ScreenSearch captures changed screens within the settings you choose, reads useful text with UI Automation and OCR, and makes that history searchable by keyword, meaning, or plain-language question. Summon the Flow overlay with a hotkey and return to the source frame without switching apps. Capture, OCR, storage, search, and AI inference stay on your Windows PC.
v0.4.2 · 14.6 MB installer · Windows 10/11 x64 · What’s new
Every major OS is adding a memory of what you’ve seen. That’s genuinely useful — and genuinely worth being careful about. ScreenSearch keeps the memory local, trims away invasive capture surfaces, and focuses recall on the content you were actually working with.
Your context is scattered across a hundred tabs, apps, and windows a day, and most of it is gone the moment you close it. You remember seeing the thing — the error, the figure, the snippet — but not where. Bookmarks and notes only help with what you thought to save.
ScreenSearch captures changed screens within your enabled settings, reads the foreground window through UI Automation with OCR fallback, and keeps searchable source frames for the retention period you choose. Default recall filters noisy chrome, while the Flow overlay lets you search or ask from your current app.
Local capture, on-device intelligence, fast recall, and carefully planned extension points. Built in Rust, quiet by default.
Continuous capture plus UI Automation and native Windows OCR read the foreground window, then an attention-first text layer keeps default results from being dominated by toolbar labels and static chrome.
Ctrl+Alt+Space opens the Flow overlay over whatever you are doing. Type a keyword, ask a question, jump to a Moment, or dismiss it with Escape when you are back in flow.
Smart Answers turns your history into something you can talk to. Ask “what did I work on this week?” and get a written answer grounded in your real activity, with source frames you can inspect.
The Command Deck gives you recent frames, capture state, enrichment progress, Recall reports, and daily summaries for the “what did I actually do today?” moment.
An optional local vision model looks at each screenshot, describes it, reads text straight from the image, and classifies the activity — all on-device. No screenshots are sent anywhere.
Hybrid search fuses keyword (FTS5) and semantic vector results with Reciprocal Rank Fusion. Search “database config” and it can surface the screen that said “connection_string.”
AI answers run through the bundled local llama.cpp sidecar. Opt-in localhost API, export, and MCP surfaces let trusted local tools work with your screen memory.
What we’re building next — and what we’re still weighing.
A pull-based context resume surface for the “what was I doing before this detour?” moment.
A hotkey to capture now and attach an optional note when you spot something you intend to act on later.
An opt-in localhost API, JSON export, and thin MCP wrapper for trusted local tools and agents.
From developers to designers, ScreenSearch helps professionals recover context from captured work.
"What did I change in the auth flow yesterday?"
Search your coding history—documentation, StackOverflow answers, terminal commands, code reviews.
"Where did I see that citation?"
Build a searchable archive of research papers, articles, and sources visible during enabled capture.
"What error message did the customer send?"
Search captured support tickets, error logs, customer conversations, and shared screens.
"I saw the perfect color palette but can’t find it."
Create a visual reference library from the work captured within your settings.
"Summarize which dashboards I looked at before the board call."
Search charts, graphs, and data visualizations across all your tools.
"I read a great quote but didn’t save the source."
Search articles, posts, and documents that appeared during enabled capture.
Four quiet steps, running in the background on your own machine.
ScreenSearch takes a snapshot every few seconds (you set the interval), only when the screen actually changes. Multi-monitor included. It sits in the system tray and stays out of your way.
Native Windows OCR pulls text from each captured frame, and an optional on-device vision model can describe what’s on screen and classify the activity. These capture and analysis steps stay local.
Captured text and source frames land in a local database, indexed for keyword and semantic retrieval.
Type a keyword and inspect a matching moment, or ask a plain-language question and get an answer linked to the retrieved source frames.
Capture, OCR, storage, keyword search, and AI inference stay on your Windows PC, with app exclusions, retention settings, and pause/delete tools in your hands. Model downloads and signed update checks use the network without uploading your screen history.
Everything you need to know about ScreenSearch.
ScreenSearch records changed screens within your capture settings, reads visible text with UI Automation and Windows OCR, and indexes it locally. Open the Flow overlay with Ctrl+Alt+Space, then search by a word, an idea, or a plain-language question and inspect the matching source frames.
Current releases are available free as a direct Windows installer download, with notes for every version on the changelog, and do not require an account.
Capture is limited to changed screens while recording is enabled. You can exclude apps, pause capture manually, rely on lock-screen auto-pause, delete captures, and choose retention settings.
Because sensitive information can still appear in an allowed app, exclude password managers, banking apps, private browsing sessions, and any other surface you do not want stored.
Both products make past screen activity searchable. ScreenSearch is an independent, open-source tool for Windows 10 and 11 that does not require Copilot+ hardware. It emphasizes user-controlled exclusions and retention, hybrid keyword and semantic search, a global hotkey overlay, and bundled local AI inference.
Microsoft Recall is a separate Microsoft product with its own hardware, security, and availability requirements.
No. Current releases do not expose a remote inference-provider setting. Answer generation runs through the bundled local llama.cpp sidecar. Model downloads and signed GitHub release checks use the network, but they do not upload your captured screen history.
See the privacy page for the full data-flow explanation.
Capture, read, index, and ask — all running locally on your own machine.
Selected screens from a normal day’s use.
Ready to search — and ask — your entire screen history?
Download for WindowsWindows 10/11 · No account · Local-first by default