A practitioner's deep dive into local-first web architecture in 2026, written from hard-won production experience. Covers the core mental model (client as a database node, not a thin view), when local-first is the wrong choice, client-side storage options (IndexedDB vs. SQLite WASM via OPFS), sync approaches (CRDTs with Yjs/Automerge, database replication with PowerSync/ElectricSQL, event sourcing), and conflict resolution strategies including field-level last-write-wins and semantic conflict flagging. Also addresses auth, authorization at the sync boundary, schema migrations across thousands of client devices, performance trade-offs (instant reads/writes vs. initial sync cost and bundle size), and testing challenges. Includes honest tool evaluations and a recommended starting stack.
Table of contents
What “Local-First” Actually Means (And The Confusion That Won’t Die)Be Honest Early: When You Should Not Do ThisReplicas, Not RequestsWhere Data Lives on the ClientThe Part That’s Actually HardConflicts: The Thing Everyone’s Afraid OfThe Tools Right NowBuilding A Real App: Architecture, Auth, And MigrationsIf I Were Starting A New Project TodayPerformance: What’s Actually Fast And What HurtsTesting This StuffWhat I’m Watching, What Worries MeFurther ReadingSort: