Rag.NET Documentation

Rag.NET is a modular Retrieval-Augmented Generation (RAG) pipeline library for .NET, built on Microsoft.Extensions.AI abstractions. These docs cover every layer from first setup to production-grade extensions.

Pages

Grouped the way the sidebar is.

Start here

Page	What it covers
Why RAG?	What RAG is, the problem it solves, and when Rag.NET is the right tool
Getting Started	Dependency injection setup, ingesting a document, and running a Q&A loop
Positioning	Where Rag.NET sits against Semantic Kernel, LangChain, LlamaIndex and Haystack — and where it loses

The pipeline

Page	What it covers
Choosing Packages	Which two or three of the 73 packages a given pipeline actually needs
Architecture	Pipeline internals, data-flow diagram, the core interfaces and models
Ingestion	Parsers, DocumentMetadata, IngestionOptions, OCR, progress reporting
Chunking	Eight strategies with a trade-off table, plus the domain-specific templates
Retrieval	RetrievalOptions, semantic and hybrid BM25+RRF search, metadata filtering, CRAG
Post-Retrieval	Lost-in-the-Middle reordering, redundancy filtering, MMR
Vector Stores	All seven stores, with the hybrid-search support matrix

Advanced retrieval

Page	What it covers
RAPTOR	Recursive abstractive tree summarisation: tree scope, retrieval modes, cluster sizing
GraphRAG	Entity extraction, community detection, local and global search, mind-map extraction
Query Techniques	HyDE, multi-query expansion, contextual compression
Answer Engines	MapReduce, Refine, FLARE and Dispatching answer strategies
Conversational Memory	In-session history trimming, token budgets, persistent cross-session recall

Sources

Page	What it covers
Data Providers	All 18 connectors: auth, delta-sync shape and options for each

Production

Page	What it covers
Security	Prompt-injection defence in depth, PII detection, RBAC, audit logging
Resilience	Retry and circuit-breaking, rate limiting, cost budgeting, fallback chains
Observability	ILogger structured logging and the ActivitySource the pipeline emits on
Pipeline Debugger	Per-query traces: chunk scores, stage latencies, guard actions
Evaluation	Embedding-distance and LLM-judge evaluators, RAGAS metrics, A/B comparison
A/B Shadow Mode	Running a second pipeline against production traffic without touching the response

Integration

Page	What it covers
Microsoft Foundry	Wiring the cloud endpoint, Foundry Local and the catalogue models, and what each one pins
MCP Server	Exposing a pipeline to an LLM agent as MCP tools, and the four deployment patterns
REST and gRPC	Serving a pipeline over HTTP or gRPC, with clients that implement IRagPipeline
CLI (ragnet)	Ingesting and querying a configured pipeline from the shell
Mediator	Dispatching ingest/retrieve/delete commands via IMediator
Extending	Implementing IDocumentParser, IVectorStore, IChunkingStrategy, IDocumentOcrEngine

Reference

Page	What it covers
Benchmarks	Measured throughput for chunking, embedding and retrieval
Retrieval Quality	BEIR results per technique, and the ablation table behind them
Library Comparison	Measured quality and cost against other RAG libraries
Comparison Scope	What each entrant was read for, cited per claim
Comparison Defaults	The defaults every entrant was measured at
OpenTelemetry	Every span and metric the pipeline emits, and how to export them
OSS Libraries	Every open-source dependency, where it is used, and why
CI and Test Tiers	Which suites run where, and what each tier needs to run at all

Quick links

• Sample applications: samples/Rag.NET.Sample — interactive console app (PgVector, Ollama/OpenAI) — and samples/Rag.NET.QuickStart — a config-driven walkthrough built on Rag.NET.Hosting
• Benchmark results: benchmarks.md
• How Rag.NET compares: quality and cost (measured) and scope (read, cited per claim)
• Feature roadmap and design notes: docs/plans/
• GitHub README: covers the quick-start and package list

Package layout

Rag.NET ships as 73 packages so that a pipeline downloads only the dependencies it actually uses. The shape is three layers — abstractions, core, and a satellite per opt-in feature — and Choosing packages walks through which two or three are yours.

Core

NuGet package	Contents
Rag.NET	Core pipeline, Text/Markdown/CSV/JSON parsers, RecursiveChunkingStrategy, in-memory vector store
Rag.NET.Abstractions	Interfaces, models and options — no implementations, no heavy dependencies. Arrives with core
Rag.NET.QueryTechniques	HyDE, multi-query expansion and contextual compression. Arrives with core

Chunking

NuGet package	Contents
Rag.NET.Chunking	Hierarchical-merge, code-aware, late, proposition, token-aware and embedding-based semantic chunking
Rag.NET.Chunking.CSharp	CSharpChunkingStrategy — Roslyn-based semantic chunking for C# source
Rag.NET.Chunking.Templates	Domain templates: Legal, Book, Academic Paper, Q&A Pairs, Email, Résumé
Rag.NET.Embeddings.Onnx	ONNX Runtime token-level embeddings, which late chunking needs

Vector stores

NuGet package	Contents
Rag.NET.VectorStores.PgVector	PostgreSQL + pgvector, with sparse-vector support
Rag.NET.VectorStores.Qdrant	Qdrant
Rag.NET.VectorStores.AzureAISearch	Azure AI Search, with native hybrid search
Rag.NET.VectorStores.Pinecone	Pinecone
Rag.NET.VectorStores.Chroma	Chroma
Rag.NET.VectorStores.Weaviate	Weaviate
Rag.NET.VectorStores.Redis	Redis (RediSearch)

Parsers

NuGet package	Contents
Rag.NET.Parsers.Pdf	PDF parser, with table extraction and Tesseract OCR
Rag.NET.Parsers.Pdf.AzureDocumentIntelligence	Whole-document OCR engine for the PDF parser (paid, per page)
Rag.NET.Parsers.Html	HTML parser (AngleSharp)
Rag.NET.Parsers.Office	Word .docx, Excel .xlsx and PowerPoint .pptx in one package (OpenXml)
Rag.NET.Parsers.Email	EML and MSG email parser (MimeKit)
Rag.NET.Parsers.Epub	EPUB parser
Rag.NET.Parsers.Archive	ZIP archive parser — parses each entry with whichever parser claims it
Rag.NET.Parsers.Audio	WAV/MP3/FLAC transcription via Whisper.net (local, no API key)
Rag.NET.Parsers.Vision	Image and video description via a vision LLM and FFMpeg

Data providers

NuGet package	Contents
Rag.NET.DataProviders	Shared OAuth, polling and watermark infrastructure. Arrives with any connector
Rag.NET.DataProviders.Web	Web crawler, sitemap loader, RSS/Atom feed loader
Rag.NET.DataProviders.Microsoft365	Exchange/Outlook mail, Teams, OneDrive and SharePoint via Microsoft Graph
Rag.NET.DataProviders.AzureBlob	Azure Blob Storage — ETag/LastModified delta sync
Rag.NET.DataProviders.GoogleDrive	Google Drive — pageToken change stream
Rag.NET.DataProviders.Dropbox	Dropbox — cursor-based delta sync
Rag.NET.DataProviders.Box	Box — events cursor delta sync
Rag.NET.DataProviders.Confluence	Confluence pages via REST API
Rag.NET.DataProviders.Jira	Jira issues via REST API
Rag.NET.DataProviders.Notion	Notion pages and blocks via REST API
Rag.NET.DataProviders.Asana	Asana tasks and subtasks via REST API
Rag.NET.DataProviders.Linear	Linear issues via GraphQL API
Rag.NET.DataProviders.Slack	Slack channel messages via REST API
Rag.NET.DataProviders.Gmail	Gmail messages via IMAP (MailKit)
Rag.NET.DataProviders.GitHub	GitHub repository files via Octokit
Rag.NET.DataProviders.GitLab	GitLab repository files via NGitLab
Rag.NET.DataProviders.Bitbucket	Bitbucket repository files via REST API
Rag.NET.DataProviders.Zendesk	Zendesk tickets and help-centre articles
Rag.NET.DataProviders.Airtable	Airtable rows and attachments
Rag.NET.Ingestion.AzureServiceBus	Consumes a queue or subscription, ingests each message end to end, and settles it (complete / abandon / dead-letter)

Advanced retrieval

NuGet package	Contents
Rag.NET.Raptor	RAPTOR — recursive abstractive tree summarisation, corpus-scoped by default
Rag.NET.Raptor.Store	Persistent leaf-chunk storage, which corpus-level RAPTOR clustering requires
Rag.NET.GraphRag	GraphRAG — entity extraction, community detection, local and global search, Mind-Map Extractor
Rag.NET.Graph	Standalone graph library — Leiden community detection, IGraphStore
Rag.NET.AnswerEngines	MapReduce, Refine, FLARE and Dispatching answer engines
Rag.NET.Memory	Persistent SQLite-backed cross-session conversation memory
Rag.NET.Reranking.Cohere	CohereReranker — hosted cross-encoder reranking
Rag.NET.Reranking.Onnx	OnnxReranker — local ONNX cross-encoder reranking (no API key)
Rag.NET.WebSearch.Tavily	Tavily web search, the corrective-RAG (CRAG) fallback source

Production

NuGet package	Contents
Rag.NET.Security	Prompt-injection defence in depth — chunk and query sanitisation, retrieval guards, prompt hardening, PII detection, RBAC
Rag.NET.Security.AspNetCore	Binds ICallerContext to ClaimsPrincipal
Rag.NET.Security.Audit.Sqlite	SQLite-backed audit log, split out so Security carries no native binary
Rag.NET.Resilience	Polly retry and circuit-breaking, token-bucket rate limiting, multi-provider chat fallback chain
Rag.NET.Caching	UseCaching() — the HybridCache implementation behind the embedding and result caches
Rag.NET.Storage.Sqlite	BM25 and parent-chunk persistence, document sidecar, content-hash record manager, embedding-version store, persistent cost ledger

Observability and evaluation

NuGet package	Contents
Rag.NET.Telemetry	AddRagNetInstrumentation() — OpenTelemetry SDK wiring, so core and its satellites take no SDK dependency
Rag.NET.Diagnostics	In-memory pipeline traces — the last N executions with chunk scores, stage latencies and guard actions
Rag.NET.Diagnostics.AspNetCore	Opt-in MapRagNetTrace() HTTP endpoint for those traces
Rag.NET.Evaluation	LLM-judge and embedding-distance evaluators, A/B comparison with confidence intervals, dataset generation, shadow capture
Rag.NET.Evaluation.Ragas	RAGAS-style metrics — faithfulness, answer relevancy, context precision and recall

Serving and integration

NuGet package	Contents
Rag.NET.Api	ASP.NET Core REST API over a pipeline
Rag.NET.Api.Client	IRagPipeline implemented over HTTP against that API
Rag.NET.Api.Grpc	gRPC service over a pipeline
Rag.NET.Api.Grpc.Client	IRagPipeline implemented over gRPC against that service
Rag.NET.Mcp	Model Context Protocol server exposing a pipeline as MCP tools
Rag.NET.Mcp.AspNetCore	HTTP transport for the MCP server, which refuses to serve an unauthenticated write surface
Rag.NET.Mcp.Tool	Self-contained MCP server as a dotnet global tool, configured entirely from appsettings.json
Rag.NET.Cli	ragnet global tool — ingest into, and query, a configured pipeline
Rag.NET.Hosting	Configuration-driven pipeline wiring for an executable
Rag.NET.Mediator	ZeroAlloc.Mediator integration — dispatch ingest/retrieve/delete via IMediator

Requirements

• .NET 10 or later
• A compatible embedding provider (OpenAI, Azure OpenAI, Ollama, etc.)
• A supported vector store (PostgreSQL+pgvector, Qdrant, or Azure AI Search)