RAGAgenticFrameworks

RAGFlow Documentation — Open-Source RAG Engine with Deep Document Understanding

by InfiniFlow

IntermediateDocumentationFreemium~2-3 hours for the Docker quickstart plus dataset configuration; ongoing reference thereafter

Stand up a self-hosted, citation-backed document QA system over messy enterprise PDFs in an afternoon.

Start LearningAdded Aug 15, 2026 · Updated Aug 15, 2026

Overview

RAGFlow is an Apache-2.0 retrieval-augmented generation engine from InfiniFlow built around deep document understanding, and its documentation is structured as an operator's manual rather than a library reference: Releases, Quickstart, User Guides (Datasets, Chat, Chat Channels, Search, Agents, Memory, File Management, Team, Models, Data Sources, Knowledge Compilation), Administrator Guides, Developer Guides, References and FAQs. The Quickstart is Docker Compose — minimum 4 CPU cores, 16 GB RAM and 50 GB disk, Docker 24.0.0 or newer and Python 3.13 or newer, with a roughly 2 GB image that unpacks to about 7 GB — and the current stable release is v0.26.4. The heart of the product is the dataset configuration page, which exposes 13 built-in chunking templates (General, Q&A, Resume, Manual, Table, Paper, Book, Laws, Presentation, Picture, One, Tag, and the newer Ingestion Pipeline added in v0.21.0), each tuned to a document layout, alongside a selectable PDF parser and an embedding model that is locked once chunks exist. DeepDoc, its layout-aware parser, handles scanned PDFs, slide decks, spreadsheets and images, and every chunk is visible and editable in the UI, so a wrong answer can be traced back to the text that produced it; generated answers carry citations to their source chunks. The 2026 changelog adds agentic workflows with MCP support, agent memory, a gVisor-sandboxed code executor, and chat channels for Feishu, Discord, Telegram and Line. 88.5k GitHub stars.

At a Glance

Topic
RAG
Level
Intermediate
Format
Documentation
Cost
Freemium
Duration
~2-3 hours for the Docker quickstart plus dataset configuration; ongoing reference thereafter
Provider
InfiniFlow
Hands-on
Yes — code/exercises
Certificate
None

What You’ll Learn

  • Choose among 13 chunking templates — General, Q&A, Paper, Book, Laws, Table, Presentation and more
  • Parse complex PDFs, scanned images, slides and spreadsheets with the DeepDoc layout-aware parser
  • Configure a dataset's embedding model and PDF parser, and why embeddings cannot be swapped later
  • Inspect and correct chunks visually before they reach retrieval, instead of trusting a blind splitter
  • Produce grounded answers with citations traceable back to the source chunk and page
  • Assemble agentic workflows using the agent builder, MCP tools and a gVisor code sandbox
  • Deploy the full stack with Docker Compose and size CPU, RAM and disk for production use
  • Wire chat channels such as Feishu, Discord, Telegram and Line onto a knowledge base

Highlights

  • 88.5k GitHub stars under Apache-2.0 — an end-to-end self-hostable RAG engine, not another library you assemble yourself
  • DeepDoc does layout-aware extraction on scanned PDFs, tables and slide decks, which is exactly where naive text splitters lose the data
  • Chunks are visible and editable in the UI, so a bad answer can be traced to the chunk that caused it
  • Ships the whole pipeline behind one Docker Compose file: ingestion, retrieval, reranking, citation-backed generation, agents and MCP
  • Original Show HN drew 230 points and 53 comments in 2024, and the project has shipped continuously since, now at v0.26.4

Who It’s For

Best For

  • Engineers who need a working document-QA system this week, not a RAG library to integrate
  • Teams with messy enterprise PDFs, scans and spreadsheets that generic chunkers mangle
  • Anyone who must show citations and defend an answer to a compliance or legal reviewer
  • Developers evaluating self-hosted alternatives to managed RAG services

Prerequisites

  • Docker and Docker Compose experience — the minimum host is 4 CPU cores, 16 GB RAM and 50 GB disk
  • Conceptual understanding of embeddings, vector retrieval and reranking
  • Comfort operating and patching a self-hosted web application that your users can reach

FAQ

What is RAGFlow Documentation — Open-Source RAG Engine with Deep Document Understanding?

RAGFlow is an open-source, self-hosted RAG engine built around deep document understanding, and its documentation reads as a practical manual for standing one up. It suits engineers who need grounded question-answering over messy enterprise PDFs, scans, slides and spreadsheets rather than another library to assemble themselves. Working through the quickstart and dataset guides, you can deploy the stack with Docker Compose, choose the right chunking template per document type, inspect and correct chunks, and ship citation-backed answers plus agentic workflows over your own corpus.

Is RAGFlow Documentation — Open-Source RAG Engine with Deep Document Understanding free?

RAGFlow Documentation — Open-Source RAG Engine with Deep Document Understanding offers free content, with paid options for certificates or premium features.

What level is RAGFlow Documentation — Open-Source RAG Engine with Deep Document Understanding for?

RAGFlow Documentation — Open-Source RAG Engine with Deep Document Understanding is aimed at a intermediate audience. Recommended background: Docker and Docker Compose experience — the minimum host is 4 CPU cores, 16 GB RAM and 50 GB disk, Conceptual understanding of embeddings, vector retrieval and reranking, Comfort operating and patching a self-hosted web application that your users can reach.

How long does RAGFlow Documentation — Open-Source RAG Engine with Deep Document Understanding take?

Expect roughly ~2-3 hours for the Docker quickstart plus dataset configuration; ongoing reference thereafter. Most learners work through it at their own pace.

What will I learn from RAGFlow Documentation — Open-Source RAG Engine with Deep Document Understanding?

You'll learn: Choose among 13 chunking templates — General, Q&A, Paper, Book, Laws, Table, Presentation and more; Parse complex PDFs, scanned images, slides and spreadsheets with the DeepDoc layout-aware parser; Configure a dataset's embedding model and PDF parser, and why embeddings cannot be swapped later; Inspect and correct chunks visually before they reach retrieval, instead of trusting a blind splitter; Produce grounded answers with citations traceable back to the source chunk and page; Assemble agentic workflows using the agent builder, MCP tools and a gVisor code sandbox; Deploy the full stack with Docker Compose and size CPU, RAM and disk for production use; Wire chat channels such as Feishu, Discord, Telegram and Line onto a knowledge base.

Topics

ragdocument parsingself-hosteddeepdoccitationsvector search

Sources

This page was written from 4 sources, 2 on domains other than ragflow.io.

  1. 1.ragflow.iodocsvendor
  2. 2.ragflow.ioconfigure knowledge basevendor
  3. 3.github.comragflow
  4. 4.hn.algolia.comhn.algolia.com