VMtree is a fast and easy way to provision self-hosted ephemeral VMs

VMtree, a lightweight self-hosted tool written in Bash, lets developers provision ephemeral VMs instantly by SSH-ing into a new subdomain, such as `ssh demo-foo.example.com`, and uses LXD/Incus containers or QEMU VMs for high density or full isolation. Developed by a DevOps team for their own use over 8+ years, it offers automatic HTTPS via Caddy, a shared `/persist/` directory, and nightly auto-cleanup, positioning itself as a free, self-hosted alternative to GitPod or GitHub Codespaces.

TL;DR: A lightweight, self-hosted alternative to GitPod or GitHub Codespaces. Provision a fresh VM instantly just by SSH-ing into a new subdomain e.g., ssh demo-foo.example.com . - 🚀 Instant Provisioning: ssh <name .example.com creates and connects you to a new VM. - 🌐 Automatic Networking: Every VM gets a dedicated HTTPS subdomain via Caddy. - 💾 Smart Storage: Shared /persist/ directory for data that survives VM resets. - 🧹 Auto-Cleanup: VMs are ephemeral and automatically destroyed nightly to save resources. - 🛠️ Pure Simplicity: Written in Bash; uses LXD/Incus containers high density or QEMU VMs full isolation . - 🤖 Not post-AI slop: It's how our DevOps team survived COVID's years of WFH. 😷 These scripts turn a server or VM into a bountiful 🌳 "VM tree", on which you can easily start up ephemeral VMs. Originally for our self-hosted cloud development environments, but usable in many ways: in a CI/CD pipeline , as acceptance testing environments , for demoing purposes , or really anything you can use an Ubuntu environment for. We didn't develop this as a product, but to scratch our own itch as a DevOps team. We've been using this and its previous in-house version for 8+ years as our dev environments, and we ❤️ that fresh VMs just "grow on the VM tree for easy picking" . It's literally how our team survived COVID's years of WFH. : It's the same philosophy as GitPod https://www.gitpod.io/ , DevPod https://devpod.sh/ , CodeSpaces https://github.com/features/codespaces , CodeSandbox https://codesandbox.io/ or Nimbus https://www.usenimbus.com/ , but self-hosted , free as in beer and speech and probably a bit more old-school: - uses just ssh , - all written in bash , - the VMs are regular Ubuntu VMs with a near-zero learning curve, - no docker or k8s involved, no .json files in git repos but you can add anything if you want to , - you are not forced to use VSCode although you can . Hello vi 😁 For the VMs, it uses LXD/Incus containers https://canonical.com/lxd or QEMU VMs https://ubuntu.com/blog/lxd-virtual-machines-an-overview . - You and your team can start up new VMs just by SSH-ing into them. - Works seamlessly with Visual Studio Code's official Remote - SSH https://marketplace.visualstudio.com/items?itemName=ms-vscode-remote.remote-ssh extension. - Uses LXD/Incus containers, which share RAM, CPU and disk space, providing high VM density. - There are "personal" and "shared" VMs. Shared VMs are the ones starting with demo- . Personal ones with username- . - Port 80 of every VM is automatically exposed over HTTPS, like https://demo-foo.example.com . A pre-configured Caddy reverse-proxy https://caddyserver.com/ deals with TLS and forwards HTTP requests to the right VM. - But every subdomain is protected with HTTP password automatically, so forgetful humans don't accidentally expose random stuff to the world. Can be disabled on a per-VM basis by sudo touch /nopassword . - The directory /persist/ is shared between a user's all personal VMs. Makes working on multiple VMs easy, even at the same time. - VMs are considered ephemeral: by default all VMs "die" at night, to protect resources from forgetful humans. Can be disabled on a per-VM basis by sudo touch /nokill . - But files in /persist/ are persistent and survive the nightly killing of VMs. So it's recommended to keep your work there. - If /persist/autoexec.sh chmod 755 exists, it will be run with user permissions right after boot. For multiple environments, we recommend detecting $HOSTNAME . - LXD/Incus containers are pre-configured to be docker -compatible. - By default VMs are running Ubuntu, but you can request different OSes just by provisioning the VM like this: ssh demo-foo-centos8.example.com Then on you can use just demo-foo.example.com . - Personal VMs are protected from other users, but can still be shared if a teammate's SSH key is put in /home/user/.ssh/authorized keys by the VM's owner. - You can use it on an Internet-based server or on-prem behind a corporate firewall: it just needs the wildcard DNS settings and a wildcard TLS certificate. We do use it both ways. - You install this on an Ubuntu server, and put a small snippet in your .ssh/config . And your team members'. - When you SSH to a 🌳VMTREE VM, your SSH client resolves the VM name you requested to the same server, because of the wildcard DNS domain. - Your .ssh/config snippet specifies to use a "jump" user called vmtree on the server. Only the people who have their SSH key in the vmtree user's authorized keys file can connect, of course. - The vmtree user's authorized keys file force-runs the /vmtree/vmtree-vm.sh script on the server. It's not possible to run anything else via SSH with this user. - The SSH snippet in your .ssh/config passes name of the VM you requested to the /vmtree/vmtree-vm.sh script. - The /vmtree/vmtree-vm.sh script does security checks regarding naming convention, etc. - The /vmtree/vmtree-vm.sh script starts an LXD/Incus container with the VM name you specified, passing it a cloud-init https://cloud-init.io/ script that pre-configures the VM with your SSH key and possibly other things . - The /vmtree/vmtree-vm.sh script attaches your "personal disk" to the VM at /persist/ . - The /vmtree/vmtree-vm.sh script waits for the VM to obtain an IP address and have SSH started. - The /vmtree/vmtree-vm.sh script connects your SSH session to the SSH port of the LXD/Incus container. - You are in You can use the LXD/Incus container just as you would with any other VM. And from cron : - Every minute, /vmtree/cron-killme.sh checks for a /killme file on the VMs, and if one's there, deletes the VM. - Every minute, /vmtree/cron-nopassword.sh checks for a /nopassword file on the VMs, and if one's there, reconfigures Caddy to NOT do HTTP authentication for the VM's subdomain. - Every day by default at 6am , /vmtree/cron-stop.sh checks for a /nokill file on the VMs, and if the file is NOT THERE, deletes the VM. If it's a personal VM, still stops the VM, but doesn't delete. This is because people forget about their VMs. - Every day /vmtree/cron-renew.sh checks if the TLS certificate needs to be renewed, and renews it if necessary, with acme.sh https://github.com/acmesh-official/acme.sh . Just a few. - You will be asked twice for SSH authorization. Once for the jump user, and once for the freshly created VM. - LXD/Incus containers are nearly full VMs, but have some security limits regarding mounting file systems, setting system parameters, etc. These rarely interfere with normal dev tasks, and when you need, you CAN start up real VMs, too see below . - The Caddy http auth protects the VMs from the world outside the server, but not from other VMs on the same server. Then again, it's assumed that your server is being used by your own team, not by your enemies. - As of now there is no support company, foundation, charity, etc behind this open-source project. On the other hand, it's barely 600 lines of bash code including comments , so I'm pretty sure your DevOps team can deal with it if necessary. When you occasionally run into LXD's limitations, , you can start real QEMU VMs instead of LXD/Incus containers by specifying -vm as the 4th part of the VM name. For example ssh demo-myrealvm1-ubuntu2004-vm8.example.com creates a QEMU VM called "demo-myrealvm1", using 8GB of memory, running an older Ubuntu version, 20.04. Full QEMU VMs are needed -- for example -- if you want to run a full Kubernetes cluster on the VM. But docker and docker-compose does work on LXD/Incus containers with the configuration that these scripts already do for you. The advantages of QEMU VMs is that they have fewer limitations, but the disadvantage is that they allocate the memory they are given memory is not shared, as it is with LXD/Incus containers . On an Ubuntu 24.04 or 22.04 server with a public IP address, run: sudo git clone https://github.com/kkovacs/vmtree.git /vmtree sudo cp ~/.ssh/authorized keys /vmtree/keys/my sudo /vmtree/install.sh Then put the snippet it prints out in your .ssh/config file on your PC, not on the server, of course . This basic setup will use ip.me https://ip.me/ , nip.io https://nip.io/ , and a self-signed certificate to set up the fullest possible functionality without using your own domain. 1. Get a domain that you will use for VMTREE. From now on: example.com . 2. Host the domain on a DNS provider supported by acme.sh https://github.com/acmesh-official/acme.sh/wiki/dnsapi . Needed for the wildcard TLS certificate. 3. Get a powerful host server that you will use for VMTREE, running Ubuntu 22.04 or 24.04. 4. Configure DNS so example.com and .example.com point to the host server's IP. Wait until it propagates. 5. On your host server, sudo git clone https://github.com/kkovacs/vmtree.git /vmtree 6. Copy your and your teams's SSH public keys ~/.ssh/id .pub files on their PCs to /vmtree/keys/<username files. 7. Run sudo /vmtree/install.sh --skip-install to do basic checks and generate a default /vmtree/.env file. 8. It will say that you need to fill out the /vmtree/.env file. Fill out: - your domain name example.com - your desired automatic HTTP AUTH username / password - your DNS provider and its token to use with acme.sh - optionally, a storage disk/partition will be formatted with ZFS 9. your domain name 10. If you want to use Incus instead of Canonical's LXD, apt-get install incus now. 11. If you need deeper customization complicated ZFS setup, etc , do it now, then install and configure LXD/Incus. 12. Run sudo /vmtree/install.sh again to finish the installation. 13. Distribute the snippet it prints out to your team, to put into their .ssh/config files.