Windows support — OCR pipeline, packaging, and foreground metadata

[JTLR]

Hey — I've been using Familiar daily on Windows and built out a full Windows port on a fork. Wanted to gauge interest before opening PRs.

What's included:

• Windows OCR pipeline using the built-in Windows.Media.Ocr API (WinRT via PowerShell 5.1) — mirrors the Apple Vision OCR pattern with layout region inference (HEADER/SIDEBAR/CONTENT from word bounding boxes)
• Foreground window metadata capture — captures active window title, app name, PID, and URL extraction for browsers at screenshot time. This addresses Capture ground-truth app/window/URL metadata per frame #11 and complements the on-screen app detector added recently (which is macOS-only)
• NSIS installer packaging — npm run dist:win builds a Windows .exe installer with PowerShell scripts properly unpacked via extraResources
• Cross-platform Launch at Login — replaces the macOS-only openAtLogin with a cross-platform setting that works on both platforms
• Platform-aware defaults — capture interval, UI text ("System Tray" vs "Dock"), extractor type auto-detection, settings normalisation for windows_ocr type

The OCR quality is solid for screen text — I've been running it daily for a few weeks. All output follows the familiar-layout-v0 markdown format with real frontmatter metadata (app, window title, URL).

Happy to break this into focused PRs if you're interested. Thinking something like:

1. Core Windows OCR + packaging
2. Cross-platform Launch at Login + capture interval
3. README updates

Fork: https://github.com/JTLR/familiar/tree/windows-support

[Hyaxia]

Hey Joe, im super happy you've found Familiar helpful.
Would you mind sharing different use cases?

Regarding the windows support, super cool that you've made it work for you!
Its definitely something we'll do eventually but for now we want to wait with it until we get the experience good enough.
Thank you!

[JTLR]

Hey, thanks for the quick response!

Happy to share how I'm using it. My main use case is feeding screen captures into a personal productivity system I've built on top of Claude Code. Familiar runs in the background throughout the day, and at end-of-day I have a script that processes the captures to summarise what I worked on, surface any tasks or follow-ups I might have missed, and clean up the raw files. It's basically an automated "what did I actually do today" review.

The window metadata capture (app name, window title, URL) has been really valuable for that. Knowing I was in Teams vs VS Code vs a browser gives the AI enough context to categorise activities without needing perfect OCR on every frame.

I also built a health check that monitors OCR success/failure rates over the last 24 hours, which surfaces warnings in my daily dashboard if the pipeline is degraded. Useful for catching issues early since the captures are only valuable if they're actually working.

On the Windows support side, I've got a clean PR split ready whenever you're open to it:

1. Core Windows OCR + packaging (PowerShell-based Windows.Media.Ocr, NSIS installer)
2. Cross-platform launch at login + configurable capture interval
3. README update

No rush on that though. I know you want to get the experience right first, and I'm happy using my fork in the meantime. Just wanted you to know the contributions are ready when the time comes.