Yesterday, the database that runs shiftsee.com lived on the same droplet as the application. Today, the database is a managed HA cluster on a separate trust boundary. The cache is a managed Valkey service. The cert is Let's Encrypt-issued and renews itself. The application sits behind a load balancer that does not blink if a single droplet does.
Same shiftsee.com. Same login. Same data. Better foundation.
This is the kind of release that should be invisible to the people who depend on the platform. That is the point.
What is different for you
If you are a business owner or a shifter, mostly nothing about today is visible.
Two things you might notice over time:
Pages load slightly faster on average. Application-to-database traffic now stays on a private network and the database is on dedicated hardware. The tail of slow page loads (the kind that used to make a Saturday-afternoon broadcast hang for two seconds) is shorter.
The reliability story is now real. A single-droplet outage no longer takes the platform down. The database survives a node failure with no intervention. The platform is built on the same managed primitives that the rest of the SaaS world uses for production-tier reliability.
What is different if you build against the API
Mobile app devs, MCP client builders, partner integrators, agency engineers, you. Here is what changed.
API docs have a stable URL. Swagger UI is at https://api.shiftsee.com/v1/docs. ReDoc is at /v1/redoc. The raw OpenAPI spec is at /v1/openapi.json. These used to be at the root (/docs, etc.) and now sit under the /v1 prefix that every other versioned endpoint uses. The root paths are retired. Update your bookmarks.
The cert is properly issued. No more "the cert chain looks weird in some clients" reports. shiftsee.com, api.shiftsee.com, and www.shiftsee.com share a Let's Encrypt cert that renews automatically every ~60 days. Pin against the trust chain, not against the leaf.
The data path is segmented. Customer data lives on a managed cluster. Application secrets live on a separate boundary. Your API calls hit the application tier and never see the database directly. That has always been true behaviorally. Today it is also structurally true.
Why we did this now
The roadmap has tiers. Tier-1 is operational floor. Tier-2 is the partner API and embeds. Tier-3 is the AI surface that went live last Saturday. Each tier sits on the tier below it. The AI rollout asked more of the infrastructure than the human-only platform ever did.
AIs do not retry politely when a database hiccups. They do not call human support. They surface the error to the human and erode trust in the platform with every glitch. If you want your AI on ShiftSee acting on your behalf for months at a stretch, the platform underneath needs to be the kind of platform that does not glitch. The two AI agents that just negotiated a Saturday line cook shift cannot have the negotiation interrupted by a primary-DB failover that takes thirty seconds. Today's release is the work that takes the failover down to "no human notices."
Specifically: the HA database survives a node failure invisibly. The cert auto-renews so the AI never hits an expired-cert error a month from now when someone forgot. The two-droplet web tier means a redeploy is rolling, not all-at-once. The cache lives on managed Valkey so the application droplets are stateless and can be recycled or scaled without a session storm.
The infrastructure release that goes uncelebrated is the one your AI quietly relies on for the next six months.
What is next
A few more weeks on tier-1. Next up: observability (uptime monitoring, alert channels, automated rollback), backup verification (we have backups; we do not yet have a tested restore), and per-service scoped database users (closing out a Phase 0 TODO).
After that, tier-2 starts in earnest. The partner API documented at /v1/docs is going to grow. Embeds for the connect-button and the schedule-widget are queued for the next several weeks.
Tier-3 keeps shipping in parallel. Phase 1B of MCP (autonomy settings, self-serve client registration) is the next AI release.
How to read today's release
If you read "tier-1 reliability work" and think "boring, where is the AI feature," the reframe is this: the AI feature that shipped on Saturday assumes a platform that does not glitch. Saturday was the announcement. Today is the substrate that makes it real. Both are required.
If your AI sleeps better tonight, that was the point.