Back to recipes
Self-HostingIntermediate30 minVerified 31 days ago

Self-Host Dify with Docker Compose

Deploy Dify — the open-source LLM app development platform — as a Docker Compose stack with Postgres, Redis, and a vector database for a self-hosted agent-building environment.

dockerdocker-composedifyself-hostingllm-platformagents
What you're deploying

Dify is an open-source platform for building LLM applications and agents — visual workflow builder, RAG pipeline, model routing, and agent orchestration in one stack. It ships as a Docker Compose stack of about half a dozen containers, including a Postgres database, a Redis cache, and a vector database. This recipe gets it running on your own infrastructure.

Prerequisites
  • Linux server with at least 4 GB RAM (2 GB minimum, 4 GB recommended)
  • Docker 20.10+ and Docker Compose v2+
  • A domain name with DNS pointing to your server (for TLS)
  • At least one LLM API key (OpenAI, Anthropic, or any OpenAI-compatible endpoint)
Maintenance
# Update Dify
cd dify
git pull
cd docker
docker compose pull
docker compose up -d

# Back up the database
docker compose exec postgres pg_dump -U postgres dify > dify_backup_$(date +%Y%m%d).sql
Security checklist
  • TLS reverse proxy in front (not plain HTTP on public interface)
  • Admin account created immediately after first boot
  • Strong SECRET_KEY generated (not the default)
  • Firewall restricts access to Postgres, Redis, and vector DB ports (internal only)
  • API keys stored in environment variables, not committed to git
  • Regular database backups configured
  • Docker images kept updated (docker compose pull regularly)
When to choose dify over alternatives
  • Choose Dify when you want a visual builder for LLM apps and agents with RAG, model routing, and a UI — without writing orchestration code
  • Choose LangGraph when you want code-level control over agent state and orchestration
  • Choose n8n when your workflow is integration-first (connecting SaaS tools) rather than LLM-first
  • Choose Open WebUI when you primarily want a ChatGPT-like interface for local models without app-building features
Resource notes

Dify's full stack with a vector database and sandbox needs at least 2 GB free RAM to start, ideally 4 GB for production use. If you're running local model inference (Ollama/vLLM) alongside Dify on the same host, plan for significantly more — the model server will be the dominant memory consumer.

Steps

1
git clone https://github.com/langgenius/dify.git
cd dify/docker
2
cp .env.example .env

Edit .env with critical settings:

# Generate a strong secret key
SECRET_KEY=$(openssl rand -hex 32)

# Set your domain
CONSOLE_API_URL=https://your-domain.com
APP_API_URL=https://your-domain.com

# Configure the built-in vector database (or point to external)
VECTOR_STORE=weaviate

# Set a strong initial admin password
INIT_PASSWORD=your-strong-password-here
3
docker compose up -d

This brings up:

  • Dify API (backend)
  • Dify Web (frontend)
  • PostgreSQL (metadata and session storage)
  • Redis (caching and queue)
  • Weaviate or your configured vector database
  • Sandbox (code execution for agent tools)
  • SSRF Proxy (safe outbound requests for RAG ingestion)
4

Dify's bundled web server listens on plain HTTP port 80 on every interface. Do not expose this directly.

Bind to loopback and add a TLS reverse proxy

Edit your docker-compose.yaml to bind the web service to 127.0.0.1:

services:
  nginx:
    ports:
      - "127.0.0.1:80:80"

Then put Caddy or Nginx in front with TLS:

# Using Caddy (automatic TLS via Let's Encrypt)
# Caddyfile
your-domain.com {
    reverse_proxy 127.0.0.1:80
}

caddy run --config Caddyfile

Create the admin account immediately

The first visitor to Dify's install page claims the admin account. After bringing the stack up, visit https://your-domain.com/install immediately to create your admin account before anyone else does.

5
  1. Log in to the Dify console
  2. Go to Settings → Model Provider
  3. Add your API key for your chosen provider (OpenAI, Anthropic, Ollama, OpenRouter, etc.)
  4. For local models, point Dify at your Ollama or vLLM endpoint
6
# Check all containers are healthy
docker compose ps

# Test the API
curl https://your-domain.com/console/api/setup

Recipe verified 2026-09-27. Commands are tested but your environment may differ.

Browse related services