Caution

This Repo is in an Early Preview state, meaning that everything is subject to change

SimpleRag

A Simple and Opinionated AI RAG Solutions for SourceCode, where you just need to setup and point to your source and we take care of the rest.

Getting Started

Step 1: Install SimpleRag Nuget + your Embedding Generator (in this sample Azure OpenAI via Semantic Kernel and SQL Server 2025 as VectorStore)

dotnet add package SimpleRag
dotnet add package Microsoft.SemanticKernel
dotnet add package Microsoft.SemanticKernel.Connectors.SqlServer

Step 2: Setup Dependency Injection of Embedding Generator and SimpleRag

string endpoint = builder.Configuration["AiEndpoint"]!;
string key = builder.Configuration["AiKey"]!;
string embeddingDeploymentName = builder.Configuration["AiEmbeddingDeploymentName"]!;
string sqlServerConnectionString = builder.Configuration["SqlServerConnectionString"]!;
string githubToken = builder.Configuration["GitHubToken"]!;

//Setup Github Credentials (optional, but needed if you wish to RAG data directly from GitHub)
builder.Services.AddSingleton(new GitHubCredentials(githubToken));

//Setup Embedding Generator
builder.Services.AddAzureOpenAIEmbeddingGenerator(embeddingDeploymentName, endpoint, key);

//Setup SimpleRag
builder.Services.AddSimpleRag(vectorStoreConfiguration, options => new SqlServerVectorStore(sqlServerConnectionString, new SqlServerVectorStoreOptions
{
    EmbeddingGenerator = options.GetRequiredService<IEmbeddingGenerator<string, Embedding<float>>>()
}));

Step 3: Ingest Data into your VectorStore (in this sample the TrelloDotNet open-source repo)

public class IngestionExample(Ingestion ingestion, IServiceProvider serviceProvider)
{
    public async Task Sync()
    {
        CollectionId collectionId = new("TrelloDotNet");
        DataSourceProviderGitHub filesProvider = new(serviceProvider)
        {
            GitHubRepository = new()
            {
                Owner = "rwjdk",
                Name = "TrelloDotNet"
            }
        };

        await ingestion.IngestAsync(
        [
            new CSharpDataSource(serviceProvider)
            {
                CollectionId = collectionId,
                Id = new SourceId("Code"),
                Recursive = true,
                Path = "src",
                FileIgnorePatterns = "TrelloDotNet.Tests",
                FilesProvider = filesProvider
            },
            new MarkdownDataSource(serviceProvider)
            {
                CollectionId = collectionId,
                Id = new SourceId("Markdown"),
                Recursive = true,
                Path = "/",
                FilesProvider = filesProvider
                LevelsToChunk = 3,
            }
        ], new IngestionOptions
        {
            OnProgressNotification = notification => Console.WriteLine(notification.GetFormattedMessageWithDetails()),
        }, cancellationToken);
    }
}

Step 4: Search your VectorStore

public class SearchExample(Search search)
{
    public async Task<string> SearchAsync(string question)
    {
        SearchResult searchResult = await search.SearchAsync(new SearchOptions
        {
            CollectionId = new CollectionId("TrelloDotNet"),
            SearchQuery = searchQuery,
            NumberOfRecordsBack = 10
        });
        
        return searchResult.GetAsStringResult();
    }
}

Legal Stuff

• PDF Extraction is done with Nuget Package PDFPig (Apache 2.0 License)
• Retry Logic is provided by Nuget Package Polly (BSD-3)