Skip to content

macOS · Apple Silicon · Bring your own AI key

Speak once.
It already knows where you are.

Press ⌥ Space in any app. Smarn reads the window you're looking at — the terminal, the error dialog, the email thread — and answers in place. No copying, no tab-switching, no re-explaining.

Free during early access. Everything you capture stays on your Mac.

The problem

Every AI micro-task costs six hops.

A question the model answers in five seconds takes thirty to sixty seconds of copying, switching, pasting and re-explaining — and breaks your flow every single time. The assistant doesn't know what's on your screen, so you become the clipboard.

  1. 1copy the content
  2. 2switch to a chat app
  3. 3paste
  4. 4re-explain the context
  5. 5read the answer
  6. 6copy it back · switch back
  7. →⌥ Space, ask, done.

How it works

Bring the AI to the context, not the context to the AI.

  1. Trigger

    ⌥ Space, or hold Right Option to talk.

  2. Capture

    Front window, selected text, terminal buffer — plus a screenshot read on-device.

  3. Ask

    Type or speak. Capture is already done before you finish.

  4. Answer

    Streamed into a small overlay as typed blocks: command, code, draft, steps.

  5. Act

    Copy, insert, or run — behind a deterministic risk gate. Nothing runs without you.

Capture is taken from the window that was frontmost before the overlay appeared. Smarn never screenshots itself, and a screenshot alone can't give you exit codes or scrollback — so it captures structured context first and uses vision to fill the gaps.

Use cases

Built for people who live in a terminal, a browser and a mail client.

Terminalcommand

“Why is this container restarting?”

docker logs --tail 200 <id> + the likely cause, risk-rated, one click to copy.

Build or test failurecode

“Why is this failing?”

The stack trace read off your screen, an explanation and a fix — or a hand-off to Claude Code with the right cwd.

Browsertext

“Explain this section simply.”

Uses your selection if you made one, otherwise the visible page. Cites the part it used.

Any error dialogsteps

“How did I get here and how do I fix it?”

Screenshot + on-device OCR of the modal → diagnosis and copyable steps.

Maildraft

“Reply saying we can do Thursday, keep it short.”

A draft in the thread's tone, ready to paste.

Memorysearch

“That Postgres vacuum thing I read last week?”

Search everything you've captured — by app, by text on screen, by date. No AI needed.

Local memory

Google Photos, for everything you've seen.

Every capture you ask about is kept on your Mac — the image and the text on it — and indexed for search. Type whatsapp and every WhatsApp screenshot appears. Type half an error message and find the moment you saw it.

  • ● Search runs on-device. No AI, no upload, no account.
  • ● Only redacted text is stored — keys and passwords never touch the disk.
  • ● Forget any capture, or everything, in one click. ⌥ ⇧ Space to open.

Privacy, by construction

Nothing leaves your Mac without you knowing.

Captures only when you trigger it

No always-on recording. No background memory. Pixels are read the moment you ask — never before.

Reads on your Mac

Screenshot OCR and speech recognition run on-device with Apple's frameworks.

Scrubs secrets first

API keys, tokens and passwords are redacted in the Rust core before anything leaves the machine.

Sends only to your AI account

The captured context and your question go to the provider you connect, under your own key and its terms. Never to us.

Never runs commands on its own

You click Run. A deterministic risk gate — code, not a prompt — decides what may run at all.

Memory never leaves the machine

Your captures are files and a database in your own Library folder. Delete them whenever you like.

Read the plain-language version: smarn.app/privacy

Put it on your Mac.

Early-access build. Signed and notarized. Bring your own Anthropic API key — it's stored in your Keychain and only ever used to make the request.

macOS 14 or later · Apple Silicon · updates itself

Coming next: ask questions across your memory, spoken answers, more AI providers.