Drop in everything you love. Amyra sees it, sorts it, finds it — and answers the questions you’d never ask a cloud AI.
Runs on your Mac
Files and model live on your own silicon.
Works offline
Planes, cafés, dead zones — all fine.
No account
No sign-up, no login. Just open it.
$0 / month
Free to keep. No trial, no subscription.
the first thing you’ll feel
meet your assistant
A full assistant, running entirely on your Mac. One chat, three modes: your board, your writing, and the documents you’d never upload. Offline, if you like.
Ask Amyra anything
on-device AI — nothing leaves your Mac
Message Amyra
live demo — click a chat tab to replay a conversation
01 · It acts
Turn Canvas mode on, and Amyra isn’t a chatbot anymore — it’s a pointer on your board. It selects, arranges, labels and leaves notes in front of you, one undoable batch at a time.
select → arrange → label → note
Amyra
select → arrange → label → note · one undoable batch
02 · It sees
Every image, video and PDF is read, described and embedded — on your Mac. Type a feeling, and Amyra finds the pictures that match it. No tags. No filenames. No cloud round-trip.
Apple Vision OCR + CLIP semantic search + keyword fusion
3 matches · 11 ms · on-device
03 · It sorts
Cluster your board by topic, color, mood, or type — or by what simply looks alike. Amyra reads the content of every item, so the groups actually make sense.
topic · color · visually similar · mood · type
04 · It notices
Drop a handful of files and Amyra quietly studies them. When it spots something — five pricing pages, a set of near-duplicates — it offers one useful move. Accept it, or ignore it. It won’t nag.
deterministic signals decide · the model only labels
These 3 look like cute pet content.
Group them into a labeled grid?
05 · It gathers
A browser docks beside your board. Hover any image on any site you’re signed into and pin it straight to the canvas — it’s indexed before you’ve scrolled on.
My Board
06 · It watches
Any board can watch a folder on your Mac — Screenshots, Downloads, a client drop. Every new image, video and PDF imports itself into a tidy grid, indexed and searchable before you’ve switched windows.
images · videos · PDFs — synced the moment they land
saved to the folder → on the board · indexed
07 · You write
Open a rich note anywhere on the canvas — headings, lists, checkable tasks, images inline. Markdown shortcuts format as you type, so the brief, the shot list and the half-formed hunch live right next to the references that sparked them.
type ## or - [ ] — it formats itself · ⌘1–4 headings
a real editor on the canvas · headings · tasks · inline images
and above all — it’s yours
Your taste is not a dataset.
Under the hood
Reasoning & vision
Gemma 4
E4B & 12B on Apple Silicon via MLX. Apache 2.0. 128K–256K context, 140+ languages.
Reading
Apple Vision OCR
Every screenshot, mockup and PDF read on-device — dense UI text included.
Visual search
CLIP · Core ML
ViT-B/32 bundled in the app. Text→image and image→image similarity, in milliseconds.
Generation
2.0× faster
Multi-token prediction with speculative decoding. Measured on-device: 64 → 128 tok/s.