Yes, to run it yourself. Dify's licence is the Apache License 2.0 with extra conditions: you need a commercial licence from LangGenius to run Dify as a multi-tenant service, where each workspace is a tenant, or to remove or change the logo and copyright notice in the Dify console and apps. Dify Cloud and Dify Enterprise are LangGenius's paid offerings. What your model provider charges for its API is separate.

Which AI models can I use?

No model is bundled, and the recipe runs none on your server. Model providers are plugins: in Integrations, Model Provider, install one such as OpenAI, Anthropic, Gemini, Azure OpenAI, Amazon Bedrock, DeepSeek, Mistral, Groq, OpenRouter or any OpenAI-API-compatible service, or a model server you run yourself, such as Ollama, LocalAI or vLLM. Click Add API Key on its card and enter the key. Under Default Models, pick the system reasoning model and, for knowledge bases, an embedding model. Installing plugins needs outbound internet access to the Dify Marketplace and PyPI.

How much RAM does Dify need?

Dify's Docker Compose guide sets a minimum of 2 CPU cores and 4 GiB of memory. The recipe caps api, worker and the plugin daemon at 2 GB each, web and worker_beat at 1 GB, the sandbox and the Squid proxy at 512 MB, and gives nginx the app's Memory limit (at least 256 MB); PostgreSQL and Redis get the panel's database default. These are upper limits, not reservations: on our test servers the Dify services used just under 2 GB together at idle, before PostgreSQL and Redis. Installed plugins run inside the plugin daemon, so its memory use grows with them.

Can my team work in Dify?

Yes. The owner and admins add people under Workspace settings, Members, with the Add button, and give each a role: Admin, Editor or Normal, where Normal can only use published apps. Dify emails the invitation only when a mail service is set: add MAIL_TYPE (smtp, resend or sendgrid), MAIL_DEFAULT_SEND_FROM and that service's settings as variables to the api and worker services on the app's Compose File page, then click Save and apply. Without it, copy the link from the Invitation sent dialog and send it yourself; links expire after 72 hours.

How do I share an app I built?

Click Publish in the app to make the latest version live. The web app link is on the app's Access Point tab, which also has the API details for calling the app with an API key, and an MCP Server card that turns the app into an MCP server for tools such as Claude Desktop and Cursor. Each publish replaces the live version right away.

Where is my data kept?

On your server, in Docker volumes that survive container updates. One PostgreSQL server with pgvector holds Dify's database, the knowledge-base vectors and the plugin daemon's own database; uploaded files sit in the storage volume and installed plugins in their own volume. The provider API keys you add are encrypted with the generated SECRET_KEY, so never change it. What you send in a chat, with the passages retrieved for it, goes to the model provider you choose.

Dify on your own server, in one click.

Build chatbots, agents, chatflows and workflows on a visual canvas with Dify, on a server you own. ServerAvatar runs Dify's core stack in Docker with its own PostgreSQL, pgvector and Redis, generates the secrets and turns HTTPS on for you.

Dify + PostgreSQL + Redis
HTTPS on its own
Your server, your data

Deploy Dify See how it installs

Section 01What is Dify

What is Dify?

Dify is an open-source platform for building apps on top of large language models. You build chatbots, agents, chatflows and workflows in the browser, mostly by connecting nodes on a visual canvas, ground them in your own documents with knowledge bases, and publish them as a web app, an API or an MCP server.

The recipe runs the core services from Dify's official Docker Compose file for release 1.17.1: the API, a background worker and scheduler, the web console, a code sandbox, the plugin daemon, a Squid proxy, PostgreSQL with pgvector as both database and vector store, and Redis. Dify runs no model itself: you install a model provider such as OpenAI, Anthropic or Ollama from the Dify Marketplace and add its API key in Dify.

Open source, under a modified Apache 2.0 licence
Workflows, chatflows, chatbots and agents in one app
Knowledge bases stored in PostgreSQL with pgvector

Section 02Features

What you can do with Dify

Workflows and chatflows on a canvas
Connect nodes for model calls, knowledge retrieval, conditions, loops, code and HTTP requests. A workflow runs once from input to result and can also start on a schedule or from a webhook; a chatflow runs behind each chat message.
Chatbots and agents
Build a chatbot from a model and a prompt, or an agent that reasons with Function Calling or ReAct and calls tools from plugins, MCP servers or your own OpenAPI services.
Knowledge bases
Upload PDFs, presentations and other documents, or import pages from Notion or a website. Dify splits them into chunks, indexes them with an embedding model and retrieves the relevant passages for your apps.
Code in a sandbox
Code nodes run Python or JavaScript in Dify's sandbox container. It has no direct route to the internet: its requests go through a Squid proxy that refuses private addresses.
Publish as a web app, API or MCP server
Click Publish to get a web app link to share, call the app through its API with an API key, or turn on its MCP server for tools such as Claude Desktop and Cursor.
Logs and annotations
The Logs tab shows every conversation or run with its token spend, latency and a node-by-node trace. Rate answers as a team and save better replies as annotations.

Section 03Requirements

What you need

A fresh Ubuntu server you can log in to as root. The installer checks all of this before it changes anything.

Dify on Docker
Requirement	Dify on Docker
Operating system	Ubuntu 24.04 or 26.04 LTS (x86-64)
CPU	2 cores or more
RAM	4 GB or more
Disk	20 GB or more (25 GB recommended)
Containers	Dify 1.17 + PostgreSQL 16 (pgvector) + Redis 6.2
Email	Optional: SMTP, Resend or SendGrid, for invitation emails
Ports	80 and 443 open

CPU and RAM are the minimum system requirements in Dify's Docker Compose guide (2 cores and 4 GiB). Disk: about 15.5 GB is in use once the panel and Dify are installed on our test server; 25 GB is Ubuntu's recommendation.

Section 04Install

How to install Dify

Every step happens in the ServerAvatar panel. No terminal needed.

Pick Dify in 1-Click Recipes
On your server, open Applications, click Create application and search for Dify under 1-Click Recipes.
Name it and choose a domain
Give the app a name and use your own domain, or a temporary one that works right away. The panel creates a system user for it.
Create it
Click Create application. The panel starts Dify's ten containers: nginx, api, worker, worker_beat, web, the code sandbox, the plugin daemon, the Squid proxy, PostgreSQL and Redis. It turns HTTPS on and shows the secrets it generated; copy them somewhere safe. On our test servers, Dify was running four to five minutes after the click.
Create your admin account right away
Open the app's domain: Dify sends you to /install. Enter an email address, a username and a password, then click Set up; that account becomes the workspace owner. Do it right after the install, because until then anyone who reaches the site can claim it. Then add a model provider in Integrations, Model Provider.

Section 05FAQ

Questions,answered.

Can't find the answer you're looking for? Reach out to our customer support.

Contact support

Risk free

Deploy your first appin 10 minutes.

Connect a server, pick an application and go live, with SSL and backups already in place.

Start free Plans & pricing

No credit card required
Free 4-day trial
Deploy in the next 10 minutes

Dify on your own server, in one click.

What is Dify?

What you can do with Dify

Workflows and chatflows on a canvas

Chatbots and agents

Knowledge bases

Code in a sandbox

Publish as a web app, API or MCP server

Logs and annotations

What you need

How to install Dify

Pick Dify in 1-Click Recipes

Name it and choose a domain

Create it

Create your admin account right away