Self-Hosted AI Agent Stack on One VPS (2026): n8n, Ollama, Open WebUI, Postgres and Caddy

What you are building

Total without local models: about 3–4 GB. With a 4-bit 7B–8B model: about 10–12 GB. That is why we suggest 8 GB for API-only and 16 GB with Ollama (see the VPS calculator).

Pick a server

Choose Ubuntu 24.04, add an SSH key, and point two DNS records (for example n8n.example.com and chat.example.com) at the server IP. New to servers? Start with cheap VPS for beginners.

The Docker Compose file

Install Docker (curl -fsSL https://get.docker.com | sh), create a folder, and save this as compose.yaml:

services:
  caddy:
    image: caddy:2
    restart: unless-stopped
    ports: ["80:80", "443:443"]
    volumes:
      - ./Caddyfile:/etc/caddy/Caddyfile
      - caddy_data:/data
  postgres:
    image: postgres:16
    restart: unless-stopped
    environment:
      POSTGRES_USER: n8n
      POSTGRES_PASSWORD: ${PG_PASSWORD}
      POSTGRES_DB: n8n
    volumes: [pg_data:/var/lib/postgresql/data]
  n8n:
    image: docker.n8n.io/n8nio/n8n
    restart: unless-stopped
    environment:
      DB_TYPE: postgresdb
      DB_POSTGRESDB_HOST: postgres
      DB_POSTGRESDB_USER: n8n
      DB_POSTGRESDB_PASSWORD: ${PG_PASSWORD}
      DB_POSTGRESDB_DATABASE: n8n
      N8N_HOST: n8n.example.com
      N8N_PROTOCOL: https
      WEBHOOK_URL: https://n8n.example.com/
      N8N_ENCRYPTION_KEY: ${N8N_KEY}
      EXECUTIONS_DATA_PRUNE: "true"
      EXECUTIONS_DATA_MAX_AGE: "168"
    volumes: [n8n_data:/home/node/.n8n]
    depends_on: [postgres]
  ollama:
    image: ollama/ollama
    restart: unless-stopped
    volumes: [ollama:/root/.ollama]
  openwebui:
    image: ghcr.io/open-webui/open-webui:main
    restart: unless-stopped
    environment:
      OLLAMA_BASE_URL: http://ollama:11434
    volumes: [openwebui:/app/backend/data]
    depends_on: [ollama]
volumes: {caddy_data: {}, pg_data: {}, n8n_data: {}, ollama: {}, openwebui: {}}

And a Caddyfile:

n8n.example.com {
  reverse_proxy n8n:5678
}
chat.example.com {
  reverse_proxy openwebui:8080
}

Create .env with PG_PASSWORD= and N8N_KEY= set to long random strings (openssl rand -hex 32), then run docker compose up -d. Pull a model with docker compose exec ollama ollama pull llama3.1:8b. In n8n, use the Ollama credential with base URL http://ollama:11434 to call the local model from AI agent nodes.

Image names and variables follow the projects' official docs as of October 2026; check each project's documentation for changes before deploying.

Security checklist

• Only Caddy publishes ports. Ollama's API has no authentication—never map 11434 to the host's public interface. Same for Postgres 5432.
• Create accounts immediately. The first user to open n8n or Open WebUI becomes the owner/admin—do it right after docker compose up.
• Firewall: allow 22, 80, 443 only. Use SSH keys and disable password login.
• Secrets: keep .env at chmod 600; never commit it.
• Updates: docker compose pull && docker compose up -d monthly; read release notes first.
• API spending limits on any hosted model provider your agents call.

Backups

1. Back up .env—especially N8N_KEY. Without it, n8n cannot decrypt saved credentials after a rebuild.
2. Dump Postgres nightly: docker compose exec -T postgres pg_dump -U n8n n8n | gzip > n8n-$(date +%F).sql.gz.
3. Copy dumps off the server (object storage, Google Drive via rclone, or your PC).
4. Take a provider snapshot before upgrades. Hostinger VPS includes weekly backups; DigitalOcean and Vultr charge extra.

Monthly cost examples

All server prices are published list prices checked 21 Oct 2026; renewals and taxes differ—model them in the true cost calculator.

Related: best VPS for AI agents · Ollama VPS requirements · n8n VPS requirements · host OpenClaw on a VPS.