Case workers for CPIMS register vulnerable children in places where connectivity is a rumour. The app couldn't degrade offline — it had to be offline-first, with the network as a bonus, not a requirement.
The problem
Most apps treat the server as the source of truth and the device as a dumb cache. That model dies the moment the signal does: forms fail to submit, data is lost, and a worker in the field has no idea whether their last hour of work saved.
Local-first storage
Each device runs a local PouchDB instance. Reads and writes hit the device first — instantly, always, signal or no signal.
const db = new PouchDB("cases");
await db.put({
_id: caseId,
child: { name, dob },
updatedAt: Date.now(),
});The UI never waits on the network. As far as the worker is concerned, the app is always online.
Conflict-safe replication
When a connection returns, PouchDB replicates bidirectionally with a central CouchDB cluster. Conflicts are inevitable — two workers, one child, two edits — so we made resolution deterministic rather than last-write-wins.
db.sync(remote, { live: true, retry: true })
.on("change", queueConflictCheck)
.on("error", scheduleBackoff);CouchDB keeps every conflicting revision; a small reconciler picks a winner by domain rules (most recent verified visit) and records the rest for audit.
Outcomes
- 100% offline-capable — full workflow with zero signal.
- Conflict-safe sync that never silently drops an edit.
- Faster field follow-ups, because nothing blocks on the network.
Offline-first isn't a feature you bolt on. It's a decision about where the truth lives — and we put it on the device.