shop-api
3 deploymentsproduction
Runningstaging
RunningPR #42 preview
Building+ Add a deployment
Your own PaaS, on a box you own.
Mangrove is a small, self-hosted platform. Point it at a Git repo, a Dockerfile, a compose file or an image, and it builds it, runs it, health-checks it and serves it over HTTPS, with blue/green deploys and automatic rollback.
On a Debian or Ubuntu server, clone it
git clone https://github.com/evanxdsouza/mangrove.git
Then run the installer
cd mangrove && sudo ./setup.sh
It installs Docker and Caddy if they're missing, builds one Go binary, and claims the instance with a single-use setup token so you become its owner. 2 GB of RAM is enough. Re-running it is safe.
Plant whatever you already have.
A new instance starts empty. Add a deployment from a GitHub repo, a Dockerfile, a docker-compose file or a pre-built image, or install a self-hosted app from a template and never write a Dockerfile at all.
New deployment
Link a repo and every push redeploys it. Staging environments track their own branch and promote the exact commit they're running; pull requests can each get a preview, with a bot comment linking to it.
Repositoryacme/shop-api
Branchmain
Builds the Dockerfile in your repo. No Dockerfile? Nixpacks, if it's installed, works out how to build the repo for you.
Build strategyDockerfile
Dockerfile pathDockerfile
Internal port3000
Health check path/healthz
Runs every service in a docker-compose file together, as one deployment.
Build strategyCompose
Compose filedocker-compose.yml
Public serviceweb
Skips the build and runs an image you've already pushed to a registry.
Imageghcr.io/acme/web:latest
Internal port8080
Common self-hosted apps, ready to install. Templates that need a database install one alongside themselves.
GhostWrite and publish posts on your own site.
VaultwardenStore and share passwords securely.
Uptime KumaGet notified if one of your sites goes down.
GiteaHost your code, like a private GitHub.
Also WordPress, Supabase, PocketBase, NocoDB, n8n and Umami, plus Postgres, MySQL, MongoDB and Redis on their own.
CancelDeploy
Then watch the tide come in.
Each deploy builds a new container beside the old one. It only takes traffic once it passes its health check. If it never does, the old one keeps serving and nothing changes for your visitors.
Every deploy stays in the history, one click from a rollback.
production
shop-api · main · a41f9c2- Queuedtriggered by push
- BuildingDockerfile
- Health checkingGET /healthz
- Runningtraffic switched, old container stopped
STDOUTRECORDING
12:02:01 queued: push to main (a41f9c2) 12:02:02 cloning acme/shop-api 12:02:04 building image from Dockerfile 12:02:39 image built in 35s 12:02:40 starting the new container 12:02:41 listening on :3000 12:02:43 GET /healthz 200 3ms 12:02:43 routing traffic to the new container 12:02:44 previous container stopped
First deploy live. This is the start of something growing.
A few weeks later, the box is full.
Projects group the deployments that belong together, each hanging off its project's root line. A project wears the tide of its worst deployment, so a build in progress or a failure never hides inside a group.
acme
home
2 deployments, from templatesblog
Runninguptime
Running+ Add a deployment
reports
1 deploymentdashboard
Sleeping+ Add a deployment
Example projects, as they appear on the dashboard's Projects page.
The parts you'd otherwise script yourself.
On a deployment's page or in the admin area, each of these is a switch or a field, not a config file to keep in your head.
production
shop-api · deployment settingsDomains
Point a domain at a deployment and Caddy gets and renews its certificate. Behind a home router with no static IP, the built-in DDNS updater keeps the record pointed at you.
shop.example.comCertificate issuedhome.example.netDDNSAccess control
Put a deployment behind a password or a Mangrove sign-in, and leave chosen paths open.
Leave open/api/webhooks
Self-healing
Restarts a service after three failed health checks in a row, and retries a failed deploy twice, a minute and then five minutes later.
Idle sleep
Quiet apps stop after a while and wake on the next visit, behind a short "waking up" page, so a small box holds more of them.
Admin
the box itselfServer health
Built and run day to day on a 2 vCPU, 2 GB box, alongside the apps it hosts.
Memory412 MB / 1.9 GB · 21%
Disk11.2 / 15.6 GB · 72%
Members
Workspaces group projects by team or client. Each person gets admin, editor or viewer in each one, and everything they do lands in the audit log.
ana@example.comAdminsam@example.comEditorclient@example.comViewerBackups and storage
One command captures the database and its encryption key together. On a home server, a plugged-in drive can be shared over SMB through a separate helper, so the control plane never needs mount rights.
$ mangrovectl backup --out mangrove.tar.gzSimple enough to hand to someone else.
Simple mode swaps projects and build strategies for a plain list of apps, with words like "Starting up" instead of status codes. Same box, same apps, one switch.
Your apps
BlogWrite and publish posts on your own site.
RunningPassword managerStore and share passwords securely.
Starting upWebsite monitorGet notified if one of your sites goes down.
Turned offghost · production
Build strategyImage
Imageghost:5-alpine
Internal port2368
Domainblog.example.com
HealthHealthy
Put down roots.
Clone it, run the installer, and your first deploy can be live in a few minutes. Then drive it from the dashboard, mangrovectl, a terminal UI with a live shell, the HTTP API, or an AI agent over MCP.