Deploy and Build AI Apps with Dify: A Complete Open‑Source Guide

Introduction

Dify is an open‑source platform for building LLM‑powered applications. It combines an intuitive visual canvas with AI workflow orchestration, Retrieval‑Augmented Generation (RAG) pipelines, agent support, model management, and observability features, enabling rapid prototyping to production.

Core Features

Workflow Builder : Design and test complex AI workflows on a canvas, leveraging all platform capabilities.

Extensive Model Support : Integrates hundreds of proprietary and open‑source LLMs and dozens of inference providers, including GPT, Mistral, Llama 3, and any OpenAI‑compatible API.

Prompt IDE : Create prompts, compare model performance, and add extra functions such as text‑to‑speech.

RAG Pipeline : Out‑of‑the‑box support for document ingestion from PDF, PPT and other common formats, with full text extraction.

Agent Framework : Define agents via LLM function calls or ReAct, with over 50 built‑in tools (Google Search, DALL·E, Stable Diffusion, WolframAlpha, etc.).

LLMOps : Monitor and analyze application logs and performance, continuously improving prompts, datasets and models based on production data.

Backend‑as‑a‑Service : All features expose APIs for easy integration into custom business logic.

Feature Comparison

The table below compares Dify with LangChain, Flowise and the OpenAI Assistant API across programming approach, model support, RAG, agent capability, workflow support, observability, enterprise features, and deployment options.

Feature          Dify.AI   LangChain   Flowise   OpenAI Assistant API
Programming      API+App   Python code App‑oriented API‑oriented
Supported LLMs    Rich      Rich        Rich      OpenAI only
RAG Engine       ✅        ✅          ✅        ✅
Agent            ✅        ✅          ❌        ✅
Workflow         ✅        ❌          ✅        ❌
Observability    ✅        ✅          ❌        ❌
Enterprise (SSO) ✅        ❌          ❌        ❌
Local Deploy     ✅        ✅          ✅        ❌

System Requirements

CPU >= 2 Core
RAM >= 4GB

Local Deployment (Docker Compose)

Clone the Dify repository and start the services with Docker Compose:

## Clone Dify project
$ git clone [email protected]:langgenius/dify.git

## Docker Compose run
$ cd dify/docker/
$ docker-compose up -d

If the official script fails, an alternative community script is available:

$ git clone [email protected]:flyeric0212/eric-dify-docker.git
$ cd eric-dify-docker
$ docker-compose up -d

Kubernetes Deployment

Deploy Dify on a K8s cluster using Helm charts. Two community charts are referenced:

## Helm Chart by @LeoQuote
https://github.com/douban/charts/tree/master/charts/dify

## Helm Chart by @BorisPolonsky
https://github.com/BorisPolonsky/dify-helm

Verification

http://localhost:8090/install

The default port is 80; it is changed to 8090 locally to avoid conflicts. After starting the containers, you can list Docker volumes to verify persistent storage:

[20:25:18] dify $ docker volume ls
DRIVER    VOLUME NAME
local     1b48b646f10961973a2abb9c885b965d7f54860dca7d7a4d42a531dc13d96b0d
local     446e6dfa7f4d14c2aed281af1b772495af71698bdae62fcc9140dbc57ac0bd5a
local     dify_app-data
local     dify_postgres-data
local     dify_redis-data
local     dify_weaviate-data

You can stop and start the Dify app without losing data.

Quick Application Building

Add models such as ChatGPT and Ollama, then create an application using a template (e.g., Code Interpreter). First query with gpt-3.5-turbo, then switch to a local model like ollama3:8b for further testing.

Adding a Knowledge Base

Select a local data source (TXT, Markdown, PDF, etc.), let Dify segment and clean the documents, then store the embeddings in a vector database. Finally, create a new application that leverages this knowledge base.

Conclusion

Dify enables almost any mainstream model to be used via templates, allowing rapid creation of AI applications, integration of diverse document types as knowledge bases, and addition of backend APIs. Compared with LangChain, which requires Python coding, Dify offers an out‑of‑the‑box, user‑friendly experience and supports private deployment.

References

[1]

https://docs.dify.ai/getting-started/readme/model-providers