Installing Dify AI on VPS – Free No-code AI Platform

If you’re looking for a platform to build AI applications without much coding, Dify is the most noteworthy name right now. This is an open-source platform with over 80,000 stars on GitHub and has been officially included in the Linux Foundation’s project catalog. In other words, this isn’t just someone’s hobby project, but a serious platform with a community and organization behind it.

Dify allows you to create chatbots, AI agents, RAG pipelines, and automation workflows using a drag-and-drop interface. You connect models (OpenAI, Anthropic, Ollama local, etc.), design processing flows, then publish them as API endpoints or web apps. The entire process requires almost no coding.

In this article, I’ll guide you through installing Dify on a private VPS using Docker Compose, connecting it to Ollama for running local models, and creating your first chatbot.

Dify vs Flowise vs Langflow

Before diving into installation, you might wonder why choose Dify over Flowise or Langflow. All three are no-code AI platforms, but their approaches differ quite clearly.

Criteria	Dify	Flowise	Langflow
GitHub Stars	80k+	35k+	50k+
Backed by	Linux Foundation	Community	DataStax
Interface	Workflow builder + Chat UI	Flow builder	Flow builder
RAG integration	Yes (Knowledge base)	Yes	Yes
Agent support	Yes (multi-tool agents)	Yes	Yes
API endpoint	Auto-generated from workflow	Yes	Yes
Observability	Built-in (logs, traces)	Limited	Via LangSmith
Installation complexity	Docker Compose (easy)	npm/Docker (easy)	Docker/pip (medium)
Best suited for	Production apps, teams	Quick prototyping	LangChain users

In summary: Flowise is lightweight and fast for prototyping. Langflow is powerful if you’re already familiar with LangChain. Dify is more comprehensive for production, with observability, user management, and ready-made API endpoints. If you want a platform that’s “ready to use after installation” for teams or real projects, Dify is a reasonable choice.

VPS Requirements

Dify runs multiple containers (web, API, worker, database, Redis, vector store), so it needs a VPS with at least the following configuration:

• CPU: 2 cores or more
• RAM: 4GB or more (8GB recommended if running Ollama)
• Disk: 40GB SSD or more
• OS: Ubuntu 22.04/24.04 or Debian 12
• Docker & Docker Compose: pre-installed

If Docker isn’t installed yet, run quickly: curl -fsSL https://get.docker.com | sh then add the user to the docker group: sudo usermod -aG docker $USER. Logout and login again to use.

Installing Dify with Docker Compose

The Dify installation process is quite simple, just a few commands and you’re done.

Step 1: Clone the repository

git clone https://github.com/langgenius/dify.git

Step 2: Create .env file and start

cd dify/docker
cp .env.example .env
docker compose up -d

The first run will take a few minutes to pull images. Dify will initialize containers: nginx, web, api, worker, db (PostgreSQL), redis, weaviate (vector store), and sandbox.

Check status:

docker compose ps

All containers must be in Up or healthy status. If any container keeps restarting, check logs with docker compose logs [container-name].

First Access

Open your browser and access http://VPS-IP:80. You’ll see the initial setup page requiring admin account creation.

• Enter email, name, and password for the admin account
• Log in to the dashboard

The Dify interface is quite intuitive: on the left is the main menu with sections for Studio, Knowledge, Tools, and Explore. You’ll mainly work in Studio to create AI applications.

If you want to access via domain instead of IP, point the domain to your VPS and configure a reverse proxy (Nginx or Caddy) with SSL. The .env file has the APP_WEB_URL variable for you to update the URL.

Connecting Ollama for Local Models

One of Dify’s strengths is supporting many model providers. You can use OpenAI, Anthropic, Google Gemini, or run local models via Ollama.

If you’ve installed Ollama on the same VPS, connect as follows:

1. Go to Settings (top right corner) → Model Provider
2. Find Ollama in the provider list
3. Enter URL: http://host.docker.internal:11434
4. Enter model name (e.g., llama3.2, qwen2.5, gemma3)
5. Select model type (LLM or Text Embedding) and save

host.docker.internal is a special address that allows Docker containers to access the host machine. On Linux, you need to add extra_hosts: ["host.docker.internal:host-gateway"] to docker-compose.yml if not present, or use the VPS’s internal IP directly.

After successful connection, you’ll see Ollama models appear in the list when creating new applications.

Creating Your First Chatbot

Now for the fun part. Go to Studio → Create from Blank → choose Chatflow or Workflow.

With Chatflow, you’ll see a canvas with drag-and-drop nodes:

• Start: receives input from user
• LLM: calls AI model (choose the model connected in previous step)
• Answer: returns result to user

The simplest flow: Start → LLM → Answer. You write the system prompt in the LLM node, select the model, then click Preview in the top right to test directly on the interface.

Want something more advanced? Add a Knowledge Retrieval node for the chatbot to query private documents (RAG), or add IF/ELSE, Code, HTTP Request to handle complex logic.

Key Features of Dify

After installation and creating your first chatbot, you’ll gradually discover more features that Dify provides:

RAG (Retrieval-Augmented Generation)

Go to Knowledge section, upload PDF, TXT, Markdown files, or crawl websites. Dify will automatically chunk, embed, and store in the vector store. When used in workflows, the chatbot will search through documents to provide more accurate answers instead of “hallucinating” responses.

Agents

Dify supports creating AI Agents capable of using tools: web search, code execution, external API calls, database queries. You configure tools in the Tools section, then attach them to Agent nodes in workflows.

Observability

Every user interaction is logged completely: input, output, token usage, latency, each step in the workflow. You can view this in the Logs section of each app. Very useful for debugging and prompt optimization.

Creating API Endpoints from Workflows

This is a feature I find very convenient. When you finish creating a workflow or chatbot, click Publish. Dify will automatically generate an API endpoint for that app.

In the app’s API Access section, you’ll see:

• API base URL
• API key for authentication
• API documentation with curl examples

Example API call with curl:

curl -X POST 'http://VPS-IP/v1/chat-messages' \
  -H 'Authorization: Bearer app-xxxxx' \
  -H 'Content-Type: application/json' \
  -d '{
    "inputs": {},
    "query": "Hello",
    "response_mode": "blocking",
    "user": "user-123"
  }'

With this feature, you can turn any workflow into an API service. Integrate into websites, mobile apps, Telegram bots, or any system that needs to call AI.

Updating Dify

Dify releases new versions quite frequently (about every 1-2 weeks). To update:

cd dify/docker
git pull origin main
docker compose pull
docker compose up -d

Before updating, consider backing up the database: docker compose exec db pg_dump -U postgres dify > backup.sql. In case new versions have breaking changes.

If there are changes in the .env.example file (new variables), you need to manually update your .env file. Dify usually clearly notes in release notes what changes need attention.

Conclusion

Dify is one of the best no-code AI platforms currently available for self-hosting. Simple installation with Docker Compose, intuitive interface, support for multiple model providers, and ready-made production features like RAG, Agents, API endpoints, and observability.

If you want to build AI applications for personal projects or teams without complete dependence on SaaS platforms, Dify on a private VPS is a very worthwhile option. You control the data, choose models as desired, and scale according to needs.

Good luck with your installation!