2026.10.15
Selfhost AI: A Single Deployment for AI and Automation Services
Selfhost AI uses Docker Compose and an interactive wizard to deploy self-hosted AI, workflow, RAG, and monitoring services. It suits users who want a unified server environment, but requires a VPS, domain, DNS setup, and ongoing operations.
What it does
Selfhost AI is a Docker Compose installer for a self-hosted AI environment. It is not just a chat application: it brings services such as n8n, Ollama, Open WebUI, OpenClaw, Dify, Flowise, Supabase, ComfyUI, and Qdrant into one deployment flow. The wizard lets users select services, generate configuration secrets, and place services behind Caddy with HTTPS and subdomain-based access.
Who it suits
The project is aimed at developers, independent builders, technical teams, and homelab users who already have a server and domain and want workflow automation, local models, agents, and RAG components in one environment. n8n covers workflow orchestration, Ollama runs local models, and Open WebUI provides a chat interface. Dify, Flowise, Letta, and OpenClaw address agent and AI application use cases. Grafana and Prometheus are available for monitoring, while Qdrant and Weaviate target vector search.
If you only need one chat interface, or do not want to manage Docker, DNS, certificates, backups, and updates, the project may be broader than necessary. It is better understood as a selectable service stack than as a single-purpose application.
Documented capabilities
According to the README, the interactive wizard controls service selection. Caddy, PostgreSQL, and Valkey are always installed, while the other services are optional. The README also lists more than 400 n8n integrations, queue mode, and an MCP server. Open WebUI can work with local and API models. Ollama configurations include CPU, NVIDIA, and AMD GPU options, with multi-GPU support mentioned in the project documentation; suitability for a particular machine still needs to be checked against the relevant configuration and upstream documentation.
Services are generally reached through subdomains. The project documents health checks, service dependencies, make doctor diagnostics, and updates intended to preserve settings. Release v1.17.0 added make pull and improved retry behavior for image-pull failures. However, the release notes also show that update paths differ by service: locally built n8n, its runners, and OpenClaw are not updated by make pull.
Limits and deployment considerations
The README requires a VPS with a public IP, Ubuntu 24.04 LTS, a domain, and a wildcard A record pointing to the server. Its reference sizing starts at 4 GB RAM, two CPU cores, and 40 GB of disk for n8n, monitoring, Databasus, and Portainer. Running all services is documented as requiring at least 20 GB RAM, four CPU cores, and 60 GB of disk. These are project guidance figures, not performance guarantees for every service combination.
Self-hosting does not remove operational responsibility. You still need to manage DNS, certificates, secrets, backups, access control, logs, image updates, and exposure. The README calls out security notes for Supabase ports, and some components can use external API keys or integrations such as Telegram. Before enabling agents with server access, code execution sandboxes, or public endpoints, review permissions and network boundaries. The repository is marked Apache License 2.0, but included upstream services may have their own licenses and operational requirements; those should be checked separately before adoption.
Sources
Sources checked: 2026-10-15