Kenhuang Academy
  • English

      • Dify Tutorial

      Welcome to the Dify tutorial. Based on the official Dify docs and the open-source langgenius/dify project, this guide shows how to build agents, agentic workflows, and chatbots on your own data with a visual, low-code studio—then publish them as a web app or REST API.

      The name Dify comes from Do It For You.

      Table of Contents

      Basics

      1. Introduction — Positioning vs LangChain and Coze
      2. Installation — Cloud vs Docker Compose, hardware, first admin user
      3. Quick Start — One model, first chatbot, first conversation

      Capabilities

      1. Model Providers — API keys, Ollama, multi-model compare
      2. Knowledge Base — Upload, chunk, embed, retrieve, citations
      3. Workflow — Visual graph, nodes, when not to use an agent
      4. Agent — Tools, ReAct / Function Calling, vs workflow

      Publish & integrate

      1. Publish & API — Web App, API keys, /v1/chat-messages
      2. Ollama Integration — Local models, Docker networking, embeddings

      Practice

      1. Practical Examples — FAQ bot, classify-then-answer workflow, Python API
      2. Resources — Official docs and related tutorials on this site

      Learning path

      StageGoalChapters
      Day 1Run a chatbotIntro → Install → Quick Start
      Days 2–3Models & RAGProviders → Knowledge Base
      Week 1OrchestrationWorkflow → Agent → Publish API
      Week 2Local models & shipOllama → Examples

      Prerequisites

      • Basic LLM / Chat API ideas (messages, tokens, prompts)
      • At least one model source: a cloud API key, or local Ollama
      • Self-hosting also needs Docker / Docker Compose (see the Docker tutorial)
      TutorialHow it relates
      LangChainSame RAG / agent ideas, but code-first
      OllamaRun models locally, then plug them into Dify
      MCPDify can also publish an app as an MCP server for Cursor and similar clients

      Two ways to run Dify

      OptionBest forTrade-off
      Dify CloudInstant start, free SandboxData lives on the managed platform
      Community Edition (self-host)Intranet, compliance, local OllamaYou own the machine and ops

      Pick a path, then continue at Installation.

      What you will have built

      After the later chapters (especially Practical Examples) you should be able to demo three things:

      1. FAQ Chatbot — upload an internal policy, cite passages, refuse when retrieval is empty
      2. Classify-then-answer flow — bucket “policy / chitchat / other”, retrieve only when needed
      3. Backend integration — Python calls official POST /v1/chat-messages; the key never enters the browser

      For a fully local stack, point chat and embeddings at Ollama. Inside Docker the base URL must be host.docker.internal:11434, not localhost.

      How to read this guide

      • Studio labels drift; we follow flows and concepts from official docs, not unverified button names
      • Comparisons with LangChain mean “same RAG / agent ideas, different delivery”
      • Commands match upstream: clone the latest release tag, cd dify/docker, cp .env.example .env, docker compose up -d

      评论