Building Flustro with a local-first constraint

Cloud voice tooling is convenient, but it is also fragile when your network is not predictable.

For Flustro, I committed to a local-first architecture early. That decision forced tradeoffs in model packaging, startup time, and desktop UX, but it removed a larger class of reliability problems.

Why local-first

The core value was simple: if I open the app, dictation should work now, not after a round-trip to a region I do not control.

What changed in implementation

• Rust + Tauri stayed as the desktop shell for predictable native performance.
• A Python sidecar handles transcription model execution.
• Product decisions started from latency and reliability, not feature count.

Current state

Flustro is still evolving, but local-first remains non-negotiable in the roadmap.