Kernel Memory with Azure

November 2024

Kernel Memory with Azure OpenAI, Blob storage and AI Search.

Azure Open AI

On AzureOpenAI resource, deploy gpt-4 chat completion model and text-embedding-ada-002 embedding model

var builder = new KernelMemoryBuilder()
    .WithAzureOpenAITextGeneration(
        new AzureOpenAIConfig
        {
            Auth = AzureOpenAIConfig.AuthTypes.APIKey,
            APIKey = "Your AzureOpenAI api key",
            Endpoint = "https://your-azure-open-ai-resource-name.openai.azure.com",
            Deployment = "gpt-4"
        })
    .WithAzureOpenAITextEmbeddingGeneration(
        new AzureOpenAIConfig
        {
            Auth = AzureOpenAIConfig.AuthTypes.APIKey,
            APIKey = "Your AzureOpenAI api key",
            Endpoint = "https://your-azure-open-ai-resource-name.openai.azure.com",
            Deployment = "text-embedding-ada-002"
        });

Azure storage account

Azure blob storage to store kenerl memory pipeline artifacts

var builder = new KernelMemoryBuilder()
    .WithAzureBlobsDocumentStorage(
        new AzureBlobsConfig
        {
            Account = "your-blob-storage-account",
            Auth = AzureBlobsConfig.AuthTypes.AccountKey,
            AccountKey = "your-blob-account-key",
            Container = "document-ingestion"
        })

Azure AI Search service

Azure AI search service as vector databases

var builder = new KernelMemoryBuilder()
    .WithAzureAISearchMemoryDb(
        new AzureAISearchConfig
        {
            Endpoint = "https://your-search-service-resource-name.search.windows.net",
            Auth = AzureAISearchConfig.AuthTypes.APIKey,
            APIKey = "your search service api key"
        })

Import some document and ask questions

await kernelMemory.ImportDocumentAsync(
    filePath: "resources/earth_book_2019_tagged.pdf",
    documentId: "earth_book_2019",
    index: "books");

var response =
    await kernelMemory.AskAsync(
        "Where is Amazon rainforest on earth?", 
        index: "books");    
   

Note the index name "books", kernel memory automatically creates Azure AI Search index name "books" if it does not exist and "books" folder in the blob container.

Sample code here