I built RAGSharp: a lightweight RAG library for .NET

I’ve been exploring Retrieval-Augmented Generation (RAG) in .NET and noticed that most approaches I tried either came bundled with more features than I needed or required setting up external services like vector databases or cloud APIs.

So I built RAGSharp - a small C# library that focuses on the basics:

load → chunk → embed → search

What it includes

• Document loading (files, directories, web, Wikipedia; custom loaders supported)
• Token-aware text chunking (SharpToken for GPT-style tokenization)
• Embeddings (works with OpenAI-compatible endpoints, so LM Studio, Ollama, vLLM, or custom providers)
• Vector stores (in-memory/file-backed, no DB required; extensible to Postgres/Qdrant)
• A simple retriever to tie it all together

Quick example

var docs = await new FileLoader().LoadAsync("sample.txt");

var retriever = new RagRetriever(
    new OpenAIEmbeddingClient("http://localhost:1234/v1", "lmstudio", "bge-large"),
    new InMemoryVectorStore()
);

await retriever.AddDocumentsAsync(docs);
var results = await retriever.Search("quantum mechanics", topK: 3);

Links: Nuget | Repo

If you’re experimenting with RAG in .NET, I’d love to hear your feedback or ideas for making it more useful in real projects.