MS-RAGS(ALL-IN-ONE) Docs

Visual architecture

From guided choices to a running RAG system

MS-RAGS(ALL-IN-ONE) is designed to make the invisible parts of RAG visible: every model, extractor, database, retriever, quality layer, permission gate, and generated deployment file has a clear place in the pipeline.

15RAG patterns

12Vector DBs

16+File types

3Generated files

MS-RAGS permission-first workbench

LLM provider Extraction Chunking Embeddings Vector DB Retrieval Reranking Compression Evaluation OTel logs Live query Code export

Use it step by step

Build Your First RAG Pipeline Without Guessing

Think of MS-RAGS(ALL-IN-ONE) like a guided setup assistant. You answer one simple question at a time, review the choices before anything is written, test the pipeline live, and then export code you can deploy.

Install and run

Create a Python environment, Install MS-RAGS(ALL-IN-ONE), then run ms-rags. The terminal opens the guided workflow.

Pick your provider

Choose the LLM provider you want. If you are new, start with one provider only so setup stays simple.

Choose the RAG style

Start with Naive RAG for learning or Advanced RAG for a better production baseline.

Add your documents

Select document types, loaders, and chunking. MS-RAGS(ALL-IN-ONE) explains what each choice means before moving on.

Choose embeddings carefully

Pick an embedding model and review its dimension. If you change embeddings later, create a new vector collection.

Approve vector storage

Review the database, collection name, sources, and embedding dimension before MS-RAGS(ALL-IN-ONE) writes vectors.

Ask real questions

Use the live query loop to test answers. If answers are weak, improve extraction and chunking before adding complexity.

Export your app

Generate standalone pipeline.py and requirements.txt, then test them in a clean environment.

Beginner path: one provider, Advanced RAG, one document type, Recursive Character chunking, one embedding model, ChromaDB locally, Dense Vector retrieval, then live questions before generated code.

Security checklist and official resources

Why MS-RAGS(ALL-IN-ONE) Is Different

Most RAG tools either give you a library and expect you to already know the architecture, or give you a demo that is hard to turn into production. MS-RAGS(ALL-IN-ONE) sits in the middle: it teaches the user what each choice means, builds a live pipeline, then exports clean standalone Python code.

Guided architecture

No hidden RAG decisions

Users are guided through 16 explicit stages: provider credentials, RAG architecture, document formats, extractors, chunking, embeddings, vector storage, ingestion, retrieval, enhancement, reranking, compression, prompts, evaluation, runtime build, and generated code.

Permission-first

Designed for user confidence

Before state-changing work, the terminal shows what will happen and asks for approval. This matters for production users who do not want to accidentally write vectors to the wrong collection or use the wrong embedding dimension.

Standalone output

A workbench, not lock-in

The generated pipeline.py, requirements.txt, and .env templates are independent of the MS-RAGS(ALL-IN-ONE) package. Users can inspect, modify, commit, deploy, and own the resulting app.

What You Can Build

Internal knowledge assistants

Index policies, handbooks, tickets, product docs, PDFs, Markdown, SQL rows, MongoDB collections, and web pages so employees can ask grounded questions.

Customer support RAG

Combine hybrid retrieval, reranking, context compression, structured prompts, evaluation gates, and monitoring export to improve support answer quality.

Research and legal workflows

Use parent-child retrieval, sentence-aware chunking, reranking, and strict grounded prompts for long documents where context preservation matters.

Agentic retrieval systems

Use LangGraph-backed Agentic RAG, Self-RAG, Corrective RAG, or Adaptive RAG when queries need routing, relevance checks, approved fallback behavior, or permission-gated tools.

Feature Coverage

Layer	What MS-RAGS(ALL-IN-ONE) includes	Why it matters
RAG architectures	15 types including Naive, Advanced, Modular, Agentic, Self-RAG, CRAG, GraphRAG, HyDE, Multi-Query, RAG-Fusion, Parent-Child, Adaptive, and Contextual Compression.	Users can choose the right architecture instead of forcing every workload into one pipeline.
Providers	OpenAI, Anthropic, Cohere, HuggingFace, Google Gemini, Mistral, Groq, Together AI, Replicate, Azure OpenAI, AWS Bedrock, Ollama local/cloud.	Teams can match their cloud, compliance, latency, and cost requirements.
Document extraction	PDF, DOCX, CSV, Excel, PPTX, HTML, Markdown, JSON, XML, URLs, YouTube, OCR images, source code, SQL, MongoDB, ePub, RTF, plain text.	RAG quality starts with extraction quality. MS-RAGS(ALL-IN-ONE) exposes the extraction choice instead of hiding it.
Chunking	Recursive, fixed, semantic, sentence, paragraph, token, Markdown-aware, HTML-aware, code-aware, agentic, document-aware.	Chunk boundaries directly control retrieval quality and answer grounding.
Vector storage	ChromaDB, FAISS, Pinecone, Qdrant, Weaviate, Milvus, Redis, PGVector, Elasticsearch, OpenSearch, Azure AI Search, MongoDB Atlas.	Users can start locally and later move to managed production infrastructure.
Quality controls	Query enhancement, reranking, compression, evaluation frameworks, CI/CD thresholds, structured logs, OpenTelemetry.	Production RAG needs measurement, tracing, and correction paths, not just vector search.

System Flow

High-level pipeline: credentials and provider setup -> document discovery -> loader/extractor -> chunk splitter -> embedding model -> vector database ingestion -> retriever -> optional query enhancement -> optional reranking -> optional context compression -> system prompt -> LLM answer -> optional evaluation and telemetry -> generated standalone code.

Documentation Map

Getting started

Install the framework, run ms-rags, understand the 16-step setup flow, review environment variables, and learn the permission gates.

RAG types

Read how each architecture works internally, when to use it, what it costs, and which retrieval or evaluation settings pair well with it.

Pipeline components

Learn providers, embeddings, extractors, chunk splitters, retrieval, enhancement, reranking, compression, evaluation, and generated-code behavior.

Vector databases

Compare Chroma, FAISS, Pinecone, Qdrant, Weaviate, Milvus, Redis, PGVector, Elasticsearch, OpenSearch, Azure AI Search, and MongoDB Atlas.

Recommendations

Pick a sensible stack for demos, local privacy, production SaaS, enterprise cloud, long documents, and high-quality support bots.

Production readiness

Use OpenTelemetry, smoke tests, generated-code workflows, and troubleshooting guidance before shipping to users.

Official resources

Find official project, provider, vector database, parser, evaluation, tracing, and framework links from one place.

Learn The Feature Before You Pick It

MS-RAGS(ALL-IN-ONE) is built for users who may not already know every RAG term. Each page explains what the feature does, when to use it, when to avoid it, what credentials it needs, what production risk it introduces, and which simpler option to try first.

Unsure which RAG type to choose?

Start with the decision guide, then read the RAG type internals before selecting advanced or graph-based architectures.

Unsure which embedding model to choose?

Read the embedding notes before ingestion. The docs explain local downloads, hosted endpoints, dimensions, and why model changes need a new collection.

Unsure which database to choose?

Compare local, managed, search-based, relational, and document-native vector stores before adding credentials.

Recommended First Pipeline

For most first-time users: Advanced RAG, OpenAI or Ollama Cloud LLM, OpenAI text-embedding-3-small or hosted HuggingFace embeddings, ChromaDB locally, Recursive Character chunking, Dense Vector retrieval, then add reranking only after measuring baseline quality.

Next: Getting started Production guide Official resources