# ScreenMind

>  AI-powered screen memory — captures, analyzes, and lets you search/chat your screen history. Powered by Gemma 4 . 100% local, 100% private.

- **Licence:** MIT
- **Latest version:** v0.2.5
- **Open source:** oss
- **Languages:** JavaScript, Python
- **Tags:** Privacy, Multi modal, Localai, Productivity, AI, Python, Fastapi, Gemma, Llama Cpp, Recall Alternative, Screenrecorder

## Description

ScreenMind is a privacy-first, self-hosted AI memory application that continuously captures your screen activity, analyzes each frame using a local Gemma 4 multimodal model (vision, audio, and reasoning), and builds a searchable, chat-able timeline of everything you do on your computer. It runs entirely on your hardware — no cloud dependencies, no telemetry. Key capabilities include smart content-change detection capture (not timer-based), hybrid semantic + keyword search across screen history, conversational RAG chat with follow-up support, voice memos and meeting transcription using the model's built-in audio encoder, an in-app Model Hub for downloading and switching Gemma 4 variants (E2B/E4B/12B) with quantization options, a Markdown/Python agent platform for automations, an MCP server for integration with AI assistants, and integrations with Obsidian, Notion, and webhooks. The web dashboard at port 7777 provides timeline, chat, search, analytics, day rewind timelapse, and settings. Data is stored in local SQLite with optional AES encryption at rest, sensitive data (credit cards, SSNs, API keys) is auto-redacted before storage, and a PIN lock secures the dashboard. It supports Windows, macOS, and Linux (X11 and Wayland via grim/XDG Portal). Minimum hardware is a GPU with 4GB+ VRAM (CPU-only works but slower), Python 3.10+, and ~5GB disk space. The project is MIT licensed and currently in pre-1.0 beta (v0.2.5).

## Links

- [Homepage](https://github.com/ayushh0110/screenmind)
- [Website](https://screenmind.space/)

---
Source: https://selfhost.directory/project/screenmind