selfhost-ai

Selfhost AI — One-Command Self-Hosted AI Stack (n8n, Ollama, Open WebUI, OpenClaw)

[GitHub stars] [Latest release] [License] [Ubuntu]

Deploy 30+ open-source AI and automation tools on your own server with a single command. Selfhost AI is a Docker Compose installer for a private AI homelab: n8n workflow automation, local LLMs with Ollama, a ChatGPT-style chat in Open WebUI, AI agents (OpenClaw, Flowise, Keelflow, Dify, Letta), RAG engines and vector databases (Qdrant, Weaviate, LightRAG, RAGFlow), Supabase, ComfyUI and Grafana monitoring. An interactive wizard picks the services, generates every secret and puts them all behind Caddy with automatic HTTPS. A free, self-hosted alternative to Zapier, Make and ChatGPT.

Quick Start

You need a VPS with a public IP running Ubuntu 24.04 LTS (4 GB RAM / 2 CPU for n8n and monitoring; 20 GB RAM / 4 CPU / 60 GB disk for everything), a domain, and a wildcard DNS record A *.yourdomain.com -> YOUR_SERVER_IP. Then run:

git clone https://github.com/kossakovsky/selfhost-ai && cd selfhost-ai && sudo bash ./scripts/install.sh

The wizard does the rest; see Installation for what it asks.

Key Features

• One-command install: an interactive wizard, generated secrets and no manual configuration
• Private AI homelab: run LLMs locally with Ollama, on CPU, NVIDIA or AMD GPUs, including multi-GPU; your data stays on your server
• ChatGPT alternative: Open WebUI for local and API models, with an optional code-execution sandbox
• Workflow automation: n8n with 400+ integrations, queue mode with scalable workers, an MCP server for AI IDEs, and 300+ optional community workflows
• AI agents and RAG: OpenClaw (with Telegram), Flowise, Keelflow, Dify, Letta, LightRAG, RAGFlow, Qdrant, Weaviate
• Automatic HTTPS: Caddy reverse proxy with Let's Encrypt; services are reached only through Caddy (see security notes for Supabase's ports); optional Cloudflare Tunnel
• Built-in monitoring: Grafana and Prometheus with an n8n workflow dashboard and ready-made alerts
• Production ready: health checks, service dependencies, make doctor diagnostics, and updates that preserve your settings
• Free and open source: Apache 2.0, no vendor lock-in

What's Included

Always installed: Caddy (reverse proxy and HTTPS), PostgreSQL, and Valkey (Redis-compatible). Everything below is optional and chosen in the wizard. "internal" means the service has no public URL and is reachable only from other containers.

AI Chat and Agents

Workflow Automation

Local LLMs and Image Generation

RAG, Vector and Graph Databases

Documents, OCR, Web Scraping and Search

Low-Code Apps and Data

Monitoring and Operations

URLs are subdomains of your domain, for example n8n.yourdomain.com.

Installation

Prerequisites

1. A domain name with a wildcard A record, *.yourdomain.com, pointing to your server's IP, configured before installing.
2. A VPS with a public IP address. Home servers, shared hosting and localhost setups are not supported.
   • Operating system: Ubuntu 24.04 LTS, 64-bit
   • n8n, Monitoring, Databasus and Portainer: 4 GB RAM / 2 CPU cores / 40 GB disk
   • All services: at least 20 GB RAM / 4 CPU cores / 60 GB disk

Run the installer over SSH:

git clone https://github.com/kossakovsky/selfhost-ai && cd selfhost-ai && sudo bash ./scripts/install.sh

The installer updates the system, configures the firewall and brute-force protection, installs Docker, generates .env with all secrets, and starts the services. It asks for:

1. Your domain (e.g. yourdomain.com).
2. Your email, used for service logins and for the Let's Encrypt certificate.
3. Which services to install, in a checklist wizard, plus follow-ups for some of them (hardware profile, Telegram bot token, Cloudflare token, ...).
4. An optional OpenAI API key (used by Supabase AI and Crawl4AI).
5. If n8n is selected: whether to import ~300 community n8n workflows (takes 20-30 minutes) and the number of n8n workers (each gets its own task runner).

At the end it prints the Welcome Page URL and login (save them). The Welcome dashboard at welcome.yourdomain.com lists every service's URL and credentials.

Upgrading

make update     # pull the latest installer, update images, restart
make git-pull   # for forks: merge from upstream instead of resetting

make update resets tracked files such as Caddyfile and docker-compose.yml, so put your customizations in the places that survive updates:

• Custom Caddy sites: caddy-addon/site-*.conf (see caddy-addon/README.md)
• Compose overrides: a docker-compose.override.yml in the project root; it takes the highest precedence
• Settings: values in .env are kept (except GOST_NO_PROXY, which is regenerated to cover new services)

Commands

Run make help for the full list.

Guides

• n8n: MCP server, Code node libraries, ffmpeg, file access and workflow monitoring
• n8n Assistant sandbox and Sysbox
• OpenClaw: AI agent with Telegram and server access
• Open Terminal for Open WebUI
• Keelflow: Moving from Flowise
• Ollama on multi-GPU hosts
• Open WebUI: SQLite or PostgreSQL
• Security notes: published ports and the firewall
• Cloudflare Tunnel
• Caddy add-ons

Troubleshooting

• Sites don't load. The server is usually out of RAM or CPU for the services you selected. Check free -h and make monitor, start with n8n only, and enable other services one by one. make doctor checks DNS, certificates and containers.

• "Dangerous site" warning right after install. Caddy briefly serves a temporary certificate while it gets one from Let's Encrypt. This normally clears within hours; if it lasts more than a day, check make logs s=caddy and your DNS records.

• Docker images fail to pull. Temporarily disable your VPN. make update retries a failed pull one service at a time. Services that still could not be updated but have an image on the server are started from it and listed at the end; finish them with make pull && make restart. If a service has no image at all, or the local n8n/OpenClaw images cannot be built, make update stops before restarting anything; run make pull, then make update again.

• make update fails. Force-sync with upstream. This discards local changes to tracked files:

  git config pull.rebase true && git fetch origin && git checkout main && git reset --hard "origin/main" && make update
  

Credits

• Based on local-ai-packaged by coleam00 and the n8n team's Self-hosted AI Starter Kit
• Community port: n8n-installer-arch by @ndrewpj for Arch, CachyOS and Manjaro

Telemetry

The installer sends anonymous usage statistics through Scarf: event type, installer version, selected services, OS, a random installation ID, and the country Scarf derives from your IP. No personal data is collected. To opt out, set SCARF_ANALYTICS=false in .env.

License

Apache License 2.0. See LICENSE.