A quiet second brain that lives in your menu bar. It watches what’s on your screen, understands what matters, and writes it down for you.
11:42
InfraClaude API billing limits
Tier 4 = 8M input tokens / min. Doubles after $4M lifetime.
11:15
DesignOnboarding flow V3
Three screens. Progressive disclosure on step 2.
10:38
Meetings1:1 with Priya — Tuesday
Shipping SSO audit log by Thursday.
09:52
ReadingarXiv 2504.07821
Sub-quadratic attention via learned pivots.
09:10
DesignFigma — design tokens cleanup
Merged 47 colors down to 9 semantic roles.
11:42 · Today · Infra
Anthropic’s tier-4 cap is 8 million input tokens per minute. It doubles automatically after $4M lifetime spend, and concurrent request limits double again at tier 5.
For our batch job that means ≈3× headroom at current volume with no PR to Anthropic.
Features
ScreenMind is designed to disappear. Everything it does happens in the background so you can stay in the work you actually care about.
Always there. Never loud.
One small glyph. No dock icon. No window to manage. Quiet until you ask.
Captured 2 min ago · from Safari
Anthropic tier-4 cap = 8M tokens/min.
Saved to today · tagged #claude
Plain English, on your timeline.
What did I decide about SSO on Tuesday? ScreenMind pulls the meeting, the ticket, and the DM into one paragraph.
Wakes on window change or long-read. Sleeps otherwise. Your battery won\u2019t notice.
Claude, GPT, or a local Ollama model. One toggle, no fallback.
Encrypted SQLite on disk. No account. No ScreenMind server.
Password fields and banking UIs are masked on-device, before anything ships.
A glanceable summary of the last eight hours \u2014 what you read, wrote, and decided.
\u2318\u21E7Space opens the question box from any app. Answer in under a second.
How it works
ScreenMind is a small program that does a small set of things, very consistently, on your machine.
ScreenMind samples your screen a few times a minute. Not a video. Not a keylogger. Just enough to know what you’re looking at.
A small, fast model on your Mac decides whether a moment is worth keeping. Most aren’t. The ones that are, get written down.
A weekly wiki, a searchable timeline, and a question box. Ask the last month anything, in plain English.
A peek inside
The app is small on purpose. It gives you back exactly three ways to look at your own attention.
Timeline
A calendar of moments worth keeping.
Ask
Plain English over the last month.
This week
Wiki
Every topic you touched, summarized.
I asked it what I decided about SSO on Tuesday. It pulled the Zoom note, the Linear ticket, and the Slack DM into one paragraph. That’s when I understood what this app actually is.
J. M.
Head of product · fintech · alpha tester
2,011 on the beta.
Engineers, PMs, researchers.
Privacy
FAQ
No. ScreenMind doesn’t record video or audio. It samples the screen a few times per minute and extracts text. The images aren’t kept.
Your notes live in a single SQLite file on your Mac, encrypted at rest. There is no ScreenMind server. We can’t see what you capture because there is nothing to see.
Anthropic Claude, OpenAI, and any Ollama model running locally. You can also point ScreenMind at a custom OpenAI-compatible endpoint.
Capture is event-driven — on window focus changes and quiet moments. On an M-series Mac in a typical workday, CPU usage is under 1% averaged.
During the private beta, ScreenMind is free. Pricing will land with 1.0 and will not require an account.
Join the beta below. We email you a signed DMG with a SHA-256 checksum, direct from us. No App Store, no account.
Private beta
Leave your email. We send the DMG and the SHA-256 checksum. That’s all.