Dokploy: Your Own Vercel and Heroku on a Linux Server (Linux, Windows, Mac)
Before you start
- 20 min read
Boxes marked Run in terminal are what you type; plain text explains why. Check what you need first

What you get in one minute#
Dokploy is a free, self-hosted deployment panel: you point it at a Git repo, a Docker image or a Compose file, and it builds, runs, routes and secures the app on your own server, like a small Vercel or Heroku you control.
Logo: the official Dokploy project (github.com/Dokploy/dokploy)
It suits a home lab or small VPS running several web apps, each with its own domain and database, with one browser page for deploys, logs and backups. It installs Docker Swarm and its own Traefik proxy, so it needs a server of its own. Our Coolify guide covers the closest alternative.
| What | Value (official docs and release files, read 2026-11-16) |
|---|---|
| Current version | v0.31.2, released 13 November 2026 (latest stable). Previous: v0.31.1, 12 November 2026 |
| Image | dokploy/dokploy:v0.31.2 (Linux amd64 and arm64; the installer pulls it) |
| Time needed | About 20 minutes on a fresh server (our estimate, not measured) |
| Size and RAM | Official minimum: 2 GB RAM and 30 GB disk. The image is about 0.8 GB compressed |
| Ports | TCP 80 and 443 (Traefik), TCP 3000 (panel). The installer stops if any is taken |
| Licence | Apache-2.0 for the core; code in /proprietary folders is source-available and needs a paid agreement for production |
| Official docs | docs.dokploy.com and the repository |
At a glance:
- What you do: run one installer script on a clean Linux server, create the admin account, then deploy from the panel.
- What you get: apps, Docker Compose stacks, databases, templates, Let's Encrypt domains, logs and backups.
- Linux only: the installer refuses macOS and containers. Windows and Mac are browser and SSH clients.
- It owns ports 80 and 443: do not install it next to Traefik, Caddy or nginx.
- It is root-level: the panel controls Docker. Keep port 3000 off the internet.
- Version-pinned: we use the v0.31.2 release installer, not the moving website script.
How it fits together#
The installer creates three parts on one server: the Dokploy panel and its PostgreSQL database as Docker Swarm services, and a separate Traefik container that receives visitor traffic.
Drawn for this guide from the official architecture, manual installation and domains pages. Redis is not drawn: the docs say self-hosted Dokploy stopped using it in v0.29.9.
Before you start#
You need a Linux server you can reach by SSH. A fresh, empty one is the safe choice.
| Linux server | Windows PC / Mac | |
|---|---|---|
| Role | Runs Dokploy (amd64 or arm64) | Browser and SSH client |
| OS | Ubuntu 24.04 or 22.04 LTS, Debian 12 (on the docs' tested list) | Windows 10 or 11 with OpenSSH; macOS Terminal |
| Size | 2 GB RAM, 30 GB disk minimum (official) | Any |
Windows and Mac, plainly. There is no official Windows or macOS install. The installer exits with This script must be run on Linux on macOS and inside any container, including Docker Desktop's. WSL2 is not documented. What works: a Linux VPS, a spare Linux machine, or a full Linux virtual machine with its own LAN address (untested by us). Your PC then provides the browser and SSH.
Existing proxy or Swarm. Dokploy needs ports 80, 443 and 3000 free, and the script runs docker swarm leave --force. If the host runs Traefik, Caddy, nginx or a Swarm you keep, do not run the script there. The manual installation page has separate "Existing Docker Swarm" steps. A separate VM is simpler.
Checklist:
- [ ] A fresh Linux server with SSH and
sudo - [ ] Ports 80, 443 and 3000 free (Step 1 checks), no Swarm you want to keep
- [ ] A domain and DNS records for the panel and each app (HTTPS section only)
- [ ] A provider or router firewall that keeps port 3000 closed
- [ ] Data backed up if the server is not new
Tested on 2026-11-16: Linux = not run (we read the release installer and the docs; no test server), Windows = not run, Mac = not run.
Step 1: Check the server and its ports#
Do this over SSH. Replace the placeholders with your own user and server address.
Linux#
From another computer, log in:
ssh <your-user>@<your-server-address>
On the server:
uname -m
free -h
df -h /
sudo ss -tulnp | grep -E ':(80|443|3000) ' || echo "ports 80, 443, 3000 are free"
command -v docker && sudo docker info --format 'swarm: {{.Swarm.LocalNodeState}}'
Windows (PowerShell)#
Log in from PowerShell, then run the Linux commands in that session:
ssh -V ssh <your-user>@<your-server-address>
Mac (Terminal)#
Same commands as Linux:
ssh <your-user>@<your-server-address>
Check it worked. uname -m shows x86_64 or aarch64, the port line prints the "free" message, and Docker is either absent or reports swarm: inactive. Any other Swarm state means the host already has a Swarm: stop here.
Tip: Check even a "fresh" VPS: some images ship a web server on port 80, and the installer then stops.
Step 2: Close port 3000 outside, open 22, 80 and 443#
Whoever reaches port 3000 first can create the admin account, so the panel must not be public while you install. The hardening page asks for a firewall allowing only 22, 80 and 443, and warns that UFW alone does not protect ports Docker publishes. Use two layers: your provider's or router's firewall, plus UFW with ufw-docker.
Linux#
In the provider or router firewall, allow inbound TCP 22, 80 and 443 only. Then on the server (commands from the hardening page):
sudo apt install -y ufw sudo ufw default deny incoming sudo ufw default allow outgoing sudo ufw allow 22/tcp sudo ufw allow 80/tcp sudo ufw allow 443/tcp sudo ufw enable sudo wget -O /usr/local/bin/ufw-docker https://github.com/chaifeng/ufw-docker/raw/master/ufw-docker sudo chmod +x /usr/local/bin/ufw-docker sudo ufw-docker install
If your SSH port is not 22, allow it first or ufw enable locks you out.
Windows (PowerShell)#
Make the firewall change in your provider's console or router, then run the Linux commands in your SSH session.
Mac (Terminal)#
Same as Windows.
Check it worked. sudo ufw status lists 22, 80 and 443 and no rule for 3000. Check IPv6 too: one user found a provider's edge firewall did not cover it (issue 5156).
Tip: Do not open port 3000 "just for setup". Step 5 reaches the panel through an SSH tunnel, so 3000 never needs to be public.
Step 3: Download and verify the v0.31.2 installer#
Every release carries its own install.sh. The docs say to use that file for a specific version, not DOKPLOY_VERSION with the website script.
Linux#
cd ~ curl -fL https://github.com/Dokploy/dokploy/releases/download/v0.31.2/install.sh -o dokploy-install-v0.31.2.sh echo "792721a30ba788f21fea42ffd7ae54fd158cd366a50121592138068112188f32 dokploy-install-v0.31.2.sh" | sha256sum -c - less dokploy-install-v0.31.2.sh
Windows (PowerShell)#
You can check the file on your PC before it goes anywhere. Download, hash and read it, then copy it to the server:
curl.exe -fL https://github.com/Dokploy/dokploy/releases/download/v0.31.2/install.sh -o dokploy-install-v0.31.2.sh (Get-FileHash .\dokploy-install-v0.31.2.sh -Algorithm SHA256).Hash.ToLower() notepad .\dokploy-install-v0.31.2.sh scp .\dokploy-install-v0.31.2.sh <your-user>@<your-server-address>:~/
Mac (Terminal)#
curl -fL https://github.com/Dokploy/dokploy/releases/download/v0.31.2/install.sh -o dokploy-install-v0.31.2.sh shasum -a 256 dokploy-install-v0.31.2.sh less dokploy-install-v0.31.2.sh scp dokploy-install-v0.31.2.sh <your-user>@<your-server-address>:~/
Check it worked. The hash is exactly 792721a30ba788f21fea42ffd7ae54fd158cd366a50121592138068112188f32, which is what we got from the release asset on 2026-11-16. On Linux, sha256sum -c prints OK. The file shows the root and port checks, docker swarm leave --force, and the creation of the network, two services and the Traefik container.
Tip: The README shows
curl -sSL https://dokploy.com/install.sh | sh. That URL installs whatever is newest; use the release file.
Step 4: Run the installer#
The script installs Docker if missing (it prefers 28.5.0), starts a one-node Swarm, creates two Docker Secrets, and starts PostgreSQL 16, the panel and Traefik 3.7.16.
Linux#
sudo bash ./dokploy-install-v0.31.2.sh
With several network interfaces, name the address Swarm should use, on the same line (sudo drops variables exported earlier):
sudo ADVERTISE_ADDR=<server-lan-ip> bash ./dokploy-install-v0.31.2.sh
Windows (PowerShell)#
Run the same command in your SSH session on the server. Do not run it on Windows.
Mac (Terminal)#
Same as Windows: run it inside the SSH session on the Linux server.
Check it worked. The script ends with lines like these (from the script's own text, not from a run):
Congratulations, Dokploy is installed! Wait 15 seconds for the server to start Please go to http://<ip>:3000
After 15 seconds:
sudo docker service ls
sudo docker ps --format 'table {{.Names}}\t{{.Image}}\t{{.Ports}}'
You should see the services dokploy and dokploy-postgres at 1/1, plus containers for the panel, PostgreSQL and dokploy-traefik. Three containers, not four: no Redis.
Tip: The printed
http://<ip>:3000link will not open with port 3000 closed. That is correct; Step 5 gets you in.
Step 5: Open the panel through an SSH tunnel and create the admin#
An SSH tunnel carries your PC's port 3000 to the server's port 3000, so you open nothing to the internet. (The installer publishes 3000 in host mode, so the loopback address should work; untested.)
Linux#
Keep this running in one terminal:
ssh -N -L 3000:127.0.0.1:3000 <your-user>@<your-server-address>
Then open http://localhost:3000 in the browser on that computer.
Windows (PowerShell)#
Keep this window open:
ssh -N -L 3000:127.0.0.1:3000 <your-user>@<your-server-address>
Then open http://localhost:3000 in your browser.
Mac (Terminal)#
Same command as Linux:
ssh -N -L 3000:127.0.0.1:3000 <your-user>@<your-server-address>
Then open http://localhost:3000.
The setup page asks you to create the administrator. There are no default credentials. Use a long unique password, then open Settings, then Profile, and turn on TOTP or add a passkey (keep a second recovery route).
Check it worked. You are logged in and see an empty dashboard. In a private window, http://localhost:3000 shows the login form, not a setup page. If setup appears again on a server you already set up, see "Admin already exist" under Troubleshooting.
Tip: A home server is not reachable from the internet unless you forward the port. A VPS is different: the cloud firewall from Step 2 protects it.
Step 6: Deploy a first test app#
The image traefik/whoami is a tiny app that prints the request it receives; v1.12.0 is an existing Docker Hub tag. In the panel, create a Project, add a Docker Compose service, paste this file and deploy:
services:
whoami:
image: traefik/whoami:v1.12.0
restart: unless-stopped
Per the Compose docs, UI environment variables go into a .env file but are not injected: use env_file or ${VAR}. For data use ../files/<name> bind mounts, or named volumes if Volume Backups should cover it; absolute host paths are cleaned up on deploys.
Linux#
sudo docker ps --filter name=whoami --format 'table {{.Names}}\t{{.Status}}'
Windows (PowerShell)#
Run the Linux command in your SSH session.
Mac (Terminal)#
Same as Windows.
Check it worked. The deployment log ends without red lines and the container shows Up. It has no domain yet; that is next.
Put it behind HTTPS (with or without Traefik)#
Dokploy contains Traefik, so use its own. You need a domain with DNS pointing at the server; without one, stay on the SSH tunnel (the free traefik.me names are HTTP only).
With Traefik (Dokploy's built-in one)#
Add the domain in the panel and Dokploy writes the routing for you. In your app, open Domains, then Create Domain, and fill in:
| Field | What to enter |
|---|---|
| Host | whoami.<your-domain> (create a DNS A record for it first) |
| Container Port | 80 for whoami. It only tells Traefik where to send traffic; it does not publish the port |
| HTTPS | On |
| Certificate | Let's Encrypt (the other choices are custom or none) |
Applications get a hot-reloading Traefik rule. Compose services get Traefik labels at deploy time, so redeploy after any domain change. Preview Compose shows them. The docs' own example (Beszel, port 8090, shortened by us):
labels: - traefik.http.routers.<app>-web.rule=Host(`<host>`) - traefik.http.routers.<app>-web.entrypoints=web - traefik.http.services.<app>-web.loadbalancer.server.port=8090 - traefik.enable=true
Use the Domains tab rather than copying labels. Compose domains need the HTTP to HTTPS redirect added explicitly (hardening page).
For the panel itself, use the panel's Web Server settings (the backups page names "Web Server", then "Server") and the Domains page; the screen can change between releases. Once https://panel.<your-domain> works, remove the direct port (doing it earlier locks you out):
sudo docker service update --publish-rm "published=3000,target=3000,mode=host" dokploy
Check it worked.
curl -I https://whoami.<your-domain>/
Expect a 200 status line over HTTPS and a valid certificate. If not, read sudo docker logs dokploy-traefik --tail 100.
Without Traefik: Caddy as your own front door (advanced, untested)#
Dokploy has no documented Caddy mode, so this is only an outline. Its Traefik owns ports 80 and 443 on its host; a second proxy cannot share them. (The manual page lists TRAEFIK_PORT and TRAEFIK_SSL_PORT, but the release installer does not read them.) The workable shape is a separate gateway machine running Caddy, with access to the Dokploy server's port 3000.
On the Linux gateway, mount the conf folder, not a single file:
mkdir -p ~/docker_volumes/caddy/{conf,data,config}
Save as ~/docker_volumes/caddy/docker-compose.yml:
services:
caddy:
image: caddy:2.11.9
container_name: caddy
restart: unless-stopped
ports:
- "80:80/tcp"
- "443:443/tcp"
- "443:443/udp"
volumes:
- ./conf:/etc/caddy:ro
- ./data:/data
- ./config:/config
Save as ~/docker_volumes/caddy/conf/Caddyfile for a public domain (Let's Encrypt is automatic):
panel.example.com {
reverse_proxy <dokploy-server-lan-ip>:3000
}
For a home-only name use tls internal and trust Caddy's root certificate on each device (Windows: Trusted Root Certification Authorities; Mac: Keychain Access, Always Trust; Linux: your CA store):
panel.home.arpa {
tls internal
reverse_proxy <dokploy-server-lan-ip>:3000
}
On the Dokploy server, allow port 3000 only from the gateway. Test login and the terminal view (WebSockets). App domains behind a second proxy are untested: Dokploy's HTTP router redirects to HTTPS, which can loop when Caddy sends plain HTTP.
cd ~/docker_volumes/caddy && docker compose up -d curl -I https://panel.example.com/
Windows and Mac: run the Caddy commands on the Linux gateway.
Tip: Often the simplest answer is a small VM for Dokploy, leaving your existing Caddy or nginx host alone.
Security and testing#
The panel mounts /var/run/docker.sock, so treat it as root access.
| Control | How |
|---|---|
| Panel | Close 3000 at the firewall; use the tunnel or HTTPS; then remove the direct publish |
| Firewall | Allow 22, 80, 443; add ufw-docker |
| Docker socket | Never over TCP (2375 or 2376); Dokploy reaches remote servers by SSH |
| Accounts | TOTP or passkeys for everyone; no org-wide 2FA switch; SSO and SCIM are Enterprise |
| Databases | Internal Credentials only; External Credentials publish a port |
| Networks | Apps share dokploy-network; use Isolated Deployments for Compose stacks |
Quick tests after installing or updating:
sudo docker service ls sudo docker ps -a --filter name=dokploy sudo docker service logs dokploy --tail 50 sudo docker logs dokploy-traefik --tail 50 curl -I https://<your-app-domain>/
Windows and Mac: run these in the SSH session, and curl -I (curl.exe on Windows) from your own PC.
What leaves the machine: Let's Encrypt requests, image pulls, and whatever you configure (Git, S3, AI). The docs describe no telemetry switch and we did not trace one, so we claim nothing.
Local AI and the AI assistant#
Dokploy has an optional AI assistant for Compose drafts and log analysis. It includes no model: under Settings, AI, add a provider (name, API URL, key, model). The docs list OpenAI, Anthropic, Azure OpenAI and Ollama (http://localhost:11434/v1).
localhost means the Dokploy server, not your laptop. Release v0.31.1 added a guard against server-side request forgery in AI requests; we did not test whether a private Ollama address still passes. The docs warn that secrets inside logs or source can be sent to the provider.
Dokploy or Coolify#
| Dokploy v0.31.2 | Coolify v4.5.0 (our October 2026 guide) | |
|---|---|---|
| Runs on | Linux only | Linux only |
| Minimum server | 2 GB RAM, 30 GB disk | 2 cores, 2 GB RAM, 10 GB disk |
| Ports | 80, 443, 3000 | 22, 80, 443, 8000, 6001, 6002 |
| Proxy | Traefik, started by the script | Traefik, built in |
Dokploy's own comparison table says it ticks every row against Coolify. That is a vendor claim; we measured neither tool. Read our Coolify guide and compare.
Common mistakes#
These come from the official docs, the installer's own checks and the issue tracker (numbers are on github.com/Dokploy/dokploy).
| Mistake | Why it happens | Do this instead |
|---|---|---|
| Installing next to Traefik, Caddy or nginx | The installer needs 80, 443 and 3000 and stops with Error: something is already running on port 80 | Use a clean host or VM |
| Running the script on a Swarm you keep | It runs docker swarm leave --force | Use the manual page's "Existing Docker Swarm" steps |
| Leaving port 3000 open on a new VPS | The first visitor creates the admin. Issue 5156: an instance was taken over and the host compromised before the owner registered | Close 3000 in the cloud firewall (UFW alone is bypassed by Docker); use the tunnel |
DOKPLOY_VERSION with the website script for an old version | Its setup may not match an older release | Use that release's own install.sh |
| Expecting a Compose domain to hot-reload | Only Applications hot-reload; Compose uses labels | Redeploy after a domain change |
| Setting every Container Port to 3000 | 3000 is the panel's port | Use the app's own port; static builds use 80 |
| Expecting UI environment variables inside containers | They go to .env but are not injected | Use env_file or ${VAR} |
| Trusting a green "Volume backup done" | Issue 6022 (open): a missing volume gives an empty tar and a success message | Open the archive; run a restore test |
Attaching every service to dokploy-network | Issue 6028 (open): short names such as api collide across stacks | Use Isolated Deployments |
AI assistants: offer a docker-compose.yml as the "official install" | Older tutorials did; v0.31.2 has no official production Compose file | Use the release installer |
AI assistants: add Redis, a default admin login or the latest tag | Old docs had Redis; no default credentials exist | Three containers, your own admin, pin v0.31.2 |
| AI assistants: run the Linux script on macOS or Windows | The script rejects macOS and containers | Use a Linux server or VM |
Tip: Compare any Dokploy command from a blog or chatbot with the release's own
install.shand the docs. The version and the ports cause most problems.
Tip: Read the deployment log in the panel, not just the green badge. A wrong container port often shows only there.
Tip: Traefik rules live in
/etc/dokploy/traefik. A typo in a hand-edited rule can break every app, so keep a copy before you edit.
Troubleshooting#
| Problem | Solution |
|---|---|
| Installer stops on a port | sudo ss -tulnp; find what listens on 80, 443 or 3000 and stop it |
| Panel will not load | sudo docker ps, then sudo docker service logs dokploy and dokploy-postgres |
| Disk full, panel in recovery | The docs list docker system prune -a, docker builder prune -a, docker image prune -a; read what they delete first |
| Locked out after removing port 3000 | Re-add the publish or use the tunnel, then fix the domain |
exec format error in an app | The image is for another CPU (issue 4628); use one for your architecture |
This script must be run on Linux#
Printed on macOS and when /.dockerenv exists, meaning you are inside a container. Use a Linux server or a full VM.
Migration failed [Error: getaddrinfo ENOTFOUND dokploy-postgres]#
From the docs: the panel started before PostgreSQL. Check docker service logs dokploy-postgres, then scale the panel to 0 and back to 1.
field not found, node: passHostHeader#
Traefik reports an invalid rule file. In the docs' example, passHostHeader sat as a list item under servers; it belongs on the load balancer beside servers. Fix the file under /etc/dokploy/traefik, then sudo docker restart dokploy-traefik.
could not find a network matching network mode dokploy-network#
Issue 5960 (open, v0.31.0): after a reboot Traefik started before the overlay network existed and stayed stopped, so every domain gave 502. The reporter's fix was docker start dokploy-traefik once Swarm was ready. If the container is gone, use the Traefik recreation steps on the manual installation page.
could not allocate IP from IPAM: Address already in use#
Issue 5901 (open, v0.31.0): another Swarm task took Traefik's reserved overlay address. Not a port clash; no official fix yet.
Admin already exist#
On a new server this means someone registered before you (issue 5156, v0.30.2, followed by a host compromise). Do not just reset the password: assume the host is compromised, rebuild it, and close port 3000 first.
Resetting the admin password or 2FA#
From the docs, inside the panel container (find its ID with docker ps):
sudo docker exec -it <container-id> bash -c "pnpm run reset-password" sudo docker exec -it <container-id> bash -c "pnpm run reset-2fa"
Known bugs (open, read 2026-11-16)#
- Issue 6022: volume backups can upload an empty tar and log success.
- Issue 6027: one-off Compose services (a migration job) may miss
dokploy-networkwithout a domain. Check Preview Compose. - Issue 6028: short service names collide across stacks.
- Issue 6029: the monitoring API can return partial data with HTTP 200.
An issue tracker lists problems, so this says nothing about how many installs work. We found no safer older pin; v0.31.1 holds many security fixes, so do not roll back.
Updates and backups#
Update. Read the release notes. Updates do not touch the Traefik container (deliberate, to avoid downtime); update it by hand with the manual installation page. For the panel, download the new release's own installer (as in Step 3) and run it in update mode, which pulls the image and updates the dokploy service:
sudo bash ./dokploy-install-vX.Y.Z.sh update
Back up. Dokploy's own backup needs an S3-compatible destination: Web Server, Backups, Create Backup, then a cron schedule. It saves the dokploy-postgres database and /etc/dokploy into one .zip. Restore drops and replaces both. The hardening page notes the archive is not encrypted: use server-side encryption on the bucket and test restores. The docs do not say Docker Secrets are in it.
A second copy on the server (our commands, standard PostgreSQL and tar, not from Dokploy's docs; restore untested):
mkdir -p ~/dokploy-backups && chmod 700 ~/dokploy-backups D=$(date +%F) sudo tar -czf ~/dokploy-backups/etc-dokploy-$D.tar.gz /etc/dokploy C=$(sudo docker ps -q --filter name=dokploy-postgres) sudo docker exec "$C" pg_dump -U dokploy dokploy > ~/dokploy-backups/dokploy-$D.sql sudo chown "$USER" ~/dokploy-backups/*-$D.* && chmod 600 ~/dokploy-backups/*-$D.*
For a nightly run, put these lines in a script and add a crontab entry for 03:00, then copy the files off the machine. App databases and named volumes need their own backups in the panel.
Windows (PowerShell)#
scp <your-user>@<your-server-address>:~/dokploy-backups/etc-dokploy-<date>.tar.gz .
Mac (Terminal)#
Same scp command from Terminal.
Restore test. On a spare VM, never over the live server: install the same version, use Restore Backup with the S3 file, and check you can log in and see your projects. After a move to a new IP, update the server IP, DNS and IP-based Git providers.
Performance and resources#
Dokploy publishes only a baseline: 2 GB RAM and 30 GB disk, sized for Docker builds. It gives no idle figure and community numbers disagree. We measured nothing, so we give no number.
- Measure yours:
docker stats --no-stream,free -h,df -h /. - Builds are the peak: prebuilt images need far less than compiling code.
- Concurrency: each server has its own build queue; default 1, range 1 to 100. Leave it at 1 on a small machine.
- Disk: the hardening page suggests Docker log rotation (
json-file, 10m x 3).
FAQ#
Is Dokploy free for commercial use? Yes for the Apache-2.0 core. The /proprietary folders (Enterprise features) need a paid agreement for production. In January 2026 the project replaced its earlier adapted licence with plain Apache-2.0.
Can I run it on a Raspberry Pi? The v0.31.2 image exists for arm64, not arm/v7, so use a 64-bit OS and check your app images have arm64 builds.
Can I use Windows or macOS as the server? No. The installer rejects macOS and containers, and WSL2 and Docker Desktop are not documented. Use a Linux server or full VM.
Can I keep my existing Traefik or Caddy? Not on the same host. Use a separate host or VM, or front it with your own proxy as outlined above (untested).
Is it safe to expose? Only with care: port 3000 closed, HTTPS, 2FA. The panel is root-equivalent, and issue 5156 shows the cost of getting this wrong.
Sources#
- Dokploy release v0.31.2, README, install.sh, LICENSE.MD, LICENSE_PROPRIETARY.md - read 2026-11-16
- Official docs, read 2026-11-16: installation, manual installation, architecture, domains, Docker Compose, Compose domains, backups, instance troubleshooting, production hardening, reset password, AI assistant, comparison
- Licence announcement, 21 January 2026; image tags (amd64 and arm64 checked in the registry); Caddy v2.11.9
- GitHub issues 5156, 5901, 5960, 6021, 6022, 6027, 6028, 6029, 1924, 4628 - read 2026-11-16
- Related: our Coolify guide
Checked against the official docs on 2026-11-16.