0.3.0 — live mic monitor & browser calls

Ported straight from the Mac's v1.3.0 — the tray panel now shows you what's happening and can capture browser calls.

• 🎙️ Live mic monitor. The panel shows, in real time, which app is currently using your microphone — even one that isn't in your call list.
• 🌐 Browser calls (Google Meet, Teams web, Whereby). When a call runs in your browser, the monitor shows it ("Chrome is using your microphone") with an Always record this app button — one click adds it to the allowlist and restarts the watcher, so browser calls are captured from then on.
• Caveat documented: recording a browser taps its whole audio — mute other tabs during calls.

Compiled green on windows-latest CI. Runtime capture of browser child-process audio (WebView2/Chromium renderers) is the main thing still to verify on real hardware — reports welcome!

0.2.0 — the tray app grows up

The tray app got its first big field-test-driven upgrade, built and live-tested inside a Windows 11 ARM64 VM:

• 🎛️ Full macOS-style settings panel — language / transcription (Whisper · Parakeet · Groq) / AI summary as segmented controls with ⓘ explainers, note-content and destination checkboxes (Nextcloud/Notion), storage paths incl. external-mirror toggle, ntfy push. Saves atomically, preserves unknown config keys.
• 📞 Active-call card with scrolling two-track waveforms (you + caller), participant editor with +/−, discard link.
• 🌒 New dark theme + real app icon; panel now dismisses on focus loss / Esc like a proper tray popover (also fixes the panel hang).
• 💡 Idle hint — the panel now tells you it's waiting for a running call (mic active); just opening Teams/WhatsApp is not supposed to trigger anything.
• 🧹 README: collapsible FAQ (incl. "I opened Teams and nothing happened?"), uninstall section — lessons from the first Mac-side community tester.

Compiled green on windows-latest CI; panel behavior verified live in the VM. Still experimental — bare-metal reports welcome!

0.1.1 — first field test: two-track capture proven

The first real-Windows field test (Windows 11 ARM64 24H2, in a UTM/QEMU VM) ran the full chain — and found the one bug that static review and CI could never catch:

• 🐛 Fixed: process loopback returned 0x88890021 (AUDCLNT_E_INVALID_STREAM_FLAG). IAudioClient::Initialize for AUDIOCLIENT_ACTIVATION_TYPE_PROCESS_LOOPBACK must pass AUDCLNT_STREAMFLAGS_LOOPBACK (as the Microsoft ApplicationLoopback sample does). Without it, capture never starts.
• ✅ Verified after the fix: a 440 Hz test tone played through the default output was recorded via process loopback at −0.1 dBFS peak (48 kHz stereo float32), with the microphone captured in parallel — the two-track core of CallNotes works on Windows.
• ✅ calltap procs and calltap setup pass; native ARM64 build works.
• 🧹 Error messages no longer blame the Windows build for flag errors; test instructions now use record --exe <player> (without --exe, silence is by design).

Still experimental: tested in a VM, not yet on bare metal, and not yet with a real VoIP call. If you have a Windows machine, be the second field test — installer/install.ps1 or bootstrap-test.ps1, and report in the issues.

CallNotes for Windows 0.1.0 — experimental

First public cut of the Windows sibling of CallNotes for macOS.

⚠️ Experimental / looking for testers. This build compiles cleanly on GitHub's windows-latest runners (CLI + tray app + pipeline), but has not yet been run on real Windows hardware. If you have a Windows machine: clone, run installer/install.ps1, make a WhatsApp/Zoom/Teams call and tell us what happens in the issues — the first field report gets eternal gratitude.

• Two-track call capture via WASAPI process loopback (no virtual drivers)
• Same pipeline as the Mac original: whisper.cpp or Parakeet TDT v3 or Groq, sherpa-onnx speaker separation, your choice of AI (Claude CLI / any OpenAI-compatible / off)
• Same config & note formats as the macOS version
• Tray app with live call state, participants popup, don't-record button, DE/EN