Selfhost AI

A self-hosted AI stack installer brings Ollama, Open WebUI and n8n to an Ubuntu server with Docker Compose and automatic HTTPS.

Selfhost AI packages AI chat, agents and workflow automation for people who want these services on a server they control. You choose which tools to include rather than deploy the whole collection.

Local models and chat: Ollama runs Llama, Qwen, Mistral and Gemma on CPU or NVIDIA and AMD GPUs, including multiple GPUs; Open WebUI provides chat with local or external API models.
Workflow automation: n8n connects applications and AI tasks, while n8n-MCP gives coding assistants such as Claude Code and Cursor access to workflow tools.
Agents and document search: Dify and Keelflow support visual agent building, Letta provides persistent agent memory, and RAGFlow supports document-based answers with citations.
Image generation: ComfyUI supports Stable Diffusion and Flux workflows, while InvokeAI includes inpainting and a canvas editor.
Monitoring and access: Grafana and Prometheus provide system and workflow dashboards and alerts, and Caddy handles automatic HTTPS.

The installer is free and open source under Apache 2.0 and builds on local-ai-packaged and n8n's Self-hosted AI Starter Kit. It targets an Ubuntu 24.04 LTS VPS with a public IP and domain; home servers and localhost deployments aren't supported. The full stack requires at least 20 GB RAM, four CPU cores and 60 GB disk.

Local model processing stays on your server; external API models send requests to their providers. The installer sends anonymous usage statistics through Scarf, with an opt-out.