How to Use Dify: What It Can Do and How to Build Your First App

This is an introduction for people looking into how to use Dify for the first time. Terms follow the English wording of the official documentation, and screen names are given as they appear in the interface.

Prices, limits and steps were checked on the official pages on October 2, 2026. Dify is updated often. Check the sources at the end for the latest details before you use it.

What Dify is: a tool for building AI apps on screen

Dify is an open source platform for building AI apps. It is developed by LangGenius, and its source code is published on GitHub.

You work in a screen called Studio. You drag parts called nodes onto the canvas and connect them with lines to build a process. What you build can be published as an app.

Inside a workspace, you can do four things: build, give apps your own data, publish, and watch how apps are used. A team can share one workspace.

What Dify can do: the types of apps you can build

What Dify can do is easiest to see through the app types. When you create a new app, you choose a type first. The official documentation recommends choosing Workflow or Chatflow.

• Workflow: takes an input, processes it once from start to finish and returns the result
• Chatflow: runs the flow on every message and replies. It can hold per-conversation variables
• Chatbot: the simplest conversational app, built with just a model and a prompt
• Agent: a conversational app in which the model reasons on its own and uses tools
• Text Generator: an app that writes text from a single input

The last three are called basic apps and are built in an older, simpler screen. Under the hood, all of them run on the same workflow engine.

Besides app types, Dify offers the following features.

• Knowledge: import internal documents and let apps search and cite them
• Publishing: release as a web app, a website embed, an API or an MCP server
• Monitoring: review usage and logs and improve answers
• Integrations: connect model providers, tools and data sources
• DSL: export an app to a YAML file and import it into another environment

Getting started: Dify Cloud or self-hosted

There are two ways to start with Dify. Use the cloud version run by Dify (Dify Cloud), or install it on your own server and run it yourself (self-hosted).

Getting started with Dify Cloud

Sign up for an account at cloud.dify.ai. New accounts start on the Sandbox plan. Sandbox comes with 200 credits for calling models from OpenAI, Anthropic, Gemini and others. These credits are granted once and do not renew each month.

On Sandbox, logs are kept for 30 days. Sandbox is enough for trying things out. Consider a higher plan when you start having other people use your apps.

Getting started with self-hosting

The version you run on your own server is the Community Edition. The official steps list at least 2 CPU cores and 4GiB of memory as the minimum. You need Docker and Docker Compose version 2.24.0 or later.

1. Get the Dify source code from GitHub
2. Move into the dify/docker folder
3. Copy .env.example to create .env
4. Start the containers with docker compose up -d
5. Open the /install page in a browser and create an admin account

When self-hosting, you handle updates, backups and incident response yourself. Use is governed by the Dify Open Source License, which is based on Apache 2.0 with added conditions.

Your first app: build and publish a chatbot

For a first app, a Chatbot takes the fewest steps. It runs once you set a prompt and a model. A Chatbot keeps up to 500 messages or 2,000 tokens of history per conversation. Beyond that, the oldest items are dropped.

How to write the prompt

The prompt sets how the model behaves throughout the conversation. The official documentation recommends covering three points.

• Role: who it answers as, and in what tone
• Output format: length, structure and style
• Constraints: what to avoid and which rules to follow

To change answers depending on the user, use variables. For example, if you make job title or language a variable, an input field appears before the conversation. A short-text variable holds up to 256 characters. If you are unsure what to write, you can use the Generate button to have AI draft it.

Test before you publish

In the preview on the right side of the screen, send real messages and check the answers. With Debug as Multiple Models, you can compare answers from up to 4 models side by side.

When you are satisfied, click Publish. The web app URL appears on the Access Point tab. Send the link, and others can use it right away. Publishing replaces the running app with the current configuration.

Model settings and API keys

Every Dify app runs on an AI model. You add model providers under Model Provider in Integrations. Only the workspace owner and admins can manage them.

• Use credits: add a supported provider and use it without an API key. Each model call counts as one, and the amount deducted varies by model
• Use your own API key: click Setup on the provider card and enter your key. Billing goes to your own provider account
• Use both: Usage Priority sets whether your key or the credits are used first

In Default Models, you set the standard models used by apps and nodes that do not specify one. Besides the model that writes text, you also set models such as the embedding model used for knowledge search.

When an app uses API keys for other services, store them in environment variables. Environment variables are stored outside the app, so the keys are not included when you share the app as a DSL file.

What to learn next: Knowledge and Workflows

Once your Chatbot is running, learning the following two features widens what you can build.

• Knowledge: import internal manuals and FAQs and have the app answer based on them
• Workflow and Chatflow: combine search, conditional branching, code execution and more with nodes

Building knowledge is covered in How to Build Dify Knowledge, and the difference between the two app types in Dify Chatflow vs. Workflow. How Dify differs from n8n and GPTs is summarized in n8n vs. Dify vs. GPTs.

List your app on Employee Store

Employee Store is a marketplace where companies adopt AI agents built by developers, through a one-time purchase or a monthly plan. Any format is accepted, including Dify Workflows. One person (or company) can list several AIs.

There is no listing fee and no upfront cost. The commission is 20% of the deal amount and applies only when a deal closes. See the seller guide for details.