Master Dify: From Local Deployment to Advanced AI Workflows in 2025

2025 Guide: Dify from Beginner to Expert

1. Introduction to Dify

Official site: Dify.AI – The Innovation Engine for Generative AI Applications.

Official Chinese site: Dify.AI – 生成式 AI 应用创新引擎.

Documentation: Product Overview – Dify Docs.

Dify is an open‑source LLM application development platform offering agent building, AI workflow orchestration, RAG retrieval, model management, and more, making it easy to create native generative AI apps. It is considered more user‑friendly than LangChain.

Dify provides two versions: Community Edition and Commercial Edition. For most users, the community version is sufficient because it can be deployed locally.

2. Local Deployment of Dify

2.1 Download Dify Source Code

Visit the GitHub repository and download version 1.1.3 source code. After downloading, unzip the archive.

In the source, locate the docker directory and rename .env.example to .env. This indicates that Docker is the only required service to run Dify locally.

2.2 Install Docker Desktop

Download Docker Desktop from the official website and follow the installation guide (including proxy settings if needed).

2.3 Run Dify

In the docker folder, open a command prompt and execute: docker-compose up -d The first run will pull necessary images; configure proxy settings in Docker Desktop if required. Once images are pulled, Dify starts successfully.

Access the installation page at http://localhost/install, set up email, username, and password, then log in to the main interface.

2.4 Ollama Integration

After deploying Dify locally, you can run large language models with Ollama (e.g., deepseek‑r1:1.5b). Ollama provides multi‑model support, easy installation across OSes, and local execution without network dependence.

2.5 Connect Dify with Ollama

Edit the .env file in the docker directory to enable custom models:

# Enable custom model
CUSTOM_MODEL_ENABLED=true
# Ollama API base URL
OLLAMA_API_BASE_URL=host.docker.internal:11434

In Dify’s UI ( http://localhost/apps), go to Settings → Model and add the Ollama model.

3. Core Features of Dify

3.1 Application Types

Chat Assistant

Create a simple conversational assistant, add prompt engineering and variables (e.g., multilingual translation).

Text Generation

Build apps for story writing, classification, translation, etc.

Agent

Agent assistants can plan, decompose tasks, call tools, and iterate without human intervention.

Chatflow

Design multi‑step dialogue workflows (e.g., start node → LLM node → end node) with input fields like city and weather.

Workflow

Automate batch processes such as high‑quality translation, data analysis, and email automation.

3.2 Tools

System Tools

Install built‑in tools from the Dify marketplace (e.g., SiliconFlow image generation).

Custom Tools

Create custom tools when built‑in ones are insufficient. Example: a custom image generation tool using SiliconFlow’s API, with a Python function to extract URLs from complex data structures.

from typing import Any, Union

def main(data: Any) -> str:
    """
    Safely extract the first URL from nested data structures.
    """
    # implementation omitted for brevity
    return {"result": extract_url(data)}

Integrate the custom tool into an agent workflow, authorize via API key, and test the image generation.

3.3 Export DSL

Dify allows exporting a case as a DSL file for sharing. Select a case, export the DSL, and import it into another Dify workspace.