# Windows resident launcher Version 0.1.7 adds a global launcher workflow. `Ctrl+Shift+Space` restores the window and selects the search field. Escape dismisses a popover first, then hides the native window. Closing the window also hides it, keeping prepared search readers and the shortcut alive. Hidden audio/video previews pause. The tray provides Show searchi and an explicit Quit action. `--background` starts without showing the main window; a duplicate background launch does not interrupt an existing session. Karl's Windows Dev installation now has a login shortcut using this flag and the existing CPU-only VM wrapper. Production acceleration remains automatic; the wrapper is a VM constraint. Validation: three native instance tests and strict desktop Clippy passed. Frontend build, 62 tests and formatting passed. The release executable compiled and was installed as 0.1.7, then observed with `--background`. Settings remained byte-identical after a consistent SQLite backup; database quick_check passed. The updated normal speech runtime was verified/staged and passed its installed self-test with forced CPU. The built-in key uses the existing focus handler. Actual native keyboard/tray interaction was not automated; available computer control does not support it in this session. These checks do not close broader GPU, macOS, clean-machine or search-quality gates. The final NSIS installer is published with [v0.1.7](https://tea.karlcom.de/karl/searchi/releases/tag/v0.1.7), from application commit `24e80edaf9c1a50cb3dde97ef21b211b52c66f73`: 101766307 bytes, SHA-256 `5ca8ad982113d086190ecf8ea684a504a4277099a377124bdffa8ca9fed6595e`. Public full-byte download matched. Extracted speech inventory/provenance (31 records) and extraction helper (two records) passed. The normal installer exited successfully on Windows Dev; installed executable matches extracted bytes: `e0cef05acfcea8b2299f118daf3d866ca2f8950704c01d8673812b2ae0139185`. The CPU-only wrapper and login shortcut were retained, and the app restarted with `--background`. The prepared Google query worker was observed using CPU.