3D desk ↗
← Jędrzej (Jed) Mejzner

02 · App

VisionSpeech

Your voice, typed. On device.

A dictation app I designed and built for myself after Vision Pro could not take dictation in Polish. Everything runs on local models, so your voice never leaves your device.

Role
Design & code
Platforms
visionOS, macOS
Year
2025 – now
Status
In progress, Mac first
VisionSpeech website on a phone
Why

Vision Pro had no Polish dictation, and the dictation apps I used every day meant another subscription and a few extra steps. I wanted one tool that worked the way I expect a good product to: instant, quiet and polished down to the last detail.

VisionSpeech homepage
The orb

On Vision Pro the app is always on: a small orb you place side by side with whatever you are working on. It listens, thinks and lands the text without pulling you out of your flow. Designing for space is different from designing for a screen, and the goal here was a sidekick, not another window.

VisionSpeech section with a green check orb
Private by default

Nothing goes to the cloud, so it is private, fast and works anywhere.

What happened

It started on Vision Pro and worked well enough to grow onto more platforms. Friends started using it, so I am now shaping it into a product, starting with the Mac.

What I learned

Models from Anthropic and OpenAI changed my role. I no longer stop at how an interface looks; I can take an idea all the way to a tool that works.

Who it is for

People who value privacy, speed and reliability. Every feature runs entirely on your own machine.