Show HN: Running a vision model on every screenshot on-device

Category: ai-ml

Tags: vision-model, privacy, screen-capture, local-ai, memory

Score: 7.5/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 7)

ScreenMind is a privacy-first, open-source alternative to Microsoft Recall that captures screenshots, analyzes them locally with the Gemma 4 multimodal model, and builds a searchable AI memory. Its innovative combination of on-device vision, audio, and reasoning with a comprehensive agent platform and MCP integration makes it a powerful tool for personal productivity and automation.

Target audience: backend devs, devops, data engineers

Repository: https://github.com/ayushh0110/ScreenMind/blob/main/README.md · Python · MIT · 123 stars

View on Hacker News