Skip to content

Self-host Ployz Cloud

You can run Ployz Cloud, the dashboard at ployz.dev, on your own server instead of using ours. Your apps still run on your servers either way. Self-hosting moves only the dashboard, which you run with Docker Compose.

Most people should use ployz.dev. It's free on your own servers, we run and upgrade it, and your apps and data stay on your servers anyway. Secrets are sealed on your device and opened only on your servers.

flowchart TD
  start(["Do you need the dashboard itself on your own infrastructure?"])
  start -- "No" --> cloud["Use ployz.dev"]
  start -- "Yes, for compliance, an air gap or just because" --> ops{"Happy to run Postgres,<br>back it up and upgrade it?"}
  ops -- "No" --> cloud
  ops -- "Yes" --> self["Self-host Ployz Cloud"]
ployz.devSelf-hosted
Where your apps runYour serversYour servers
Who runs the dashboardPloyzYou: setup, backups, upgrades
PriceFree on your own serversFree, plus your server for the dashboard
SetupSign in with GitHubTwo GitHub apps, a server, a public https address
Still uses Ployz-hosted partsAll of themThe relay, hosted DNS, the installer and release binaries

A self-hosted Ployz Cloud still reaches your servers through the relay at relay.ployz.dev, gets each organization's generated domain from our hosted DNS, and installs from ployz.sh and GitHub releases.

Ployz Cloud's TLS proxy and a Ployz server that takes web traffic both need ports 80 and 443. If both run on one server, the proxy wins: your domains show Something else answers on port 80 of … and visitors get an SSL handshake error.

Your setupWhat to do
A separate server for Ployz CloudPut Caddy or nginx on it, as in step 3.
One server for everythingSkip the proxy. Publish Ployz Cloud with a Cloudflare Tunnel to http://localhost:3000, so ports 80 and 443 stay free for your apps.
A server where only Ployz Cloud's proxy uses 80 and 443On the server's page, turn off Takes web traffic under Web traffic and confirm Stop. Your apps' domains then need another server.

You need:

  • A Linux server with Docker Compose that can run linux/amd64 images.
  • A public https address for the dashboard, like https://cloud.example.com, from your proxy or tunnel.

Ployz Cloud needs two GitHub apps: an OAuth App for signing in, and a GitHub App for reading repositories. In both, replace https://cloud.example.com with your address.

OAuth App. Create it under Settings → Developer settings → OAuth Apps:

FieldValue
Homepage URLhttps://cloud.example.com
Authorization callback URLhttps://cloud.example.com/api/auth/callback/github

Keep its Client ID and a new client secret for step 2.

GitHub App. Create it under Settings → Developer settings → GitHub Apps, on your account or your GitHub organization:

FieldValue
Homepage URLhttps://cloud.example.com
Callback URLhttps://cloud.example.com/api/github/callback
Request user authorization (OAuth) during installationChecked
Webhook URLhttps://cloud.example.com/api/github/webhook, with Active checked
Webhook secretThe output of openssl rand -hex 32. Keep it for step 2.
Repository permissionsContents: Read-only. Checks: Read and write. Metadata: Read-only. Actions: Read and write. Pull requests: Read-only.
Organization permissionsMembers: Read-only. Ployz Cloud uses it to check that the person connecting a GitHub organization is one of its owners.
Subscribe to eventsPush, Check suite, Workflow run, Pull request

Then generate a private key and a client secret on the app's page, and note its App ID, its Client ID, and its slug, the name in its URL github.com/apps/<slug>. Ployz Cloud uses the app's client ID and secret to ask GitHub which installations the person connecting owns.

Your Ployz Cloud keeps the private key and the client secret, and makes its own GitHub tokens with them. Each token gets only what its step needs. A token for reading code, pull requests and checks can only read. A token for starting, following or cancelling a build, or for posting the Ployz check, works in that one repository.

One app for both. You can skip the OAuth App and sign in with the GitHub App: set GITHUB_CLIENT_ID and GITHUB_CLIENT_SECRET to the GitHub App's client ID and secret. Then the GitHub App needs the account permission Email addresses: Read-only, so it can read private emails at sign-in, and both Callback URLs, with https://cloud.example.com/api/github/callback first and https://cloud.example.com/api/auth/callback/github second. GitHub sends a fresh installation back to the first Callback URL. Every sign-in then hands Ployz Cloud a token that can do what the GitHub App can, in every repository the person signing in reaches. With a separate OAuth App, a sign-in token only reads the person's profile and email addresses.

  1. Download the compose file and the example settings from the latest release into a new directory:

    mkdir ployz-cloud && cd ployz-cloud
    curl -fsSLo compose.yml https://github.com/getployz/ployz/releases/latest/download/ployz-cloud-compose.yml
    curl -fsSLo .env https://github.com/getployz/ployz/releases/latest/download/ployz-cloud.env.example

    releases/latest is the newest stable release. To run a prerelease, like a beta, replace latest/download with download/<tag>, for example download/v0.3.0-beta.7.

  2. Generate the secrets, one per line:

    for name in POSTGRES_PASSWORD BETTER_AUTH_SECRET APP_ENCRYPTION_SECRET INNGEST_EVENT_KEY INNGEST_SIGNING_KEY; do
      echo "$name=$(openssl rand -hex 32)"
    done
  3. Fill in .env:

    VariableValue
    APP_URLYour public address, like https://cloud.example.com
    POSTGRES_PASSWORD, BETTER_AUTH_SECRET, APP_ENCRYPTION_SECRET, INNGEST_EVENT_KEY, INNGEST_SIGNING_KEYThe generated secrets
    GITHUB_CLIENT_ID, GITHUB_CLIENT_SECRETThe OAuth App's client ID and secret
    GITHUB_APP_ID, GITHUB_APP_SLUGThe GitHub App's ID and slug
    GITHUB_APP_CLIENT_ID, GITHUB_APP_CLIENT_SECRETThe GitHub App's client ID and secret, not the OAuth App's
    GITHUB_APP_WEBHOOK_SECRETThe GitHub App's webhook secret
    GITHUB_APP_PRIVATE_KEYThe whole private key PEM, in quotes
    PLOYZ_HOSTED_DNS_URLOptional. compose.yml sets https://dns.ployz.app/ when it is unset

    Leave every POLAR_* and POSTHOG_* variable unset: that turns off billing and product analytics. If a required value is missing or malformed, web and worker stop at startup, and docker compose logs web worker names each variable to fix. To turn on the Ployz agent, set ANTHROPIC_API_KEY. To count visits on an environment's Analytics page, set ANALYTICS_ENABLED=true. Email is optional too. Ployz Cloud sends none while POSTSHIBA_API_KEY, POSTSHIBA_TEAM_ID, POSTSHIBA_CLUSTER_ID, and MAIL_FROM are unset. Set all four to send through PostShiba. With only some of them set, web and worker stop at startup. Each person chooses which emails they get under Account → Notifications.

Compose uses ployz-cloud-self-host as the project name, even in another directory. A second copy with that name replaces the first install's containers and reuses its database. Before starting a second install, set a unique COMPOSE_PROJECT_NAME in its .env, such as COMPOSE_PROJECT_NAME=ployz-staging. Compose uses it instead of the name in compose.yml. Set a different WEB_PORT too.

Keep the project name unchanged for an existing install. Changing it selects a different volume and makes the database appear empty. With a custom name, back up <COMPOSE_PROJECT_NAME>_postgres-data and use the network <COMPOSE_PROJECT_NAME>_default for a containerized proxy. The default names below still apply when you leave it unset.

  1. Create the database tables, then start everything:

    docker compose run --rm web npm run db:migrate
    docker compose up -d
  2. Point your proxy or tunnel at port 3000, or at WEB_PORT if you set it. With Caddy on the same server as Ployz Cloud, the whole config is:

    cloud.example.com {
      reverse_proxy localhost:3000
    }

    Ployz Cloud listens only on 127.0.0.1, so other computers can't skip your proxy. If your proxy or cloudflared runs in a Docker container on this server, attach it to the ployz-cloud-self-host_default network and point it at http://web:3000. If your proxy runs on another server, set WEB_BIND in .env to a private address of this server that only the proxy can reach. Don't set it to 0.0.0.0: ports that Docker publishes skip firewalls like UFW, so anyone could reach Ployz Cloud over plain http, past your proxy's TLS and rate limits.

  3. Run docker compose ps. Ployz Cloud is ready when worker shows healthy.

  1. Open your address and sign in with GitHub.
  2. Open Organization → GitHub and click Connect GitHub. GitHub asks you to install the app and to authorize it. See Who owns the connection for which GitHub accounts you can connect.
  3. Add a server. You need one even if it's the server Ployz Cloud runs on: Ployz Cloud is only the dashboard, and your apps run on the servers you add. The command the dashboard gives you already points at your address.

To install a server with a Fleet Token, use the install line Add servers or ployz server token create shows. Both already point at your address. The Add servers line installs the release your Ployz Cloud runs. The ployz server token create line installs the release of the ployz CLI that made it, so if your CLI and your Ployz Cloud run different releases, use the dashboard's line or upgrade the CLI first. To write the line yourself, add your address, or the server checks in with ployz.dev. An address that answers but isn't Ployz Cloud ends the install with an error:

curl -fsSL https://ployz.sh | sudo PLOYZ_CLOUD_URL=https://cloud.example.com sh -s -- VERSION --enroll

The CLI signs in to ployz.dev unless you point it at yours:

ployz login --cloud-url https://cloud.example.com

In CI, set PLOYZ_CLOUD_URL next to PLOYZ_TOKEN; see Make an organization token.

Pick the release to move to on the releases page. If you run a beta, pick the newest release marked Pre-release. Don't use releases/latest here: it's the newest stable release, so it would move a beta back to an older release. If the database contains migrations the image does not ship, db:migrate, web, and worker refuse to start. The error names the image version and asks you to use a newer image. To roll back to an older image anyway, set PLOYZ_ALLOW_OLDER_IMAGE=1 in .env. It starts with a warning, and requests that need the newer tables fail. Remove it once you run a newer image. Rolling forward again needs nothing beyond the usual db:migrate.

Once 0.3.6 has made any generated value (Create → Database makes one for every new database's password), don't roll Cloud back to an earlier release. It starts without a warning, but it stops deploys for the whole organization (once the organization has a volume, as every database does) until you roll forward. Apps keep running. Rolling forward to 0.3.6 or later restores everything, and every value stays as it was.

Set VERSION to that release's tag, then run the rest. It refuses a tag older than the one you run now:

VERSION=vX.Y.Z
current=$(docker compose config --images | sed -n 's|^ghcr.io/getployz/ployz-cloud:||p' | head -n 1)
verdict=$(awk -v from="$current" -v to="$VERSION" '
  function key(tag, p, n) {
    if (tag !~ /^v[0-9]+\.[0-9]+\.[0-9]+(-[a-z]+\.[0-9]+)?$/) return ""
    n = split(substr(tag, 2), p, "[.-]")
    return sprintf("%06d.%06d.%06d.%s.%06d", p[1], p[2], p[3], n > 3 ? "a" p[4] : "z", p[5])
  }
  BEGIN {
    if (key(to) == "") print "Set VERSION to a release tag, like v1.2.3 or v1.2.3-beta.4."
    else if (key(from) == "") print "Could not read the current release from compose.yml."
    else if (key(to) < key(from)) print to " is older than " from ". Not downgrading."
  }')
if [ -n "$verdict" ]; then
  echo "$verdict"
else
  curl -fsSLo compose.yml "https://github.com/getployz/ployz/releases/download/$VERSION/ployz-cloud-compose.yml" &&
    docker compose pull &&
    docker compose run --rm web npm run db:migrate &&
    docker compose up -d
fi

Nothing migrates the database on start, so run the migration on every upgrade. docker compose up -d can take up to 30 minutes: worker finishes the work it's running before it stops.

Before v0.3.5, Ployz Cloud encrypted secret variables with APP_ENCRYPTION_SECRET. Now only your Servers open them. Each time worker starts, it reseals any such variable it finds for your Servers. It needs the same APP_ENCRYPTION_SECRET you had before, and a Server that has joined. You don't run anything for it.

Until the reseal finishes, a deploy that reads one of those variables stops before it changes anything, and says the variable is waiting to be resealed. Your apps keep running with their values. Deploy again once it's done, usually within a minute. docker compose logs worker shows 0 legacy values for each organization when nothing is left. If a deploy keeps stopping on one variable, set that variable again.

  • No root, no cloud lookups. web and worker run as the unprivileged node user, and by default don't contact a cloud provider's metadata address such as 169.254.169.254 at startup. When telemetry is on, setting OTEL_NODE_RESOURCE_DETECTORS=all turns those lookups back on.
  • No telemetry by default. web starts the OpenTelemetry SDK only when you set an OTEL_* variable, such as OTEL_EXPORTER_OTLP_ENDPOINT, even to an empty value. OTEL_NODE_RESOURCE_DETECTORS, which the image sets, doesn't count. Without one, web starts no tracing and sends nothing.
  • Config Store warnings. When the Config Store or a deploy hits a problem it works around, such as a branch it couldn't update in time, web and worker write one JSON line for it to stderr, shown in docker compose logs. A line says what went wrong, which branch, volume or machine it hit, and the error's first line. It never holds a variable's value: anything else reads [redacted]. To change which lines appear, set PLOYZ_SDK_LOG in .env to a tracing filter. It replaces the default, ployz=warn, so keep that in it: for example PLOYZ_SDK_LOG=ployz=warn,ployz_store=info. Lines below warnings mostly read [redacted].
  • Back up one volume. Everything Ployz Cloud stores, your organizations, projects and settings included, is in ployz-cloud-self-host_postgres-data.
  • Restarting Postgres. While it's down, pages say This page couldn't load and try again every few seconds. They come back by themselves within seconds of Postgres starting.
  • Rate-limit requests in your proxy. Ployz Cloud limits only failed sign-ins from the ployz CLI, per address. If your instance is public, set a per-address request rate limit on /api/ and /auth/ in your proxy or tunnel. Leave out /api/github/webhook, since GitHub doesn't resend a rejected delivery, and leave room for Servers that share one address.
  • Pass each client's own address. Ployz Cloud takes the client's address from X-Forwarded-For when your proxy reaches it from a private or local address, as in the setups above. Caddy and cloudflared add that header for you. Otherwise every client shares the proxy's address. Then someone guessing tokens can lock every ployz user out for up to an hour.
  • Adding a GitHub permission later. Existing installations keep working without it until their owner approves the new permissions on GitHub.
  • No uptime checks yet. Self-hosted Ployz Cloud doesn't run the uptime checker, so services show no up or down status. Their Monitoring tab shows CPU and memory, with no uptime checks or incidents.
  • Behind Cloudflare? Your apps' domains need SSL/TLS set to Full (strict). See My domain is behind Cloudflare. Your apps see the visitor's address, not Cloudflare's.