ScreenMind
Remember everything. Share Nothing
Start new thread

ScreenMind - Remember everything. Share Nothing.

by•
Unlike Microsoft Recall (closed, needs $1000+ hardware) or Screenpipe (requires separate OCR + Whisper + external LLM), ScreenMind uses one local Gemma 4 model for everything — screen vision, audio transcription, chat, and reasoning. No cloud, no telemetry, no accounts. Captures only when content changes. Search or chat with your full screen history in plain English. Build automations in Markdown. Runs on a 4GB GPU. MIT licensed, pip install.

Add a comment

Replies

Best
Maker
📌
Hey PH! I launched ScreenMind back in May and honestly got zero traction. Since then I've put in 73 commits — Model Hub UI, desktop launcher, PyPI packaging, 166 tests, a proper landing page. The core idea hasn't changed: one local model that sees your screen, remembers everything, and lets you search/chat over it. No cloud, no telemetry, no subscriptions. I built this because I kept losing context — tabs I closed, errors I saw, things people said in meetings. Microsoft showed the world wants this with Recall, but I wanted something open and private. Happy to answer any questions. Feedback on the landing page especially welcome — I had fun building it: screenmind.pages.dev

the local-only angle is genuinely appealing, especially for handling sensitive screens at work. one thing that would seal the deal for me is adding a way to manually redact or exclude specific apps or windows from capture, since I don't want every banking session or private DM ending up in searchable memory even if it stays on my machine.

Maker

 Already there! Settings has a blocked apps list — add any app and it'll never be captured. On top of that, there's an automatic sensitive data filter that redacts credit card numbers, API keys, passwords, and SSNs from the text before it ever touches the AI model. So even if a banking app slips through, the actual sensitive data gets stripped out.

pip install was painless and the "what was that error 2 hours ago" demo actually pulled up the right stack trace from earlier. Wish the local model was a bit snappier on my aging laptop, but the privacy angle makes it worth the wait.

Maker

 If it feels slow, try switching to "Fast" mode in Settings — it skips the heavy thinking pass and brings analysis down to ~12s per frame. Also check the Model Hub and make sure you're on E2B (the smallest variant). Should help a lot on older hardware.

Searched back through my screen for an error from earlier and it actually pulled up the exact stack trace without me having to scroll anywhere. Loving that everything stays local too.

Maker

 Glad it worked for you! exactly why i build it

This is honestly really cool, especially the local-only angle since I don't trust cloud tools with my screen. One thing though, I'd love a way to pause or exclude specific apps from being captured, like games or banking stuff. A simple blacklist with a system tray toggle would make it feel way less invasive day to day.

Maker

hi  , thanks! And what makes you think I didn't add that (jk) — it's already built in! You can add apps to a blocklist in Settings, and they'll never get captured. There's also auto-pause for heavy apps like games so it stays out of your way.