AI that stays on your device
By the end, you can run AI in your browser, on your own device, with no cloud call for the actions it covers.
4 steps · 1-minute video · Recorded November 5, 2026
Watch the full walkthrough · about 2 minutes
-
Settings, Local AI models
It lives in Settings, under Local AI models. First, Glassy checks your browser. Full WebGPU support means you're ready for every model.
Go deeper: Without WebGPU
-
Without WebGPU
Without WebGPU, the writing and voice models won't run on your device. Related notes and tag suggestions still can, just more slowly.
-
-
Set up recommended models
Then get the models. The recommended set is three models, about two gigabytes in all, and one click sets it up. Or choose them yourself.
Go deeper: Writing, voice, related notes · Pick a size that fits
-
Writing, voice, related notes
Qwen helps with your writing. Whisper turns voice recordings into text. And the small embedding model is what finds related notes.
-
Pick a size that fits
Choosing yourself? Each model lists its size and the memory it wants. Lite suits lighter machines. Enhanced wants sixteen gigabytes or more, for the best rewrites and summaries.
-
-
Prefer local AI
Last, switch on Prefer local AI. With it on, eligible actions run on your device instead of the cloud. That includes cleanup, summaries, rewrites, and voice transcription. That's the whole setup.
-
Made on your device
Here's what that looks like in a note. The suggestion is marked On-Device, so you know where it came from. And you see the change before anything moves.
Go deeper: Apply or discard
-
Apply or discard
Keep it with Apply Changes, or let it go with Discard. Changed your mind after applying? Control Z puts it back.
-