Your AI forgets everything. Loomis doesn’t.

⌥ N to jot, paste, or drag anything into a floating panel. ⌥ Space to ask for it back. Everything stays on your Mac, threaded with what it’s related to, and open to whatever AI you already use.

macOS · early builds this fall · no spam, one email when it’s ready.

⌥ N ⌘ ⏎ ⌥ Space ⏎

CaptureLoomis

youtube.comAndrej KarpathyMachine Learning Advice for Beginners

Tab to nest · drop images anywhereFlush ⌘ ⏎

Kept. 3 captures threaded into Machine Learning · deduped · embedded locally

Karpathy’s argument: you don’t understand a model until you’ve implemented it. Build a tiny one from scratch before reading papers.⁠[1]

[1]Andrej Karpathy · Machine Learning Advice for Beginners12:34

⏎ open at 12:34⌘ C copy link⌘ ⇧ C copy answer

Your whiteboard sketch from Tuesday: a three-layer net with the loss circled.⁠[1]

[1]whiteboard.jpgTue · ⌥N

⏎ open⌘ C copy image

Andrej Karpathy · Machine Learning Advice for Beginnersyoutube.com

…and you don’t really understand it until you’ve implemented it yourself, from scratch.

Opened at the cited moment12:34

⌥ N

Capture without leaving

A floating panel over whatever you’re doing. Type, paste a link, drag an image, nest with Tab. ⌘ ⏎ and it’s out of your head and into your vault.

⌥ Space

Ask, get the source

Spotlight for your own mind. Type a half-remembered thing, get the answer with the link it came from, copy it, move on. Dedup and auto-threads keep it coherent underneath.

MCP

Let your AI ask too

The same vault, served to Claude Code, Cursor, and Claude Desktop with citations. loomis_ask "what did I save about X?". No more re-explaining yourself.

Your AI forgets everything. Loomis doesn’t.

Capture without leaving

Ask, get the source

Let your AI ask too

Five concepts a day. Until they stick.