Docker AI Stack, self-hosted LLM/STT/TTS/MCP in one compose file
One-command local AI stack bundling Ollama, Whisper, and MCP Gateway.
Docker image to run a Headscale server, a self-hosted, open-source Tailscale-compatible mesh VPN coordination server. Auto-generates config and pre-auth key on first start. MagicDNS support, helper script for user/node management. Multi-arch: amd64, arm64.
Convenience wrapper for Headscale when the official Docker image already exists.
Self-hosters and homelab enthusiasts
Headscale official image · Tailscale · ZeroTier
On first start it:
* Generates config * Creates a pre-auth key. * Allows a client to connect within a few minutes
For non-Docker setups, there is also a simple install script: hwdsl2/headscale-install
That script is intended for a one-command VPS setup. It installs and configures Headscale with sensible defaults.
This follows the same approach as some of my earlier projects (e.g. docker-ipsec-vpn-server, setup-ipsec-vpn), focusing on making things work out of the box with minimal maintenance.
I’ve also been experimenting with related projects in the same space:
* docker-wireguard * docker-openvpn
Feedback is welcome, especially from others running Headscale.
One-command local AI stack bundling Ollama, Whisper, and MCP Gateway.
Docker sandboxing for agents when most runners just trust whatever the LLM outputs.
OpenClaw but in a container—fixes security by default, ships Docker isolation instead of promises.
Config-as-code uptime monitoring beats dashboard vendor lock-in with git-driven workflows.
Outbound-only Pilot Agent beats Portainer for NAT traversal without exposing Docker socket.
Portainer alternative keeping compose files on disk with outbound-only agent connectivity.