Uptime
Where we host, what we watch, and how we tell you when something breaks. The stack is small and boring on purpose.
Updated October 2026Four providers
Vercel
Websites and client sites run on Vercel's edge network, with its 99.99% target. Region: Washington, D.C.
Hetzner
Automations, Postgres, and tracing run on a Hetzner server in Ashburn, Virginia. A second server in Falkenstein, Germany is the disaster-recovery target.
Neon
Each database lives in its own Neon Postgres instance. Point-in-time recovery, branched dev databases, encrypted at rest.
Cloudflare
DNS and edge caching. R2 holds the off-site backups.
When something breaks
1.
We notice.
Synthetic checks hit the public site every 60 seconds. Internal services report their health. A sev-1 pages the founder's phone.
2.
First note in fifteen minutes.
A public note that says what we know and what we do not. We do not wait for a clean explanation.
3.
Updates while it is open.
Every thirty minutes for a sev-1, every hour for a sev-2. Affected clients get a direct email on the same clock.
4.
A write-up after.
Every sev-1 gets a public post within seven days: timeline, cause, what changed. Posted on labs.spirecoast.com.
Backups run nightly to Cloudflare R2, in a different region from the database. Point-in-time recovery covers the last seven days. Once a quarter we rehearse a full restore, with a target of under four hours for a complete rebuild. A redacted copy of the runbook goes to any client whose scope asks for it.
A public status page at status.spirecoast.com is in the works. Until it ships, active clients get incident emails directly.