_SEARCH_QUERY_TYPE
    • AZURE_SEARCH_PERMITTED_GROUPS_COLUMN
    • AZURE_SEARCH_STRICTNESS
    • AZURE_OPENAI_EMBEDDING_NAME

  3. Start the app with start.cmd. This will build the frontend, install backend dependencies, and then start the app. Or, just run the backend in debug mode using the VSCode debug configuration in .vscode/launch.json.

  4. You can see the local running app at http://127.0.0.1:50505.

NOTE: You may find you need to set: MacOS: export NODE_OPTIONS="--max-old-space-size=8192" or Windows: set NODE_OPTIONS=--max-old-space-size=8192 to avoid running out of memory when building the frontend.

Local Setup: Enable Chat History

To enable chat history, you will need to set up CosmosDB resources. The ARM template in the infrastructure folder can be used to deploy an app service and a CosmosDB with the database and container configured. Then specify these additional environment variables:

  • AZURE_COSMOSDB_ACCOUNT
  • AZURE_COSMOSDB_DATABASE
  • AZURE_COSMOSDB_CONVERSATIONS_CONTAINER
  • AZURE_COSMOSDB_ACCOUNT_KEY

As above, start the app with start.cmd, then visit the local running app at http://127.0.0.1:50505. Or, just run the backend in debug mode using the VSCode debug configuration in .vscode/launch.json.

Local Setup: Enable Message Feedback

To enable message feedback, you will need to set up CosmosDB resources. Then specify these additional environment variable:

/.env

  • AZURE_COSMOSDB_ENABLE_FEEDBACK=True

Local Setup: Enable SQL Server

To enable SQL Server, you will need to set up SQL Server resources. Then specify these additional environment variables:

  • DATASOURCE_TYPE (Should be set to AzureSqlServer)
  • AZURE_SQL_SERVER_CONNECTION_STRING
  • AZURE_SQL_SERVER_TABLE_SCHEMA

Deploy with the Azure CLI

NOTE: If you've made code changes, be sure to build the app code with start.cmd or start.sh before you deploy, otherwise your changes will not be picked up. If you've updated any files in the frontend folder, make sure you see updates to the files in the static folder before you deploy.

You can use the Azure CLI to deploy the app from your local machine. Make sure you have version 2.48.1 or later.

If this is your first time deploying the app, you can use az webapp up. Run the following two commands from the root folder of the repo, updating the placeholder values to your desired app name, resource group, location, and subscription. You can also change the SKU if desired.

  1. az webapp up --runtime PYTHON:3.11 --sku B1 --name <new-app-name> --resource-group <resource-group-name> --location <azure-region> --subscription <subscription-name>
  2. az webapp config set --startup-file "python3 -m gunicorn app:app" --name <new-app-name>

If you've deployed the app previously, first run this command to update the appsettings to allow local code deployment:

az webapp config appsettings set -g <resource-group-name> -n <existing-app-name> --settings WEBSITE_WEBDEPLOY_USE_SCM=false

Check the runtime stack for your app by viewing the app service resource in the Azure Portal. If it shows "Python - 3.10", use PYTHON:3.10 in the runtime argument below. If it shows "Python - 3.11", use PYTHON:3.11 in the runtime argument below.

Check the SKU in the same way. Use the abbreviated SKU name in the argument below, e.g. for "Basic (B1)" the SKU is B1.

Then, use these commands to deploy your local code to the existing app:

  1. az webapp up --runtime <runtime-stack> --sku <sku> --name <existing-app-name> --resource-group <resource-group-name>
  2. az webapp config set --startup-file "python3 -m gunicorn app:app" --name <existing-app-name>

Make sure that the app name and resource group match exactly for the app that was previously deployed.

Deployment will take several minutes. When it completes, you should be able to navigate to your app at {app-name}.azurewebsites.net.

Add an identity provider

After deployment, you will need to add an identity provider to provide authentication support in your app. See this tutorial for more information.

If you don't add an identity provider, the chat functionality of your app will be blocked to prevent unauthorized access to your resources and data.

To remove this restriction, you can add AUTH_ENABLED=False to the environment variables. This will disable authentication and allow anyone to access the chat functionality of your app. This is not recommended for production apps.

To add further access controls, update the logic in getUserInfoList in frontend/src/pages/chat/Chat.tsx.

Common Customization Scenarios (e.g. updating the default chat logo and headers)

The interface allows for easy adaptation of the UI by modifying certain elements, such as the title and logo, through the use of environment variables.

  • UI_TITLE
  • UI_LOGO
  • UI_CHAT_TITLE
  • UI_CHAT_LOGO
  • UI_CHAT_DESCRIPTION
  • UI_FAVICON
  • UI_SHOW_SHARE_BUTTON
  • UI_SHOW_CHAT_HISTORY_BUTTON

Feel free to fork this repository and make your own modifications to the UX or backend logic. You can modify the source (frontend/src). For example, you may want to change aspects of the chat display, or expose some of the settings in app.py in the UI for users to try out different behaviors. After your code changes, you will need to rebuild the front-end via start.sh or start.cmd.

Scalability

You can configure the number of threads and workers in gunicorn.conf.py. After making a change, redeploy your app using the commands listed above.

See the Oryx documentation for more details on these settings.

Debugging your deployed app

First, add an environment variable on the app service resource called "DEBUG". Set this to "true".

Next, enable logging on the app service. Go to "App Service logs" under Monitoring, and change Application logging to File System. Save the change.

Now, you should be able to see logs from your app by viewing "Log stream" under Monitoring.

Configuring vector search

When using your own data with a vector index, ensure these settings are configured on your app:

  • AZURE_SEARCH_QUERY_TYPE: can be vector, vectorSimpleHybrid, or vectorSemanticHybrid,
  • AZURE_OPENAI_EMBEDDING_NAME: the name of your Ada (text-embedding-ada-002) model deployment on your Azure OpenAI resource.
  • AZURE_SEARCH_VECTOR_COLUMNS: the vector columns in your index to use when searching. Join them with | like contentVector|titleVector.

Changing Citation Display

The Citation panel is defined at the end of frontend/src/pages/chat/Chat.tsx. The citations returned from Azure OpenAI On Your Data will include content, title, filepath, and in some cases url. You can customize the Citation section to use and display these as you like. For example, the title element is a clickable hyperlink if url is not a blob URL.

    <h5 
        className={styles.citationPanelTitle} 
        tabIndex={0} 
        title={activeCitation.url && !activeCitation.url.includes("blob.core") ? activeCitation.url : activeCitation.title ?? ""} 
        onClick={() => onViewSource(activeCitation)}
    >{activeCitation.title}</h5>

    const onViewSource = (citation: Citation) => {
        if (citation.url && !citation.url.includes("blob.core")) {
            window.open(citation.url, "_blank");
        }
    };

Using Entra ID

The app uses Azure OpenAI on your data (see documentation). To enable Entra ID for intra-service authentication

  1. Enable managed identity on Azure OpenAI
  2. Configure AI search to allow access from Azure OpenAI
    1. Enable Role Based Access control on the used AI search instance (see documentation)
    2. Assign Search Index Data Reader and Search Service Contributor to the identity of the Azure OpenAI instance
  3. Do not configure AZURE_SEARCH_KEY and AZURE_OPENAI_KEY to use Entra ID authentication.
  4. Configure the webapp identity
    1. Enable managed identity in the app service that hosts the webapp
    2. Go to the Azure OpenAI instance and assign the role Cognitive Services OpenAI User to the identity of the webapp

Note: RBAC assignments can take a few minutes before becoming effective.

Best Practices

We recommend keeping these best practices in mind:

  • Reset the chat session (clear chat) if the user changes any settings. Notify the user that their chat history will be lost.
  • Clearly communicate to the user what impact each setting will have on their experience.
  • When you rotate API keys for your AOAI or ACS resource, be sure to update the app settings for each of your deployed apps to use the new key.
  • Pull in changes from main frequently to ensure you have the latest bug fixes and improvements, especially when using Azure OpenAI on your data.

A note on Azure OpenAI API versions: The application code in this repo will implement the request and response contracts for the most recent preview API version supported for Azure OpenAI. To keep your application up-to-date as the Azure OpenAI API evolves with time, be sure to merge the latest API version update into your own application code and redeploy using the methods described in this document.

Environment variables

Note: settings starting with AZURE_SEARCH are only needed when using Azure OpenAI on your data with Azure AI Search. If not connecting to your data, you only need to specify AZURE_OPENAI settings.

App Setting Value Note
AZURE_SEARCH_SERVICE The name of your Azure AI Search resource
AZURE_SEARCH_INDEX The name of your Azure AI Search Index
AZURE_SEARCH_KEY An admin key for your Azure AI Search resource.
AZURE_SEARCH_USE_SEMANTIC_SEARCH False Whether or not to use semantic search
AZURE_SEARCH_QUERY_TYPE simple Query type: simple, semantic, vector, vectorSimpleHybrid, or vectorSemanticHybrid. Takes precedence over AZURE_SEARCH_USE_SEMANTIC_SEARCH
AZURE_SEARCH_SEMANTIC_SEARCH_CONFIG The name of the semantic search configuration to use if using semantic search.
AZURE_SEARCH_TOP_K 5 The number of documents to retrieve from Azure AI Search.
AZURE_SEARCH_ENABLE_IN_DOMAIN True Limits responses to only queries relating to your data.
AZURE_SEARCH_CONTENT_COLUMNS List of fields in your Azure AI Search index that contains the text content of your documents to use when formulating a bot response. Represent these as a string joined with "
AZURE_SEARCH_FILENAME_COLUMN Field from your Azure AI Search index that gives a unique identifier of the source of your data to display in the UI.
AZURE_SEARCH_TITLE_COLUMN Field from your Azure AI Search index that gives a relevant title or header for your data content to display in the UI.
AZURE_SEARCH_URL_COLUMN Field from your Azure AI Search index that contains a URL for the document, e.g. an Azure Blob Storage URI. This value is not currently used.
AZURE_SEARCH_VECTOR_COLUMNS List of fields in your Azure AI Search index that contain vector embeddings of your documents to use when formulating a bot response. Represent these as a string joined with "
AZURE_SEARCH_PERMITTED_GROUPS_COLUMN Field from your Azure AI Search index that contains AAD group IDs that determine document-level access control.
AZURE_SEARCH_STRICTNESS 3 Integer from 1 to 5 specifying the strictness for the model limiting responses to your data.
AZURE_OPENAI_RESOURCE the name of your Azure OpenAI resource (only one of AZURE_OPENAI_RESOURCE/AZURE_OPENAI_ENDPOINT is required)
AZURE_OPENAI_MODEL The name of your model deployment
AZURE_OPENAI_ENDPOINT The endpoint of your Azure OpenAI resource (only one of AZURE_OPENAI_RESOURCE/AZURE_OPENAI_ENDPOINT is required)
AZURE_OPENAI_MODEL_NAME gpt-35-turbo-16k The name of the model
AZURE_OPENAI_KEY One of the API keys of your Azure OpenAI resource (optional if using Entra ID)
AZURE_OPENAI_TEMPERATURE 0 What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic. A value of 0 is recommended when using your data.
AZURE_OPENAI_TOP_P 1.0 An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. We recommend setting this to 1.0 when using your data.
AZURE_OPENAI_MAX_TOKENS 1000 The maximum number of tokens allowed for the generated answer.
AZURE_OPENAI_STOP_SEQUENCE Up to 4 sequences where the API will stop generating further tokens. Represent these as a string joined with "
AZURE_OPENAI_SYSTEM_MESSAGE You are an AI assistant that helps people find information. A brief description of the role and tone the model should use
AZURE_OPENAI_PREVIEW_API_VERSION 2024-02-15-preview API version when using Azure OpenAI on your data
AZURE_OPENAI_STREAM True Whether or not to use streaming for the response. Note: Setting this to true prevents the use of prompt flow.
AZURE_OPENAI_EMBEDDING_NAME The name of your embedding model deployment if using vector search.
UI_TITLE Contoso Chat title (left-top) and page title (HTML)
UI_LOGO Logo (left-top). Defaults to Contoso logo. Configure the URL to your logo image to modify.
UI_CHAT_LOGO Logo (chat window). Defaults to Contoso logo. Configure the URL to your logo image to modify.
UI_CHAT_TITLE Start chatting Title (chat window)
UI_CHAT_DESCRIPTION This chatbot is configured to answer your questions Description (chat window)
UI_FAVICON Defaults to Contoso favicon. Configure the URL to your favicon to modify.
UI_SHOW_SHARE_BUTTON True Share button (right-top)
UI_SHOW_CHAT_HISTORY_BUTTON True Show chat history button (right-top)
SANITIZE_ANSWER False Whether to sanitize the answer from Azure OpenAI. Set to True to remove any HTML tags from the response.
USE_PROMPTFLOW False Use existing Promptflow deployed endpoint. If set to True then both PROMPTFLOW_ENDPOINT and PROMPTFLOW_API_KEY also need to be set.
PROMPTFLOW_ENDPOINT URL of the deployed Promptflow endpoint e.g. https://pf-deployment-name.region.inference.ml.azure.com/score
PROMPTFLOW_API_KEY Auth key for deployed Promptflow endpoint. Note: only Key-based authentication is supported.
PROMPTFLOW_RESPONSE_TIMEOUT 120 Timeout value in seconds for the Promptflow endpoint to respond.
PROMPTFLOW_REQUEST_FIELD_NAME query Default field name to construct Promptflow request. Note: chat_history is auto constucted based on the interaction, if your API expects other mandatory field you will need to change the request parameters under promptflow_request function.
PROMPTFLOW_RESPONSE_FIELD_NAME reply Default field name to process the response from Promptflow request.
PROMPTFLOW_CITATIONS_FIELD_NAME documents Default field name to process the citations output from Promptflow request.
DATASOURCE_TYPE Type of data source to use for using the 'on-your-data' api. Can be AzureCognitiveSearch, AzureCosmosDB, Elasticsearch, Pinecone, AzureMLIndex, AzureSqlServer or None

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

When contributing to this repository, please help keep the codebase clean and maintainable by running the formatter and linter with npm run format this will run npx eslint --fix and npx prettier --write on the frontebnd codebase.

If you are using VSCode, you can add the following settings to your settings.json to format and lint on save:

{
    "editor.codeActionsOnSave": {
        "source.fixAll.eslint": "explicit"
    },
    "editor.formatOnSave": true,
    "prettier.requireConfig": true,
}

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.