Platforms

Run Noite on Coolify or Railway — env, domains, downtime model, and platform-specific settings.

Coolify

Point a Docker Compose resource at docker/compose.yaml (repo root, branch main): the same file the installer uses, driven by env. In Environment Variables, set BASE_DOMAIN to your domain (defaults to localhost) and BETTER_AUTH_URL / GIT_PUBLIC_BASE to match, the secrets (BETTER_AUTH_SECRET, RUNNER_TOKEN, and RUSTFS_ACCESS_KEY + RUSTFS_SECRET_KEY or your own bucket’s keys) and leave NOITE_IMAGE on latest. Nothing auto-generates secrets; dev defaults are refused on real domains. Deploy, then paste the real hostnames once on the noite service’s Domains field: https://app.<domain>:80,https://api.<domain>:80,https://git.<domain>:80. The apex stays on your marketing site. Behind a terminating proxy set CADDY_AUTO_HTTPS=off.

Downtime model

Coolify does not do rolling updates for Compose: docker compose up stops a service before starting its replacement. A new image restarts the runner and every tenant fleet with it (they cold-boot, roughly a minute for many apps), so batch upgrades and deploy them off-peak.

Tenant subdomains (<slug>.<domain>) are fully automatic: no per-app steps, no wildcard cert, no DNS provider. A Traefik TCP router forwards every *.<domain> SNI straight to Noite’s Caddy on :443, which mints a per-slug certificate on demand (ask-gated by the runner at /v1/edge/tls-ask, so only live tenant and platform hosts get certs).

One-time setup (besides *.<domain> DNS pointing at the server): save this file under Servers > server > Proxy > Dynamic Configurations in Coolify. It is static text that Coolify cannot mangle, and noite-tenants@docker resolves the TCP service the noite service label defines:

tcp:
  routers:
    noite-tenants:
      entryPoints: [https]
      rule: 'HostSNIRegexp(`^.+\.<domain>$`)'
      service: noite-tenants@docker
      tls: { passthrough: true }

Then redeploy once and confirm the noite-tenants router in the Traefik dashboard. The first visit to a new slug pauses a few seconds for certificate issuance; certs persist in the noite-data volume. If passthrough misbehaves, the fallback is https://<slug>.<domain>:80 per app (exact hostnames use the plain HTTP challenge).

Railway

One service from the image, one volume, one bucket:

noite takes the same variables as a Compose install: BASE_DOMAIN, CONTROL_SUBDOMAIN=app, BETTER_AUTH_URL=https://app.<domain> (set before the first registration: passkeys bind to it), GIT_PUBLIC_BASE=https://git.<domain>, BETTER_AUTH_SECRET, RUNNER_TOKEN, the bucket (S3_ENDPOINT=http://rustfs.railway.internal:9000 and its keys, or your R2/Tigris endpoint), plus CADDY_AUTO_HTTPS=off (Railway terminates TLS) and RAILWAY_DEPLOYMENT_DRAINING_SECONDS=35 so the runner’s stop budget fits. For a healthcheck, set PORT=8080 (Railway probes the port in PORT; the runner always listens on 8080 and Caddy on 80, so this changes nothing else), path /ready, and RAILWAY_HEALTHCHECK_TIMEOUT_SEC=600: a first boot deploys the control bundle and can take minutes.

One wildcard domain is the whole DNS story: add the * CNAME, the _acme-challenge CNAME and the ownership TXT the dashboard returns for *.<domain> on the noite service. That covers app., git., api. and every {slug}. host, because the edge dispatches by Host.

The volume holds the runner’s SQLite, git mirrors, builds and certificates; the runner also snapshots its SQLite into the bucket, so a lost volume restores from there on the next boot. Back up the bucket (Railway volume backups for a RustFS service, provider versioning for R2/Tigris).