A mind of its own

Your screen,
remembered.

Thousands of screen-moments, connected like neurons — running entirely on your machine.

pip install screenmind
1,500+ downloads · MIT · runs on 4 GB VRAM
Understand

Every synapse
is a moment.

One local model reads every frame — apps, text, mood, layout — and remembers what it means.

Learn how it works ↯
Automate

Agents that live
in your memory.

Drop a Markdown file and it runs on your screen history — no code needed.

Build an agent ↯
Private

All in your
head. Literally.

Zero cloud. Zero telemetry. Your mind stays yours.

See how it stays private ↯
Overview

What ScreenMind does

ScreenMind captures your screen, understands it with a single local model, and turns it into a searchable, chat-able memory of everything you do — all on your machine.

1,500+
DOWNLOADS
~180
GITHUB STARS
4 GB
MIN VRAM
MIT
LICENSE
📸
Smart capture
Content-change detection with perceptual hashing — it captures when the screen actually changes, not on a dumb timer.
🔬
Gemma 4 vision
Every frame analyzed into structured info: app, activity, mood, a rich scene description, and layout regions.
🔍
Hybrid search
Semantic embeddings (MiniLM) fused with FTS5 keyword search. Find things by meaning, not just exact words.
💬
Chat with memory
Conversational RAG over your history with follow-ups, and a vision fallback that reads the screenshot when text isn't enough.
🧠
In-app Model Hub
Download, switch and manage models from the UI with live progress — no terminal. Chat/Summary stay locked until the model is ready.
🎙️
Voice memos
Hold a hotkey and talk — Gemma 4's native audio encoder transcribes it, with a screenshot captured alongside.
🎤
Meeting transcription
Auto-detects Zoom/Teams/Meet/Discord, records mic + system audio, transcribes, and writes a structured summary.
📊
Analytics & Rewind
Category breakdowns, top apps, an hourly heatmap, and a timelapse player to scrub through your whole day.
🤖
Agents
Automations in plain-English Markdown or full Python. Drop a file in the agents folder and it runs on a schedule.
🔌
MCP + integrations
Expose your history to Claude/Cursor/VS Code over MCP; push summaries to Obsidian, Notion, or webhooks.
🔒
Private by design
100% local, zero telemetry, encryption at rest, and automatic redaction of cards / API keys / passwords.
⌨️
System hotkeys
Bookmark a moment, pause/resume capture, or record a voice memo — all with global shortcuts.
🔔
Smart notifications
Gentle nudges — distraction alerts, break reminders, and keyword auto-bookmarks like 'git push' or 'deploy'.
🧑‍💻
Dev-aware
Detects your git repo, branch and recent changes while you code, so coding activity is tracked with real context.

How it compares

Microsoft showed the world wants screen-aware AI with Recall — but it drew heavy privacy backlash. ScreenMind is the open-source, fully-local alternative to Recall and Screenpipe.

ScreenMindScreenpipeMS Recall
LicenseMIT — fully openSource-available (paid for commercial)Proprietary
CostFree foreverFree personal / paid commercialNeeds $1000+ Copilot+ PC
PrivacyZero network, zero telemetryLocal-first, optional cloudTelemetry opt-in
Min hardwareAny GPU ≥4 GB (or CPU)8 GB RAM, modern CPU40 TOPS NPU + 16 GB RAM
AI architectureOne model — vision+audio+reasoningOCR + Whisper + external LLMProprietary NPU model
Audio / meetingsNative (Gemma audio encoder)Whisper-basedNot supported
SearchSemantic + FTS5 hybridSemantic + keyword + a11ySemantic only
Chat with memoryFull RAG + follow-ups
AgentsMarkdown + Python + MCPPipes (TS) + MCP
EncryptionAES (Fernet) + OS keyringOptionalTPM + BitLocker
PlatformWindows / macOS / LinuxWindows / macOS / LinuxWindows 11 only
Fun fact: the whole thing was built and benchmarked on a 4 GB GTX 1650 — the model literally spills into system RAM and still works. Any GPU with ≥6 GB runs it 3–5× faster.
GitHubPyPIWalkthroughRaise an issueXLinkedInEmail
ScreenMind — 100% local, MIT-licensed.
↯ back to the brain