Building a privacy-first dictation tool

“Privacy-first” is easy to say and harder to execute when releasing quickly.

For Murmr, it forced a few concrete decisions early:

• keep capture and refinement on Apple-native frameworks
• avoid building account infrastructure into the core workflow
• keep text delivery simple (clipboard, paste, files)

Tradeoffs are real

A privacy-first approach means I cannot outsource difficult product moments to external services by default.

That affects:

• error handling
• fallback paths
• cross-device behavior

It is more work, but the architecture is cleaner and easier to explain.

Reliability before growth hacks

Right now the priority is not “growth loops.” It is confidence:

• capture should be dependable
• refinement should be predictable
• output should be ready to use

If those are strong, everything else compounds naturally.

Why this matters

Dictation should feel like part of your operating system, not a black box.

That is the bar Murmr is built against.