Dify Documentation
by Dify (LangGenius)
The manual for the largest open-source LLM app platform — visual agent workflows, a node-based RAG pipeline, and a self-hostable backend.
Overview
Dify (github.com/langgenius/dify, ~155k stars and ~24.5k forks, pushed 2026-09-09) is the largest open-source LLM application platform by GitHub stars, and these are its official docs — roughly 264 pages for Cloud and 276 for Self-Host, mirrored in English, Chinese and Japanese, plus about 152 pages of developer resources per track. The docs open with a Quick Start that builds a first app, then split by deployment: Dify Cloud, or the self-hosted Community Edition on Docker Compose (v2.24.0+, minimum two cores and 4 GB RAM). Beyond that sit an API Reference with OpenAPI specs, a CLI section, a Plugin Development track and a tutorials library. The product surface the docs cover maps to seven capabilities: a visual Workflow canvas for building and testing pipelines; support for hundreds of models across dozens of providers; a Prompt IDE for drafting prompts and comparing model output; a RAG pipeline with document ingestion including PDF and PPT; agent capabilities with 50+ built-in tools; LLMOps monitoring, logging and annotation; and backend-as-a-service APIs for embedding an app in an existing product. Two more recent additions are documented in depth — the Knowledge Pipeline, which puts the RAG ETL path on the same canvas so source connection, parsing and chunking strategy are each a visible node, and the Agent Node, which hands a chosen step inside an otherwise fixed workflow to autonomous multi-step tool reasoning.
At a Glance
- Topic
- Frameworks
- Level
- Intermediate
- Format
- Documentation
- Cost
- Freemium
- Duration
- ~3-4 hours for Quick Start plus a first RAG app; ~276-page self-host reference
- Provider
- Dify (LangGenius)
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Build an agentic workflow on a visual canvas and publish it as a callable API
- ✓Design a Knowledge Pipeline so each RAG ETL step is an inspectable node
- ✓Configure chunking strategy, embedding model and retrieval settings per knowledge base
- ✓Use the Agent Node to delegate one workflow step to multi-step tool reasoning
- ✓Self-host the Community Edition with Docker Compose and size the deployment correctly
- ✓Build and publish plugins against the plugin SDK and the Dify marketplace
- ✓Integrate a finished app through the backend-as-a-service REST endpoints
- ✓Monitor, log and annotate production traffic for ongoing LLMOps work
Highlights
- •The single biggest open-source LLM app platform on GitHub, so the docs describe something with a real deployment base
- •Knowledge Pipeline exposes RAG ETL as visible nodes instead of hiding chunking behind a toggle
- •Documents both the low-code canvas and the full API, CLI and plugin SDK, so you are not trapped in the GUI
- •Self-host instructions are specific — Docker Compose v2.24.0+, two cores, 4 GB RAM — rather than 'deploy anywhere'
- •Check the licence before commercial use: GitHub reports a custom licence, not a stock OSI one
Who It’s For
Best For
- ✓Teams prototyping LLM apps before committing engineers to a bespoke codebase
- ✓Engineers who need a self-hosted RAG and agent platform for data-residency reasons
- ✓Platform teams exposing internal LLM apps to product teams as APIs
Prerequisites
- •Docker and Docker Compose basics if you plan to self-host
- •An API key for at least one LLM provider
- •Familiarity with RAG concepts — chunking, embeddings and retrieval
FAQ
What is Dify Documentation?
Official documentation for Dify, an open-source platform for building agents, agentic workflows and chatbots over your own data and publishing them as web apps or APIs. It is written for engineers who want a self-hostable platform rather than a framework they assemble themselves. After working through it you can build a RAG-backed agent on the canvas, self-host it with Docker Compose, and call it from your own product.
Is Dify Documentation free?
Dify Documentation offers free content, with paid options for certificates or premium features.
What level is Dify Documentation for?
Dify Documentation is aimed at a intermediate audience. Recommended background: Docker and Docker Compose basics if you plan to self-host, An API key for at least one LLM provider, Familiarity with RAG concepts — chunking, embeddings and retrieval.
How long does Dify Documentation take?
Expect roughly ~3-4 hours for Quick Start plus a first RAG app; ~276-page self-host reference. Most learners work through it at their own pace.
What will I learn from Dify Documentation?
You'll learn: Build an agentic workflow on a visual canvas and publish it as a callable API; Design a Knowledge Pipeline so each RAG ETL step is an inspectable node; Configure chunking strategy, embedding model and retrieval settings per knowledge base; Use the Agent Node to delegate one workflow step to multi-step tool reasoning; Self-host the Community Edition with Docker Compose and size the deployment correctly; Build and publish plugins against the plugin SDK and the Dify marketplace; Integrate a finished app through the backend-as-a-service REST endpoints; Monitor, log and annotate production traffic for ongoing LLMOps work.
Topics
Sources
This page was written from 2 sources, 1 on domains other than docs.dify.ai.