MetadevXInnovate. Code. Create.MetadevXInnovate. Code. Create.

(01) The challenge

The planning tool stopped at the site gate

TaskForge's crews run inspections from vans, warehouses and rooftop plant rooms — everywhere a browser tab fails slowly. Planning lived in a web app that nobody could trust offline, so the work reverted to paper and someone re-typed it in the evening.

The cost was not the typing. It was that the day's plan was only visible the next morning, by which point the exceptions had already been absorbed by the crew.

  • · Average 40 minutes per crew lost to re-keying at end of shift
  • · No shared view of what was outstanding while out of coverage
  • · Previous pilot crashed on large photo attachments
Offline-First Desktop Task Manager — The challenge

(02) Approach

A native shell, a real local database

We built the app in Rust with a Tauri front end so the binary stays small and starts instantly, and made the local store the source of truth rather than a cache. Every write lands on disk first; sync is a background reconciliation pass that can be interrupted at any point.

The interface is deliberately ordinary — dashboard, board and calendar over the same records — because the difficult work is underneath it.

  • · Rust core with SQLite locally and an append-only change log
  • · Conflict resolution by field-level merge, last editor never silently wins
  • · Signed installers for Windows, macOS and Linux from one pipeline
  • · Silent background updates with a pinned rollback version
Offline-First Desktop Task Manager — Approach

(03) Outcome

Two weeks of work per crew, per month, back

Across the first quarter of deployment, end-of-shift re-keying disappeared and exceptions became visible the same day. The binary is under 12 MB and opens in well under a second on the oldest machines in the fleet.

100%

Tasks captured offline

12 MB

Installed binary size

3

Platforms from one codebase

Offline-First Desktop Task Manager — Outcome

(04) What we did

We proved sync before building screens

The first milestone was a headless sync harness, not a user interface. Two clients on the same account were driven against each other with network cuts in the middle until the change log converged every time. Only then did the product work start.

  • · Sync modelled as an append-only log with deterministic replay
  • · Property-based tests over randomised offline sequences
  • · Field pilot on six machines for a month before the wider rollout
  • · Update channel staged by device group
Next projectOn-Chain Settlement Rail→