Rag.NET 1.8.2

Rag.NET

A modular RAG pipeline for .NET, built on Microsoft.Extensions.AI. Ships as 74 packages so a pipeline downloads only what it calls. Documentation · Repository

The core Retrieval-Augmented Generation pipeline for .NET: AddRagNet() wires ingestion (parse → chunk → embed → store) and retrieval (query → search → rerank → answer) into any IServiceCollection, exposed to your code as one IRagPipeline.

Install

dotnet add package Rag.NET

Setup

Rag.NET builds on Microsoft.Extensions.AI: register any IChatClient and IEmbeddingGenerator<string, Embedding<float>> (OpenAI, Azure OpenAI, Ollama, …) before calling AddRagNet(). Out of the box this package parses text and Markdown, chunks recursively, and can run entirely in memory — perfect for a first spike before you attach a real vector store package such as Rag.NET.VectorStores.PgVector.

using Microsoft.Extensions.DependencyInjection;
using Rag.NET.Abstractions;
using Rag.NET.DependencyInjection;
using Rag.NET.Storage;

// chatClient / embeddingGenerator: your Microsoft.Extensions.AI implementations.
services.AddSingleton(chatClient);
services.AddSingleton(embeddingGenerator);

services.AddSingleton<IVectorStore>(new InMemoryVectorStore()); // nothing persisted
services.AddRagNet();

Example

using Microsoft.Extensions.DependencyInjection;
using Rag.NET.Abstractions;
using Rag.NET.Models;

var pipeline = provider.GetRequiredService<IRagPipeline>();

using var stream = File.OpenRead("notes.md");
var ingested = await pipeline.IngestAsync(stream, new DocumentMetadata
{
    DocumentId  = new DocumentId("notes"),
    FileName    = "notes.md",
    ContentType = "text/markdown",
});

var response = await pipeline.AskAsync("What do the notes say about pricing?");
Console.WriteLine(response.Answer);

foreach (var source in response.Sources)
    Console.WriteLine($"[{source.Score:F2}] {source.Chunk.Text}");

RetrieveAsync returns raw ranked chunks, AskStreamingAsync streams the answer token-by-token, and the RagBuilder passed to AddRagNet(rag => …) activates self-query, parent-document retrieval, MMR, corrective RAG, conversation memory and more.

Full guide

• Architecture
• Retrieval options
• Ingestion