Between March 30 and April 1, Xeref.ai shipped four versions. Here's what each one actually involved.
v1.1 — Payments with Creem
We chose Creem over Stripe for one reason: less configuration noise. The integration covers checkout, confirmation emails, and webhook-based subscription handling. When a user upgrades, their plan is reflected immediately — no manual sync, no delays.
The hardest part wasn't the payment flow. It was the OAuth mismatch between local development and production. We fixed that in v2.1 cleanup.
v1.2 — Dynamic Model Routing
Not every task needs the most capable model. v1.2 introduced a routing layer that selects the right model based on task type, user plan, and context length. Basic tier users route through the Xeref model. Opus Plan Mode is reserved for architectural reasoning tasks where multi-step thinking matters.
This cut inference costs without touching output quality for 80% of tasks.
v1.3 — AI SDK v6
This upgrade was overdue. AI SDK v5 had accumulated rendering inconsistencies that were hard to reproduce and even harder to fix without upgrading. v6 corrected message rendering and streaming behavior in one pass.
v1.4 — Home View and Stats
The home view shipped with a chat/tasks toggle and a live stats dashboard. Small change, outsized impact on how users orient themselves when they log in.
The operational lesson: batch infrastructure updates, ship them fast, and move on. Don't let them accumulate.