Merge pull request 'Feature chore bug 010' (#25) from feature_chore_bug_010 into main

Reviewed-on: #25

README.md

@@ -11,36 +11,39 @@ on-device acceleration (**NPU → GPU → CPU**), labels speakers, structures th
 Markdown notes, and optionally augments them with a locally hosted LLM (Ollama). Audio and
 transcripts **never leave the machine** unless you explicitly configure a destination.
 
-> **Status: working application (v0.5.2).** Capture (system audio **+ your microphone**, with a
+> **Status: working application (v0.7.3).** Capture (system audio **+ your microphone**, with a
-> live **dual level meter**), transcription (CPU / Intel NPU / Vulkan GPU) with a **fluid live
+> live **dual level meter** and an **M-key mic mute**), transcription (CPU / Intel NPU / Vulkan
-> transcript**, speaker diarization, storage + crash recovery, local-LLM summaries, AI tags,
+> GPU) with a **fluid live transcript**, speaker diarization that labels **You** live from your
+> own mic channel, storage + crash recovery, local-LLM summaries, AI tags, **AI-enhanced notes**,
 > opt-in recording with **in-app playback**, at-rest encryption, self-hosted sync, **Outlook
-> `.pst`/calendar import with optional auto-record**, **importing an existing recording from a
+> `.pst`/calendar import with optional auto-record**, **background import of an existing
-> file or URL**, and a loopback **MCP server** for coding-agent handoff are all implemented and
+> recording from a file or URL** (with a step-by-step progress tracker), **close-to-tray**, and a
-> ship as a single signed **MSI + NSIS** universal installer. Build order and remaining tasks
+> loopback **MCP server** for coding-agent handoff are all implemented and ship as a single
-> are in [`docs/05-roadmap.md`](docs/05-roadmap.md).
+> **MSI + NSIS** universal installer. On Intel Core Ultra machines, transcription runs with the
+> **CPU near-idle**: Whisper's encoder on the **NPU**, its decoder on the **iGPU**. Build order
+> and remaining tasks are in [`docs/05-roadmap.md`](docs/05-roadmap.md).
 
 ## Why WhispAssist — Granola vs Meetily vs WhispAssist
 
-| | **Granola** | **Meetily** | **WhispAssist** |
+|                                                | **Granola**                                     | **Meetily**                       | **WhispAssist**                                             |
-|---|---|---|---|
+| ---------------------------------------------- | ----------------------------------------------- | --------------------------------- | ----------------------------------------------------------- |
-| **Positioning** | Cloud AI notepad | Open-source self-hosted assistant | Local, Windows-native assistant |
+| **Positioning**                                | Cloud AI notepad                                | Open-source self-hosted assistant | Local, Windows-native assistant                             |
-| **Platform** | macOS, Windows | macOS, Windows, Linux | Windows 10/11 |
+| **Platform**                                   | macOS, Windows                                  | macOS, Windows, Linux             | Windows 10/11                                               |
-| **Bot-free system-audio capture** | ✅ | ✅ | ✅ WASAPI loopback |
+| **Bot-free system-audio capture**              | ✅                                              | ✅                                | ✅ WASAPI loopback                                          |
-| **Transcription** | ☁️ cloud | ✅ local (Whisper) | ✅ local (whisper.cpp) |
+| **Transcription**                              | ☁️ cloud                                        | ✅ local (Whisper)                | ✅ local (whisper.cpp)                                      |
-| **NPU / GPU auto-acceleration** | n/a (cloud) | ⚠️ CPU/GPU, manual | ✅ **NPU→NVIDIA→AMD→Intel→CPU** ladder, one binary |
+| **NPU / GPU auto-acceleration**                | n/a (cloud)                                     | ⚠️ CPU/GPU, manual                | ✅ **NPU→NVIDIA→AMD→Intel→CPU** ladder, one binary          |
-| **Speaker diarization** | ✅ cloud | ⚠️ limited | ✅ offline (sherpa-onnx) |
+| **Speaker diarization**                        | ✅ cloud                                        | ⚠️ limited                        | ✅ offline (sherpa-onnx)                                    |
-| **Summaries / notes AI** | ☁️ cloud LLM | ✅ local (Ollama) / BYO | ✅ local (Ollama, localhost **or LAN**) + optional hosted |
+| **Summaries / notes AI**                       | ☁️ cloud LLM                                    | ✅ local (Ollama) / BYO           | ✅ local (Ollama, localhost **or LAN**) + optional hosted   |
-| **Interface localization (UI language)** | ⚠️ English-only UI (multilingual transcription) | ⚠️ English-only UI (more planned) | ✅ **20 languages** + RTL (Arabic/Urdu) |
+| **Interface localization (UI language)**       | ⚠️ English-only UI (multilingual transcription) | ⚠️ English-only UI (more planned) | ✅ **20 languages** + RTL (Arabic/Urdu)                     |
-| **Default data egress** | ☁️ audio + notes to cloud | 🔒 local (cloud optional) | 🔒 **none** — everything off by default, allowlist-enforced |
+| **Default data egress**                        | ☁️ audio + notes to cloud                       | 🔒 local (cloud optional)         | 🔒 **none** — everything off by default, allowlist-enforced |
-| **Calendar / Outlook `.pst` context** | ✅ cloud calendar | ❌ | ✅ local `.pst` import + **auto-record on events** |
+| **Calendar / Outlook `.pst` context**          | ✅ cloud calendar                               | ❌                                | ✅ local `.pst` import + **auto-record on events**          |
-| **Import an existing recording (file/URL)** | ❌ live capture only | ❌ | ✅ any file or URL (ffmpeg + yt-dlp) |
+| **Import an existing recording (file/URL)**    | ❌ live capture only                            | ❌                                | ✅ any file or URL (ffmpeg + yt-dlp)                        |
-| **Opt-in recording + consent notice** | ⚠️ partial | ❌ | ✅ off by default, one-time consent |
+| **Opt-in recording + consent notice**          | ⚠️ partial                                      | ❌                                | ✅ off by default, one-time consent                         |
-| **At-rest encryption** | ☁️ server-side | ❌ | ✅ vault: Argon2id + XChaCha20-Poly1305 |
+| **At-rest encryption**                         | ☁️ server-side                                  | ❌                                | ✅ vault: Argon2id + XChaCha20-Poly1305                     |
-| **Self-hosted sync w/ client-side encryption** | ❌ | ⚠️ | ✅ WebDAV + OAuth, **encrypt-before-upload** |
+| **Self-hosted sync w/ client-side encryption** | ❌                                              | ⚠️                                | ✅ WebDAV + OAuth, **encrypt-before-upload**                |
-| **Coding-agent (MCP) server** | ✅ **cloud** MCP (notes via their servers) | ❌ | ✅ **local, loopback-only**, token-gated, adds no egress |
+| **Coding-agent (MCP) server**                  | ✅ **cloud** MCP (notes via their servers)      | ❌                                | ✅ **local, loopback-only**, token-gated, adds no egress    |
-| **License** | Proprietary | Open source (MIT) | Open source (MIT / Apache-2.0) |
+| **License**                                    | Proprietary                                     | Open source (MIT)                 | Open source (MIT / Apache-2.0)                              |
-| **Cost** | Subscription | Free | Free |
+| **Cost**                                       | Subscription                                    | Free                              | Free                                                        |
 
 <sub>Comparison reflects each project's public positioning as of mid-2026. Granola now ships an
 MCP server too, but it is **cloud-hosted** — an agent reaching it pulls your notes through
@@ -56,29 +59,36 @@ accelerated, zero-egress-by-default** option: it exploits the NPU/GPU in modern
 everything on the device unless you opt in, and adds Windows-specific context (Outlook) and a
 coding-agent handoff.
 
-## What's built (v0.5.2)
+## What's built (v0.7.3)
 
-- **Bot-free capture — now both sides, with a live dual meter** — WASAPI loopback records the
+- **Bot-free capture — now both sides, with a live dual meter and mic mute** — WASAPI loopback
-  system mix (all participants), and an optional **microphone** path captures your own voice,
+  records the system mix (all participants), and an optional **microphone** path captures your
-  mixed into both the live transcript and the saved recording. While recording, a **level meter
+  own voice, mixed into both the live transcript and the saved recording. While recording, a
-  overlays the system and microphone signals in two colours** so you can see both sides are being
+  **level meter overlays the system and microphone signals in two colours** so you can see both
-  picked up. Pick a specific output/input or turn the mic off in **Settings ▸ Hardware**. No
+  sides are being picked up, and the **M key (or the mic button) mutes/unmutes your mic
+  mid-meeting**. Pick a specific output/input or turn the mic off in **Settings ▸ Hardware**. No
   meeting bot, no per-app plumbing.
 - **Local transcription with a hardware ladder — and a fluid live transcript** — whisper.cpp via
   `whisper-rs` on CPU; the **Intel NPU** via ONNX Runtime + OpenVINO; **GPU via Vulkan** (a single
   binary that runs on NVIDIA, AMD, and Intel). WA detects the hardware, picks the best backend
-  (**NPU → NVIDIA → AMD → Intel → CPU**), and shows it in the UI. The **live transcript streams a
+  (**NPU → NVIDIA → AMD → Intel → CPU**), and shows it in the UI. On NPU machines the engine
-  growing line that refreshes ~once a second and commits at natural pauses** — words appear as
+  **splits Whisper across accelerators — encoder on the NPU, KV-cache decoder on the Intel iGPU —
-  they're spoken instead of in fixed multi-second blocks, so sentences aren't chopped across
+  so transcription leaves the CPU near-idle**. The **live transcript streams a growing line that
-  lines.
+  refreshes ~once a second and commits at natural pauses** — words appear as they're spoken
-- **Speaker diarization** — `sherpa-onnx` (pyannote segmentation + speaker-embedding clustering),
+  instead of in fixed multi-second blocks, so sentences aren't chopped across lines.
-  fully offline. Install the two diarization models in **Settings ▸ Hardware** and finished
+- **Speaker diarization that knows who _you_ are** — `sherpa-onnx` (pyannote segmentation +
-  recordings are split by speaker (Speaker 1, Speaker 2, …); the microphone speaker is
+  speaker-embedding clustering), fully offline. Install the two diarization models in
-  auto-labelled from a short voiceprint.
+  **Settings ▸ Hardware** and recordings are split by speaker (You, Speaker 2, …). With the mic
-- **Import an existing recording** — add a meeting from a **local audio/video file or a URL**
+  enabled, your voice is recorded on its **own channel**, so your lines are labelled **You** live
-  (YouTube, a streaming page, or a direct media link). WhispAssist transcribes and diarizes it
+  — from the first pass, not just after the meeting — and attributed by per-channel voice
-  just like a live recording. Uses **`ffmpeg`** (and **`yt-dlp`** for URLs), which you install
+  activity rather than clustering guesswork. Renaming a speaker updates the transcript **and the
-  yourself — neither is bundled.
+  generated notes**.
+- **Import an existing recording — in the background, with a progress tracker** — add a meeting
+  from a **local audio/video file or a URL** (YouTube, a streaming page, or a direct media link);
+  it transcribes quietly in the background behind a four-step tracker (_Transcode → Transcribe →
+  Identify speakers → Finalize_) while you keep working. Uses **`ffmpeg`** (and **`yt-dlp`** for
+  URLs), which you install yourself — neither is bundled, and one-click download buttons are in
+  the Import dialog.
 - **Notes & summaries** — Markdown notes; local-LLM summaries via **Ollama** on `localhost`
   **or a private LAN endpoint** (RFC-1918), with a full advanced-parameter panel (system prompt,
   `think`, `keep_alive`, `num_ctx`, sampling/repetition/mirostat, etc.).
@@ -90,8 +100,10 @@ coding-agent handoff.
   recordings are decrypted **in memory on the fly** (nothing plaintext is written to disk).
   Recordings are 16-bit for roughly half the size, and an accidental recording can be **cancelled**
   (audio + transcript deleted).
-- **Notes, summaries & AI tags** — Markdown notes with an **Editor/Preview** toggle; local-LLM
+- **Notes, summaries & AI tags** — Markdown notes with an **Editor/Preview** toggle, a formatting
-  summaries and one-click **tag generation** with a chip-based tag editor and tag filtering.
+  toolbar, and **slash commands** (`/todo`, `/h1`, `/quote`, …); **AI-enhance** turns rough notes
+  into clean structured Markdown via your local LLM (with one-step Undo); local-LLM summaries and
+  one-click **tag generation** with a chip-based tag editor and tag filtering.
 - **At-rest encryption vault** — Argon2id key derivation + XChaCha20-Poly1305; transcripts,
   notes, summaries, and recordings sealed on disk; startup unlock gate; keys zeroized on lock.
 - **Self-hosted sync (optional, off by default)** — WebDAV (Nextcloud, ownCloud, Cloudreve,
@@ -103,7 +115,11 @@ coding-agent handoff.
 - **Outlook `.pst` / calendar context, with auto-record** — import events and attendees from a
   local Outlook `.pst` backup (read-only, range-limited, de-duplicated, with cleanup), attach
   meetings to events, and — opt-in — **auto-start recording when a calendar event begins** while
-  the app is open (a one-shot timer, no background polling).
+  the app is open (a one-shot timer, no background polling). The import shells out to `readpst`
+  (libpst) — **Settings ▸ Calendar** has a one-click download button for it.
+- **Close to tray** — closing the window keeps WhispAssist running in the background; reopen it
+  from the tray icon, quit fully from the tray menu. On by default, toggleable in
+  **Settings ▸ Recording**.
 - **Local MCP server for coding-agent handoff** — hand meeting context to your own coding agents
   (Claude, Codex, Copilot, OpenCode) over a **loopback-only, token-gated** MCP server that is off
   by default, scope-limited, audited, and **adds no egress** — data leaves only via the agent's
@@ -116,9 +132,10 @@ coding-agent handoff.
   Sinhala (සිංහල), Spanish — Spain & Mexico (Español), Swedish (Svenska), Tamil (தமிழ்), and
   Urdu (اردو). Selecting Arabic or Urdu flips the interface to **right-to-left**; any untranslated
   string falls back to English rather than showing a raw key. Adding a language is a one-file job.
-- **One universal installer** — a single signed MSI and NSIS `-setup.exe` that covers every
+- **One universal installer** — a single MSI and NSIS `-setup.exe` that covers every machine:
-  machine: Vulkan for all GPUs, the Intel NPU path (+ DirectML fallback), and CPU. The Vulkan
+  Vulkan for all GPUs, the Intel NPU path (+ DirectML fallback), and CPU. The Vulkan loader is
-  loader is bundled so it launches even on machines without a GPU driver.
+  bundled so it launches even on machines without a GPU driver. Every release ships **SHA-256
+  checksums** in its release notes.
 
 **In progress:** Microsoft Graph calendar (cloud calendar via OAuth) and an optional CUDA
 (NVIDIA-only) build variant.
@@ -132,30 +149,39 @@ during setup).
 1. Download the latest **`WhispAssist_<version>_x64_en-US.msi`** from the
    [releases page](https://git.dou.bet/iamdoubz/WhispAssist/releases).
 2. Double-click the `.msi` and follow the prompts, accepting the UAC prompt to install. The
-   installer is signed; if Windows SmartScreen appears, choose **More info → Run anyway**.
+   installers are currently unsigned, so if Windows SmartScreen appears, choose
+   **More info → Run anyway** (verify the download against the SHA-256 checksums in the release
+   notes first).
 3. Launch **WhispAssist** from the Start menu.
 
 On first run WA detects your hardware and picks the best transcription backend
 (**NPU → NVIDIA → AMD → Intel → CPU**). It works on your **CPU or GPU** (GPU via Vulkan) out of the
 box; to use an Intel **NPU**, open **Settings ▸ Hardware** and download the one-time NPU
-acceleration package. The app runs without admin rights, never adds itself to startup, and keeps all
+acceleration package. The app runs without admin rights, does not add itself to startup unless you
-data under `%LOCALAPPDATA%\WhispAssist`.
+opt in (**Settings ▸ Recording ▸ Launch at login**), and keeps all data under
+`%LOCALAPPDATA%\WhispAssist`.
 
-Prefer the NSIS installer? Grab **`WhispAssist_<version>_x64-setup.exe`** from the same page.
+Prefer the NSIS installer? Grab **`WhispAssist_<version>_x64-setup.exe`** from the same page — it
+lets you choose a **current-user** (no admin) or **all-users** install.
+
+**Deploying to many machines?** See [`docs/enterprise-deployment.md`](docs/enterprise-deployment.md)
+for silent install, custom install location, and presetting defaults with a `wa-defaults.ini` file.
 
 ## Optional dependencies
 
-If you do not have these installed, WhispAssist will still work, but some features will be unavailable.
+If you do not have these installed, WhispAssist will still work, but some features will be
+unavailable. The app links to each one where it's needed: `ffmpeg`/`yt-dlp` in the Import dialog,
+`libpst` in **Settings ▸ Calendar**.
 
-- [yt-dlp](https://github.com/yt-dlp/yt-dlp/releases)
+- [yt-dlp](https://github.com/yt-dlp/yt-dlp/releases) — importing from URLs
-- [ffmpeg](https://www.ffmpeg.org/download.html#build-windows)
+- [ffmpeg](https://www.ffmpeg.org/download.html#build-windows) — importing files/URLs
-- [libpst](https://sourceforge.net/projects/ezwinports/files/libpst-0.6.63-w32-bin.zip/)
+- [libpst](https://sourceforge.net/projects/ezwinports/files/libpst-0.6.63-w32-bin.zip/) — Outlook `.pst` calendar import
 
 ## Technology
 
 WhispAssist is a **Tauri 2** application: a small Rust core with a compiled **Svelte + TypeScript**
 frontend rendered through the OS WebView2 (no bundled browser → low idle memory). Every decision
-and its alternatives are recorded in [`docs/adr/`](docs/adr/) (ADR-0001–0011).
+and its alternatives are recorded in [`docs/adr/`](docs/adr/) (ADR-0001–0012).
 
 - **Shell / IPC:** Tauri 2 (Rust ⇄ WebView2)
 - **Audio capture:** WASAPI loopback
@@ -176,7 +202,7 @@ and its alternatives are recorded in [`docs/adr/`](docs/adr/) (ADR-0001–0011).
 WhispAssist/
 ├── docs/                 # The engineering plan (read this first)
 │   ├── 00-overview.md … 07-research-findings.md
-│   └── adr/                  Architecture Decision Records (0001–0011)
+│   └── adr/                  Architecture Decision Records (0001–0012)
 ├── src-tauri/            # Rust core — implemented service modules:
 │   └── src/{audio,transcription,diarization,storage,llm,calendar,
 │            hardware,notes,sync,mcp,vault}

RELEASE_NOTES_0.5.0.md

@@ -1,96 +0,0 @@
-# WhispAssist v0.5.0
-
-**Privacy-first, Windows-native meeting assistant — everything on-device, nothing leaves unless you say so.**
-
-This release is about **living with your recordings**: play back a meeting while following along in
-the transcript, manage action items by hand, move recordings between computers, drop a meeting into
-your Obsidian vault, and — new in this release — **use WhispAssist in your own language**. Everything
-stays off-by-default and local-first.
-
-One universal installer (**MSI** and **NSIS**) covers every machine: **Vulkan** for all GPUs
-(NVIDIA/AMD/Intel), the Intel **NPU** (OpenVINO), a **DirectML** fallback, and **CPU**.
-
----
-
-## ✨ Highlights
-
-### Click the transcript to play that moment
-The recording player and the transcript now talk to each other. **Click any transcript line to jump
-the audio to that moment** and start playing, and as playback runs the **current line highlights and
-scrolls into view** so you never lose your place. If you scroll by hand, auto-scroll steps aside for
-a few seconds so it doesn't fight you.
-
-### Interface language selector (i18n)
-WhispAssist can now be **fully translated**. A new **Settings ▸ Language** picker switches the
-interface language, English ships as the baseline, and **every** user-facing string across the app —
-the shell, meetings list, transcript/notes, summary panel, and all of Settings — now flows through a
-single translation layer. Adding a new language is as simple as translating **one JSON file**; no
-code changes. (This release ships English; the groundwork is in place for community translations.)
-
-### Manage action items yourself
-Action items are no longer just whatever the summary extracted. You can now **add, edit, and delete
-them by hand** in the summary panel, set an owner and due date, and toggle a local reminder. Your
-edits are the source of truth and are reconciled cleanly — deleting an item also cancels its reminder.
-
-### Move recordings between computers (Export / Import)
-A new **Export & import** section in **Settings ▸ Storage** writes each meeting as a portable
-**bundle folder** (audio, transcript, notes, summary, and a `meeting.json` manifest) and imports them
-back on another machine. Imported meetings get a fresh id, so re-importing never overwrites anything.
-Export to any folder — a synced drive, a USB stick, or a sync target's local mount — and carry it across.
-
-### Export a meeting to Obsidian
-A new **Obsidian** export writes a single self-contained vault note — YAML frontmatter
-(title, date, duration, participants, tags) plus notes, summary, action items, and a timestamped
-transcript — **without the audio**. Drop it in your vault and the transcript is fully searchable.
-
----
-
-## 🚀 Also new since v0.4.0
-
-- **Per-segment transcript timestamps** — every line now shows a quiet `m:ss` (or `h:mm:ss`) time
-  prefix, in both the live and finalized views.
-- **Auto-resync on edit** — when sync is enabled, editing a meeting's notes, summary, transcript,
-  tags, or action items **re-uploads just the changed artifacts** automatically (deduped by hash, so
-  an unchanged save uploads nothing). Off unless sync is configured.
-
-## 🐛 Fixes & polish
-- Transcript scroll-intent handling refined so playback auto-scroll never yanks you back while you're
-  reading.
-- Import preserves each meeting's original date rather than stamping the import time.
-
----
-
-## 📦 Install
-
-**Requirements:** Windows 10 or 11 (x64). WhispAssist needs **WebView2** (preinstalled on Windows 11;
-the installer fetches it on Windows 10). Importing from a file/URL additionally needs **`ffmpeg`**
-(and **`yt-dlp`** for URLs) on your PATH — install those separately.
-
-1. Download **`WhispAssist_0.5.0_x64_en-US.msi`** (or the NSIS **`WhispAssist_0.5.0_x64-setup.exe`**).
-2. Run it and accept the UAC prompt. If SmartScreen appears, choose **More info → Run anyway**.
-3. Launch **WhispAssist** from the Start menu.
-
-On first run WA picks the best transcription backend (**NPU → NVIDIA → AMD → Intel → CPU**). It runs
-without admin rights, never adds itself to startup, and keeps all data under
-`%LOCALAPPDATA%\WhispAssist`.
-
-## 🔐 Checksums (SHA-256)
-
-```
-b423feed1a1171384e46c5e0b5aa63cbf912bfe8f29cc89a5a027bb15358075c  WhispAssist_0.5.0_x64_en-US.msi
-2dce26b5603a5d02094533c594d814a3d72e05512e1d735ea69e1e309f15b710  WhispAssist_0.5.0_x64-setup.exe
-```
-
-Verify after download:
-
-```powershell
-Get-FileHash .\WhispAssist_0.5.0_x64_en-US.msi -Algorithm SHA256
-```
-
----
-
-## Privacy, unchanged
-Everything optional is **off by default**. With nothing configured, WhispAssist makes **no content
-egress at all**. Recording is opt-in; sync/AI credentials live only in the OS credential store; the
-MCP server is loopback-only and adds no egress. The reachable-host allowlist is derived from your
-settings and enforced in the core.

RELEASE_NOTES_0.5.1.md

@@ -1,73 +0,0 @@
-# WhispAssist v0.5.1
-
-**Privacy-first, Windows-native meeting assistant — everything on-device, nothing leaves unless you say so.**
-
-v0.5.0 put the **i18n groundwork** in place. v0.5.1 fills it in: WhispAssist now ships in
-**20 languages**, so you can run the whole app — the shell, meetings list, transcript/notes, summary
-panel, and every corner of Settings — in your own language. Everything stays off-by-default and
-local-first; this is a UI-language release with no change to what leaves your device (nothing, by default).
-
-One universal installer (**MSI** and **NSIS**) covers every machine: **Vulkan** for all GPUs
-(NVIDIA/AMD/Intel), the Intel **NPU** (OpenVINO), a **DirectML** fallback, and **CPU**.
-
----
-
-## ✨ Highlights
-
-### 20 interface languages
-Pick your language under **Settings ▸ Language**. Alongside English, this release adds full
-translations for:
-
-- **Arabic** (العربية) · **Bengali** (বাংলা) · **German** (Deutsch)
-- **Spanish** — Spain (Español, España) and **Mexico** (Español, México)
-- **Finnish** (Suomi) · **French** — France (Français, France) and **Canada** (Français, Canada)
-- **Hindi** (हिन्दी) · **Korean** (한국어) · **Norwegian Bokmål** (Norsk bokmål)
-- **Portuguese** — Brazil (Português, Brasil) and **Portugal** (Português, Portugal)
-- **Russian** (Русский) · **Sinhala** (සිංහල) · **Swedish** (Svenska)
-- **Tamil** (தமிழ்) · **Urdu** (اردو) · **Mandarin Chinese, Simplified** (中文简体)
-
-Your choice persists across launches, and any untranslated string quietly falls back to English rather
-than showing a raw key — so partial translations degrade gracefully.
-
-### Right-to-left layout
-Selecting **Arabic** or **Urdu** flips the whole interface to **right-to-left**, so those languages
-read and lay out correctly rather than being crammed into an LTR shell.
-
-Adding a further language remains a one-file job — drop in a single JSON translation, no code changes.
-
----
-
-## 📦 Install
-
-**Requirements:** Windows 10 or 11 (x64). WhispAssist needs **WebView2** (preinstalled on Windows 11;
-the installer fetches it on Windows 10). Importing from a file/URL additionally needs **`ffmpeg`**
-(and **`yt-dlp`** for URLs) on your PATH — install those separately.
-
-1. Download **`WhispAssist_0.5.1_x64_en-US.msi`** (or the NSIS **`WhispAssist_0.5.1_x64-setup.exe`**).
-2. Run it and accept the UAC prompt. If SmartScreen appears, choose **More info → Run anyway**.
-3. Launch **WhispAssist** from the Start menu.
-
-On first run WA picks the best transcription backend (**NPU → NVIDIA → AMD → Intel → CPU**). It runs
-without admin rights, never adds itself to startup, and keeps all data under
-`%LOCALAPPDATA%\WhispAssist`.
-
-## 🔐 Checksums (SHA-256)
-
-```
-911618e6996a079bad1cbe46731c89ee247f3f5f17c09b60091d7e6a5e7cbef0  WhispAssist_0.5.1_x64_en-US.msi
-ffabbd74def981da176931349b2813a3927d7376c0d18e9d191891abfc17c385  WhispAssist_0.5.1_x64-setup.exe
-```
-
-Verify after download:
-
-```powershell
-Get-FileHash .\WhispAssist_0.5.1_x64_en-US.msi -Algorithm SHA256
-```
-
----
-
-## Privacy, unchanged
-Everything optional is **off by default**. With nothing configured, WhispAssist makes **no content
-egress at all**. Recording is opt-in; sync/AI credentials live only in the OS credential store; the
-MCP server is loopback-only and adds no egress. The reachable-host allowlist is derived from your
-settings and enforced in the core.

RELEASE_NOTES_0.5.2.md

@@ -1,64 +0,0 @@
-# WhispAssist v0.5.2
-
-**Privacy-first, Windows-native meeting assistant — everything on-device, nothing leaves unless you say so.**
-
-A small but important bug-fix release. v0.5.2 fixes a startup crash that could stop WhispAssist
-from opening after an upgrade, and makes any future database problem show a clear message instead
-of failing silently. No feature or privacy changes — everything optional stays off-by-default and
-local-first.
-
-One universal installer (**MSI** and **NSIS**) covers every machine: **Vulkan** for all GPUs
-(NVIDIA/AMD/Intel), the Intel **NPU** (OpenVINO), a **DirectML** fallback, and **CPU**.
-
----
-
-## 🐛 Fixes
-
-### Fixed: app failing to launch after an upgrade
-On some machines, upgrading could leave the app unable to open at all — no window, no message,
-just nothing. The cause was a database-migration mismatch during startup: WhispAssist would hit
-the error while opening its local database (`wa.db`) and, because the failure happened before the
-window existed, the process exited silently and Windows reported only a generic crash.
-
-### Startup database errors now explain themselves
-Instead of that silent exit, a failure to open `wa.db` now shows a **native error dialog** naming
-the problem and pointing at the database location, then exits cleanly. Your recordings and notes
-are never touched — the message tells you exactly what happened and how to recover, rather than
-leaving you with an app that won't start.
-
----
-
-## 📦 Install
-
-**Requirements:** Windows 10 or 11 (x64). WhispAssist needs **WebView2** (preinstalled on Windows 11;
-the installer fetches it on Windows 10). Importing from a file/URL additionally needs **`ffmpeg`**
-(and **`yt-dlp`** for URLs) on your PATH — install those separately.
-
-1. Download **`WhispAssist_0.5.2_x64_en-US.msi`** (or the NSIS **`WhispAssist_0.5.2_x64-setup.exe`**).
-2. Run it and accept the UAC prompt. If SmartScreen appears, choose **More info → Run anyway**.
-3. Launch **WhispAssist** from the Start menu.
-
-On first run WA picks the best transcription backend (**NPU → NVIDIA → AMD → Intel → CPU**). It runs
-without admin rights, never adds itself to startup, and keeps all data under
-`%LOCALAPPDATA%\WhispAssist`.
-
-## 🔐 Checksums (SHA-256)
-
-```
-b173b6b15590a65e96d3b24d0ad97666b8d1da905169d807ecc1c51df0584d3e  WhispAssist_0.5.2_x64_en-US.msi
-441b2e7ec7c68be39f9d2381af59ab9eadf916d6d23d0c5541d78bdd88c588b2  WhispAssist_0.5.2_x64-setup.exe
-```
-
-Verify after download:
-
-```powershell
-Get-FileHash .\WhispAssist_0.5.2_x64_en-US.msi -Algorithm SHA256
-```
-
----
-
-## Privacy, unchanged
-Everything optional is **off by default**. With nothing configured, WhispAssist makes **no content
-egress at all**. Recording is opt-in; sync/AI credentials live only in the OS credential store; the
-MCP server is loopback-only and adds no egress. The reachable-host allowlist is derived from your
-settings and enforced in the core.

RELEASE_NOTES_0.7.0.md

@@ -0,0 +1,78 @@
+# WhispAssist v0.7.0
+
+**Privacy-first, Windows-native meeting assistant — everything on-device, nothing leaves unless you say so.**
+
+This release is about **control over install and startup** — for a single user and for admins rolling
+WhispAssist out across a fleet. You can now launch WhispAssist at login, choose a per-user (no-admin)
+or all-users install, preset every default with a deployment file, and re-scan your hardware without
+restarting. No feature here changes the privacy posture: everything optional stays off-by-default and
+local-first.
+
+One universal installer (**MSI** and **NSIS**) covers every machine: **Vulkan** for all GPUs
+(NVIDIA/AMD/Intel), the Intel **NPU** (OpenVINO), a **DirectML** fallback, and **CPU**.
+
+---
+
+## ✨ New
+
+### Launch at login (opt-in)
+A new **Settings ▸ Recording ▸ Launch WhispAssist at login** toggle starts WhispAssist when you sign
+in to Windows. It's **off by default**, needs **no admin** (a per-user startup entry), and does **not**
+begin recording on its own.
+
+### Choose a per-user or all-users install
+The NSIS setup (`WhispAssist_0.7.0_x64-setup.exe`) now lets you install for **just yourself**
+(no admin rights required) or for **all users**. The MSI remains the per-machine, all-users installer.
+
+### Customize deployments with a file (`wa-defaults.ini`)
+Admins can preset WhispAssist's defaults for every machine using native Windows tooling (Group Policy,
+SCCM, Intune, `msiexec`) — no management console. Drop a `wa-defaults.ini` next to the installer or in
+`%PROGRAMDATA%\WhispAssist\`, and each machine's **first launch** seeds its settings from it: record-
+by-default, preferred backend, a model to auto-download, retention, storage location, AI provider, and
+more. You can also set a **custom install location** with standard `msiexec INSTALLDIR=…` / NSIS `/D=`
+flags. **Secrets are never read from this file** — API keys and tokens stay in the OS credential store.
+Full key reference and silent-install examples in
+[`docs/enterprise-deployment.md`](docs/enterprise-deployment.md).
+
+### Refresh hardware
+A **Refresh** button in **Settings ▸ Hardware** re-detects your GPU/NPU on the spot — handy after
+installing a driver or plugging in an eGPU — no restart needed.
+
+---
+
+## 📦 Install
+
+**Requirements:** Windows 10 or 11 (x64). WhispAssist needs **WebView2** (preinstalled on Windows 11;
+the installer fetches it on Windows 10). Importing from a file/URL additionally needs **`ffmpeg`**
+(and **`yt-dlp`** for URLs) on your PATH — install those separately.
+
+1. Download **`WhispAssist_0.7.0_x64_en-US.msi`** (or the NSIS **`WhispAssist_0.7.0_x64-setup.exe`**).
+2. Run it and accept the UAC prompt. If SmartScreen appears, choose **More info → Run anyway**.
+3. Launch **WhispAssist** from the Start menu.
+
+On first run WA picks the best transcription backend (**NPU → NVIDIA → AMD → Intel → CPU**). It runs
+without admin rights, does **not** add itself to startup unless you opt in, and keeps all data under
+`%LOCALAPPDATA%\WhispAssist`.
+
+Deploying to many machines? See [`docs/enterprise-deployment.md`](docs/enterprise-deployment.md).
+
+## 🔐 Checksums (SHA-256)
+
+```
+730676580dbef5c4bb2c155f46a73a5791722139ab144a3fc0286e3a4449c432  WhispAssist_0.7.0_x64_en-US.msi
+6a34c16eb6f876fc92c7a79d37414b4fdd2c09f4f40873008149db36cd0a30aa  WhispAssist_0.7.0_x64-setup.exe
+```
+
+Verify after download:
+
+```powershell
+Get-FileHash .\WhispAssist_0.7.0_x64_en-US.msi -Algorithm SHA256
+```
+
+---
+
+## Privacy, unchanged
+Everything optional is **off by default**. With nothing configured, WhispAssist makes **no content
+egress at all**. Recording is opt-in; sync/AI credentials live only in the OS credential store; the
+MCP server is loopback-only and adds no egress; the deployment file never carries secrets. The
+reachable-host allowlist is derived from your settings and enforced in the core.

RELEASE_NOTES_0.7.1.md

@@ -0,0 +1,94 @@
+# WhispAssist v0.7.1
+
+**Privacy-first, Windows-native meeting assistant — everything on-device, nothing leaves unless you say so.**
+
+A quality-of-life release focused on **importing meetings, taking notes, and living in the
+background**. Adding a meeting from a file or link now runs without freezing the app and shows you
+exactly where it's up to; notes gained AI cleanup and quick formatting; and WhispAssist can now mute
+your mic mid-meeting and tuck itself into the system tray. No feature here changes the privacy
+posture: everything optional stays off-by-default and local-first.
+
+One universal installer (**MSI** and **NSIS**) covers every machine: **Vulkan** for all GPUs
+(NVIDIA/AMD/Intel), the Intel **NPU** (OpenVINO), a **DirectML** fallback, and **CPU**.
+
+---
+
+## ✨ New
+
+### Add a meeting — now in the background, with a progress tracker
+Importing a recording (a local audio/video file or a YouTube/streaming/direct URL) no longer blocks
+the app while it works. Click **Import** and the meeting appears in your list immediately with a
+**four-step progress tracker** — *Transcode → Transcribe → Identify speakers → Finalize* — where the
+current step pulses and finished steps show how long they took. A 25-minute video that used to lock
+the window for ~13 minutes now transcribes quietly in the background.
+
+- **Pick the transcription model** right in the dialog, and see **"Transcribed with …"** on the
+  finished meeting so you always know how it was produced.
+- **One-click links** to download `ffmpeg` and `yt-dlp` (still external, not bundled).
+- A failed import stays in your list marked **error** instead of vanishing.
+
+### Better notes
+- **AI-enhance** (✨): turn rough notes into clean, structured notes using your local LLM, with a
+  one-step **Undo**. Off unless you have a local model configured.
+- **Slash commands & a formatting toolbar**: type `/todo`, `/h1`, `/quote`, … or use the toolbar for
+  headings, lists, checkboxes, quotes, and dividers.
+- **Fix:** notes no longer show stale text after re-transcribing a meeting — the pane updates in
+  place, no restart needed.
+
+### Mute your microphone — press **M**
+Mute/unmute the mic mid-meeting with the **M** key or the new mic button by the level meter. The mic
+channel goes silent (recording, live transcript, and meter) while system/loopback audio keeps
+capturing.
+
+### Close to system tray
+Closing the window now **keeps WhispAssist running in the background** instead of quitting. Reopen it
+from the tray icon; the tray's **Quit** exits fully. On by default — toggle it in
+**Settings ▸ Recording ▸ Close to system tray**. (This release also fixes a bug that showed **two**
+WhispAssist icons in the tray — there's now just one.)
+
+### Privacy & hardware odds and ends
+- **Vault lock card** in **Settings ▸ Privacy**: lock/unlock the encrypted store and change its
+  password at a glance.
+- **Test your audio devices**: a live level meter for your mic and system audio, plus a test tone.
+- **Quick hardware stress test**: benchmark the available backends against your installed models and
+  apply the fastest real-time combination.
+- Tidier recording header.
+
+---
+
+## 📦 Install
+
+**Requirements:** Windows 10 or 11 (x64). WhispAssist needs **WebView2** (preinstalled on Windows 11;
+the installer fetches it on Windows 10). Importing from a file/URL additionally needs **`ffmpeg`**
+(and **`yt-dlp`** for URLs) on your PATH — the Import dialog now links to both.
+
+1. Download **`WhispAssist_0.7.1_x64_en-US.msi`** (or the NSIS **`WhispAssist_0.7.1_x64-setup.exe`**).
+2. Run it and accept the UAC prompt. If SmartScreen appears, choose **More info → Run anyway**.
+3. Launch **WhispAssist** from the Start menu.
+
+On first run WA picks the best transcription backend (**NPU → NVIDIA → AMD → Intel → CPU**). It runs
+without admin rights, does **not** add itself to startup unless you opt in, and keeps all data under
+`%LOCALAPPDATA%\WhispAssist`.
+
+Deploying to many machines? See [`docs/enterprise-deployment.md`](docs/enterprise-deployment.md).
+
+## 🔐 Checksums (SHA-256)
+
+```
+bd96a059db3658a9bee81161edc709cfc4741ed99d6c66a12194403b339f9369  WhispAssist_0.7.1_x64_en-US.msi
+711fa7df618c8fc3f03c18d543abad2288d56f750b532288204e0ec0f9426595  WhispAssist_0.7.1_x64-setup.exe
+```
+
+Verify after download:
+
+```powershell
+Get-FileHash .\WhispAssist_0.7.1_x64_en-US.msi -Algorithm SHA256
+```
+
+---
+
+## Privacy, unchanged
+Everything optional is **off by default**. With nothing configured, WhispAssist makes **no content
+egress at all**. Recording is opt-in; sync/AI credentials live only in the OS credential store; the
+MCP server is loopback-only and adds no egress; the deployment file never carries secrets. The
+reachable-host allowlist is derived from your settings and enforced in the core.

RELEASE_NOTES_0.7.3.md

@@ -0,0 +1,94 @@
+# WhispAssist v0.7.3
+
+**Privacy-first, Windows-native meeting assistant — everything on-device, nothing leaves unless you say so.**
+
+A performance release: on Intel Core Ultra machines, transcription now leaves the **CPU nearly
+idle**. The Whisper encoder keeps running on the **NPU**; the decoder — previously the dominant
+CPU cost — now decodes with a **KV cache** and runs on the **Intel iGPU**. Nothing here changes
+the privacy posture: the one new download is the upgraded transcription model, fetched from the
+same source as before.
+
+One universal installer (**MSI** and **NSIS**) covers every machine: **Vulkan** for all GPUs
+(NVIDIA/AMD/Intel), the Intel **NPU** (OpenVINO), a **DirectML** fallback, and **CPU**.
+
+---
+
+## ⚡ Faster, cooler transcription on NPU systems
+
+### KV-cache decoding
+The NPU engine previously re-processed the entire token sequence for every new token it
+generated. It now uses the upgraded *merged* Whisper decoder and caches attention state between
+steps, so each step does a small, constant amount of work. This benefits **every** machine using
+the NPU/DirectML engine, whatever device the decoder runs on.
+
+### Decoder on the Intel iGPU
+On NPU systems the decoder now runs on the **Intel iGPU** via the same OpenVINO runtime already
+installed for the NPU — no new runtime download. Division of labor while transcribing:
+
+- **Encoder** → NPU (the short periodic spikes on Task Manager's NPU graph)
+- **Decoder** → Intel iGPU
+- **CPU** → audio plumbing only
+
+Measured on a Core Ultra 5 135U (base.en): decoding on the iGPU ran **~4.7× faster** than the
+same KV-cache decode on the CPU (1.29 s vs 6.10 s for a 19 s clip), with identical transcripts —
+and that work no longer occupies the CPU at all.
+
+**Fallbacks:** no usable Intel iGPU → the decoder automatically drops back to the CPU (still
+with the KV-cache speedup), and the log says so. Machines with an NVIDIA/AMD GPU can keep using
+the Vulkan backend (Settings ▸ Hardware), which runs the whole model on that GPU. The startup
+log names both devices: `transcription engine: ONNX base.en (encoder EP: OpenVINO/NPU, decoder
+EP: OpenVINO/GPU)`.
+
+### Model update (automatic, one-time)
+The engine's decoder artifact changed to the merged export (`decoder_model_merged.onnx`,
+~208 MB, same trusted source). WhispAssist fetches it automatically in the background on first
+launch after updating; until it arrives, transcription simply uses the whisper.cpp engine, as on
+a fresh install. The superseded decoder file is cleaned up automatically.
+
+---
+
+## ✨ Small addition
+
+**Settings ▸ Calendar** now has a one-click **libpst (readpst)** download button — the same style
+as the Import dialog's `ffmpeg`/`yt-dlp` buttons — linking to the ezwinports Windows build.
+`.pst` calendar import shells out to `readpst`, which isn't bundled; the button just opens the
+download page in your browser.
+
+---
+
+## 📦 Install
+
+**Requirements:** Windows 10 or 11 (x64). WhispAssist needs **WebView2** (preinstalled on Windows 11;
+the installer fetches it on Windows 10). Importing from a file/URL additionally needs **`ffmpeg`**
+(and **`yt-dlp`** for URLs) on your PATH — the Import dialog links to both.
+
+1. Download **`WhispAssist_0.7.3_x64_en-US.msi`** (or the NSIS **`WhispAssist_0.7.3_x64-setup.exe`**).
+2. Run it and accept the UAC prompt. If SmartScreen appears, choose **More info → Run anyway**.
+3. Launch **WhispAssist** from the Start menu.
+
+On first run WA picks the best transcription backend (**NPU → NVIDIA → AMD → Intel → CPU**). It runs
+without admin rights, does **not** add itself to startup unless you opt in, and keeps all data under
+`%LOCALAPPDATA%\WhispAssist`.
+
+Deploying to many machines? See [`docs/enterprise-deployment.md`](docs/enterprise-deployment.md).
+
+## 🔐 Checksums (SHA-256)
+
+```
+91a32a17addc31fbc00f38966d52402ca0c5e9e7a581a0431c383ca5a60532bd  WhispAssist_0.7.3_x64_en-US.msi
+764b758544b075b5aefe9dde37b04e4966df528912fb884ed1be5bd57905ecdb  WhispAssist_0.7.3_x64-setup.exe
+```
+
+Verify after download:
+
+```powershell
+Get-FileHash .\WhispAssist_0.7.3_x64_en-US.msi -Algorithm SHA256
+```
+
+---
+
+## Privacy, unchanged
+Everything optional is **off by default**. With nothing configured, WhispAssist makes **no content
+egress at all**. Recording is opt-in; sync/AI credentials live only in the OS credential store; the
+MCP server is loopback-only and adds no egress; the deployment file never carries secrets. The
+reachable-host allowlist is derived from your settings and enforced in the core.

docs/02-architecture.md

@@ -84,10 +84,14 @@ issues from feature briefs.
 4. Provisional speaker turns (from cheap segmentation) tag segments live; the user may name a
    speaker, applied immediately to past/future segments (FR-SPK-2).
 5. The notes view renders Markdown that updates as segments arrive (FR-NOTE-1, FR-NOTE-6).
-6. On `stop_recording`: finalize the WAV; run full `diarization`, align speaker IDs to segments,
+6. On `stop_recording`: finalize the WAV; run `diarization`, align speaker IDs to segments,
-   apply name mappings; persist transcript JSON + metadata via `storage` (FR-STORE-1). If "Record
+   apply name mappings; persist transcript JSON + metadata via `storage` (FR-STORE-1). When the mic
-   this meeting" is **off**, delete the working WAV **after** the transcript is finalized; if **on**,
+   is on, `audio.wav` is a **dual-channel split** (left = mic, right = loopback; ADR-0005 Phase 3.5,
-   keep it as `audio.wav` (ADR-0009, FR-REC-1/4).
+   FR-SPK), so attribution is **per-stream**: sherpa clusters the **right channel only** →
+   `Speaker N`, and "You" comes from **left-channel** voice activity. Blind whole-signal clustering
+   + voiceprint is the mic-off (`summed`) fallback. If "Record this meeting" is **off**, delete the
+   working WAV **after** the transcript is finalized; if **on**, keep it as `audio.wav` (ADR-0009,
+   FR-REC-1/4).
 7. If an LLM provider is configured: `llm` builds a prompt (transcript + metadata + template),
    streams a summary/decisions/action-items into the summary panel (FR-LLM-2/4).
 8. Action items are parsed and presented for confirmation; confirmed ones persist and may raise

docs/03-data-model.md

@@ -14,7 +14,8 @@ Default root: `%LOCALAPPDATA%\WhispAssist\` (user-configurable, FR-STORE-2).
 │   └── spk-eres2net.onnx
 └── meetings\
     └── <meeting_id>\           # one folder per meeting (uuid)
-        ├── audio.wav           # canonical recording — present ONLY if "Record" was on (ADR-0009)
+        ├── audio.wav           # canonical recording — present ONLY if "Record" was on (ADR-0009). Mic on → stereo split (L=mic/"You", R=loopback/"Speaker", `audio_layout=split`); mic off → summed. Playback folds to mono; export folds to dual-mono (ADR-0005 Phase 3.5)
+        ├── voiceprint.wav      # mic-only 16kHz sample for "You" re-matching on reprocess of a SUMMED recording (FR-SPK); lives/dies with audio.wav
         ├── transcript.json     # canonical transcript (segments+speakers+timings)
         ├── manual_notes.json   # raw user-authored notes captured live during recording
         ├── notes.md            # the final notes document: manual notes + transcript, merged at finalize
@@ -86,6 +87,7 @@ CREATE TABLE meetings (
     language        TEXT,                         -- detected/selected language code
     backend_used    TEXT,                         -- npu|nvidia|amd|intel|cpu
     model_used      TEXT,                         -- e.g. whisper-base
+    audio_layout    TEXT,                         -- split|summed|NULL; how audio.wav's channels are laid out (0009, ADR-0005 Phase 3.5)
     calendar_event_id TEXT,                       -- FK -> calendar_events.id (nullable)
     template_id     TEXT,                         -- note-template id (T8.1, FR-NOTE-5); catalog is
                                                     -- a built-in Rust list (notes::templates), not a table
@@ -380,6 +382,13 @@ segment (the M1 grounding invariant, asserted by the golden-transcript test).
     "expose_recordings": false, // never serve .wav unless explicitly true (FR-MCP-3)
   },
   "privacy": { "encrypt_at_rest": false },
+  // Launch WhispAssist at login (NFR-RES-4). Opt-in, OFF by default. Toggling this
+  // via `set_auto_start` also writes a per-user `HKCU\...\Run` entry (no admin);
+  // startup reconciles the OS entry to this flag (e.g. after a reinstall).
+  "auto_start": false,
+  // Closing the window hides WhispAssist to the system tray (keep running in background) instead of
+  // quitting; ON by default. Tray "Quit" is the real exit. Enforced in the Rust on_window_event handler.
+  "close_to_tray": true,
   // Optional MS Graph calendar source (M4.4, T8.9, FR-CAL-6). Opt-in, explicit consent via OAuth
   // PKCE — OFF by default. `credential_ref` points into the OS credential store; the token itself
   // is never written here (same invariant as sync credentials, FR-SYNC-6).
@@ -387,6 +396,20 @@ segment (the M1 grounding invariant, asserted by the golden-transcript test).
 }
 ```
 
+## First-run deploy seeding (`wa-defaults.ini`)
+
+For enterprise mass-deployment, the **first** launch on a machine (before `settings.json` exists)
+optionally seeds its defaults from an admin-supplied INI. First file found wins:
+
+1. `%PROGRAMDATA%\WhispAssist\wa-defaults.ini` — machine-wide (GPO / SCCM / Intune file copy).
+2. `<install dir>\wa-defaults.ini` — the bundled template (shipped fully commented → no-op).
+
+Keys are flat `key = value` matching `settings.json` field names (bools/ints coerced), plus the
+special `auto_download_model = true` which fetches the configured `whisper_model` in the background.
+**No secrets** — any key containing `key`/`token`/`secret`/`credential`/`password` is ignored; those
+live only in the OS credential store. After first run the file is never read again. See
+`docs/enterprise-deployment.md` and `src-tauri/src/deploy.rs`.
+
 ## Retention & recovery semantics
 
 - **Retention** (FR-STORE-2): a background job deletes whole meeting folders + rows once a meeting

docs/04-api-contracts.md

@@ -23,6 +23,9 @@ start_recording(input: { meetingTitle?: string; calendarEventId?: string; record
 stop_recording(input: { meetingId: MeetingId }): MeetingSummaryRef
 pause_recording(input: { meetingId: MeetingId }): void
 resume_recording(input: { meetingId: MeetingId }): void
+// Mute/unmute the mic mid-meeting (FR-CAP-7): mic channel goes silent (recording + transcript + meter),
+// loopback keeps capturing. Returns the new muted state; emits recording://mic. Errs if the mic is off.
+toggle_microphone_mute(input: { meetingId: MeetingId }): boolean
 set_recording_retention(input: { meetingId: MeetingId; record: boolean }): void  // toggle mid-meeting (FR-REC-1)
 acknowledge_recording_consent(): void                                            // one-time (FR-REC-2)
 
@@ -36,8 +39,14 @@ update_live_notes(input: { meetingId: MeetingId; markdown: string }): void
 set_segment_note(input: { meetingId: MeetingId; anchorMs: number; text: string }): void
 
 // ---- Hardware ----
-hardware_status(): { backends: BackendInfo[]; active: BackendId; modelSize: string; estRtf: number }
+hardware_status(): { backends: BackendInfo[]; active: BackendId; modelSize: string; estRtf: number }  // re-detects fresh each call — backs the Settings ▸ Hardware "Refresh" button
 set_preferred_backend(input: { backend: BackendId | "auto" }): void
+// Launch-at-login (NFR-RES-4). Writes/removes a per-user OS Run entry (no admin) and persists auto_start. Opt-in, off by default.
+set_auto_start(input: { enabled: boolean }): void
+// Device test: opens a mic ("input") or the render device in loopback ("loopback") for a few seconds and streams device://level (no recording, no retained audio). Refused while recording.
+monitor_audio_level(input: { kind: "input" | "loopback"; deviceId?: string; durationMs?: number }): void
+// Quick stress test: benchmarks each available backend × installed model (≤3 sizes) on a fixed sample, returns per-pair real-time factor + the most-accurate real-time-capable recommendation. Emits stress://progress. Refused while recording.
+stress_test_hardware(): { results: { backend: string; model: string; rtf: number; realtime: boolean }[]; recommended: { backend: string; model: string } | null }
 
 // ---- Transcription / models ----
 // language (T8.7, M4.2): omitted reuses the meeting's current language rather than resetting it.
@@ -79,6 +88,8 @@ export_meeting(input: { meetingId: MeetingId; dest: string; format: "md" | "pdf"
 // they enqueue for finalize-trigger targets and pump in the background. SHA-256 dedup means an
 // edit that didn't alter a file uploads nothing.
 update_notes(input: { meetingId: MeetingId; markdown: string }): void
+// AI-enhance rough notes into structured Markdown grounded in the transcript (Granola-style), via the configured LlmProvider (no new egress). Takes the live buffer, returns the enhanced text WITHOUT persisting — the UI keeps or undoes it. Refused while recording; errors with no provider.
+enhance_notes(input: { meetingId: MeetingId; notes: string }): string
 // SearchHit = MeetingListItem fields (id, title, started_at, duration_secs, status, tags) + snippet: string
 search(input: { query: string }): SearchHit[]                 // FTS (FR-SEARCH-1)
 set_tags(input: { meetingId: MeetingId; tags: string[] }): void
@@ -93,6 +104,13 @@ bulk_export_meetings(input: { destDir: string; format: "md" | "pdf" | "docx" | "
 // folder of them (from a bulk export). Each is reconstructed under a fresh meeting id (original
 // title/date/duration/speakers/tags/action items preserved). Returns the count imported.
 import_meeting_bundle(input: { dir: string }): number
+// Add a meeting from an existing recording: a local audio/video file path or a URL (YouTube/
+// streaming page or direct media URL). Needs ffmpeg (+ yt-dlp for URLs) on PATH; neither bundled.
+// `model` overrides the Settings whisper model for this import (recorded as meeting.model_used).
+// Returns the new meeting id IMMEDIATELY (status "transcribing"); transcode→transcribe→diarize→
+// finalize run in the background, streaming import://progress and ending with transcript://finalized.
+// A failed import is left in the list with status "error" (not deleted).
+import_media(input: { source: string; title?: string; model?: string }): MeetingId
 
 // ---- LLM / AI provider (ADR-0007/0011) ----
 // provider ∈ ollama | custom | anthropic | openai | off. Hosted-provider API keys are passed to
@@ -183,9 +201,11 @@ privacy_self_check(): {
 "recording://state"      { meetingId, state: "recording"|"paused"|"stopped"|"cancelled", elapsedMs }
 "recording://level"      { meetingId, rms: number, peak: number }            // waveform (FR-CAP-5)
 "recording://device"     { meetingId, recovered: boolean, message: string }  // capture device change (FR-CAP-6)
-"transcript://segment"   { meetingId, segment: TranscriptSegment }            // live segments (FR-TRX-2)
+"recording://mic"        { meetingId, muted: boolean }                        // mic mute toggled (FR-CAP-7)
+"transcript://segment"   { meetingId, segment: TranscriptSegment }            // live segments (FR-TRX-2); may re-emit a committed segment with a refined `speaker` — replace by `segment.id`
 "transcript://finalized" { meetingId, segmentCount }
-"diarization://updated"  { meetingId, speakers: SpeakerInfo[] }              // after post-pass (FR-SPK)
+"import://progress"      { meetingId, phase: "prepare"|"transcribe"|"diarize"|"finalize", state: "active"|"done"|"error", elapsedMs: number|null, error: string|null }  // background import_media tracker
+"diarization://updated"  { meetingId, speakers: SpeakerInfo[] }              // post-pass AND live 15s provisional passes (FR-SPK); carries "You" once the mic voiceprint matches
 "llm://token"            { meetingId, text }                                  // streamed summary (FR-LLM-4)
 "llm://done"             { meetingId, summary: SummaryFile }                  // full summary.json contents, not just a pointer
 "model://progress"       { id, receivedBytes, totalBytes }
@@ -193,6 +213,8 @@ privacy_self_check(): {
 "calendar://linked"      { ok: boolean, error?: string }                     // MS Graph OAuth handshake settled (M4.4)
 "calendar://progress"    { processed, total }                                // MS Graph import (M4.4)
 "hardware://changed"     { active: BackendId, reason: string }               // fallback occurred (FR-HW-4)
+"device://level"         { kind: "input"|"loopback", rms?, peak?, done?: boolean }  // Settings device test meter; done=window ended
+"stress://progress"      { backend: string, model: string }                  // quick stress test, per pairing benchmarked
 "recording://retention"  { meetingId, record: boolean }                      // retention toggled (FR-REC-1/3)
 "sync://job"             { jobId, meetingId, targetId, artifact, status, bytesSent, bytesTotal }  // FR-SYNC-5
 "sync://done"            { meetingId, targetId, uploaded: number, failed: number }

docs/adr/0004-hardware-acceleration.md

@@ -44,3 +44,20 @@ Detection mapping:
 ## Revisit if
 Windows ML supersedes DirectML for our models, or a vendor SDK gives materially better NPU
 throughput than the ONNX/DirectML path.
+
+## Update (2026-07-16): NPU path uses OpenVINO; decoder split off the CPU
+
+As built (T3.4 + follow-up), the NPU tier runs on **ONNX Runtime + the OpenVINO EP**
+(`device_type=NPU`), not DirectML — Intel's prebuilt OpenVINO ORT runtime is downloaded on
+demand (DirectML remains the *GPU* fallback EP for AMD/Intel in non-Vulkan builds). The engine
+splits Whisper across devices:
+
+- **encoder** (fixed shape): OpenVINO **NPU**;
+- **decoder** (autoregressive, KV-cache greedy via Optimum's merged export): OpenVINO **GPU**
+  (Intel iGPU, same runtime bundle) when available, else the CPU EP —
+  `WA_ONNX_DECODER_DEVICE=cpu|gpu` overrides.
+
+Rationale: the dynamic-shaped decode loop is a poor NPU fit, and pre-KV-cache it re-ran the full
+token prefix every step on the CPU — the dominant CPU cost while transcribing. With the merged
+decoder + iGPU placement, sustained CPU during transcription is near-idle on NPU+iGPU systems.
+The engine logs both EPs at load (`transcription engine: ONNX …`).

docs/adr/0005-diarization-sherpa-onnx.md

@@ -32,6 +32,54 @@ labels in the post-meeting pass; this keeps latency low while improving final ac
 - Speaker IDs (`S1`, `S2`, …) are internal and stable per meeting; name mappings live in the DB
   and are applied at render/export time, never destructively rewritten onto segments.
 
+### Phase 3 refinement — per-stream "You" attribution (FR-SPK, 2026-07-14)
+
+The single-pass-over-the-whole-recording model above blind-clusters a **summed mono** signal
+(mic + loopback), then guesses which cluster is the user via a voiceprint match. In practice that
+clustering is unreliable in both directions on the summed signal — it over-split a 2-speaker call
+into 83 clusters at one threshold and merged two clearly distinct voices into one at another — and
+the voiceprint can only *label* a cluster, never *create* the separation.
+
+So when the **microphone is enabled**, WA no longer relies on clustering to find the user:
+
+- A live **mic-activity timeline** (`audio::MicActivity`) records, per 100 ms of `audio.wav` frame
+  time, when the mic was speech-level — captured in the loopback writer, the one place the mic and
+  loopback exist separately in the recording's own timebase. Those ranges become **"You"** spans
+  directly (no clustering, no embedding).
+- sherpa clustering is then run over the recording with the "You" ranges **masked out** (zeroed),
+  so it only ever splits the **far side** into `Speaker N`. The mic can never pollute or merge into
+  a far-side cluster.
+- The timeline is persisted as `mic_activity.json` (retained meetings) so `reprocess_transcript`
+  re-attributes identically without the live capture.
+
+The original blind-clustering + voiceprint pass **remains the fallback** for mic-off recordings and
+imports (no timeline). Naming stays uniform (`build_name_map`: "You", then `Speaker 2…`). Segment
+IDs and the names-in-DB rule are unchanged; only the *source* of the spans changes.
+
+### Phase 3.5 refinement — dual-channel capture (supersedes Phase 3's timeline, 2026-07-14)
+
+The Phase 3 timeline above reconstructed the mic/far-side split *after the fact* from a summed-mono
+`audio.wav` plus a `mic_activity.json` sidecar, masking the mic ranges before clustering. That was
+reliable at stop but **fragile on reprocess** (re-aligning a sidecar against a mono mix — it
+collapsed to a single speaker) and carried a parallel persistence path.
+
+**Decision:** when the mic is enabled, record `audio.wav` as **stereo with the streams separated —
+left = microphone ("You"), right = system/loopback ("Speaker")** instead of summing them. The
+separation is then intrinsic to the recording:
+
+- Diarization runs on the **right channel only** → `Speaker N`; "You" comes from **left-channel**
+  voice activity (`vad_spans`). The mic is never clustered, by construction.
+- Reprocess recomputes both from the file — no sidecar, no masking; stop and reprocess agree.
+- Transcription still downmixes (L+R) to the same summed mono; playback folds to mono and bundle
+  export folds to dual-mono so shared/played audio is normal.
+- A `meetings.audio_layout` flag (`summed` | `split`, carried in the bundle manifest) distinguishes
+  recordings; existing `summed` recordings keep the blind-clustering + voiceprint path. On a stereo
+  render endpoint this is **size-neutral** (the file was already 2ch with the mic summed into both).
+
+This **retires** `MicActivity` / `mic_activity.json` / masked diarization from Phase 3 (see the
+retirement note at `attribute_split` in `commands.rs`); the capture-silence handling,
+`diarize_samples`, and `build_name_map` naming are retained. Segment IDs and names-in-DB unchanged.
+
 ## Revisit if
 A single model gives joint ASR + diarization with better accuracy, or whisper.cpp gains
 production diarization.

docs/adr/0012-startup-and-enterprise-deployment.md

@@ -0,0 +1,54 @@
+# ADR-0012 — Launch-at-login & enterprise deployment defaults
+
+- **Status:** Accepted
+- **Date:** 2026-07-14
+- **Context source:** User request (2026-07-14) — auto-start on boot; customize an installation
+  (install location, per-user/all-users, default settings) with native Windows tooling for
+  mass-deployment.
+
+## Context
+
+Two related needs. (1) Users want WhispAssist to **launch automatically at login**. NFR-RES-4
+forbids adding WA to OS startup without explicit consent, so this must be opt-in. (2) An admin
+mass-deploying WA to many machines wants to **customize the deployment** — install location, whether
+it installs per-user (no admin) or all-users, and the app's default settings (record-by-default,
+preferred backend, a model to pre-download, retention, AI provider) — using **native Windows tools**
+(GPO / SCCM / Intune / `msiexec` / silent NSIS), not a bespoke management console.
+
+## Decision
+
+1. **Launch-at-login is opt-in, off by default.** A `set_auto_start` command uses
+   `tauri-plugin-autostart` to write a **per-user** `HKCU\...\Run` entry (no admin) and persists an
+   `auto_start` setting. Startup reconciles the OS entry to that flag (restores it after a reinstall).
+   Nothing runs on a timer — this is a registry entry, not a background process (NFR-RES-1).
+2. **Install location & scope are native, no app code.**
+   - Location: `msiexec INSTALLDIR=…` (MSI) / NSIS `/D=…` (silent).
+   - Scope: NSIS `installMode: "both"` — the `.exe` setup lets the user choose **current-user
+     (no admin)** or **all-users (admin)**. The MSI stays per-machine as the enterprise all-users
+     artifact.
+3. **Default settings via a first-run `wa-defaults.ini`.** On a machine's **first** launch (before
+   `settings.json` exists) WA reads an admin-supplied INI — `%PROGRAMDATA%\WhispAssist\wa-defaults.ini`
+   first, else the bundled `<install dir>\wa-defaults.ini` — and seeds `settings.json` from it, with an
+   optional `auto_download_model` to pre-fetch the model. The shipped template is fully commented, so
+   a normal install is unaffected. This is deployable purely by copying a file with existing Windows
+   management tooling; no WiX custom actions.
+
+## Consequences
+
+- **Positive:** opt-in startup honors NFR-RES-4 with zero idle cost; install location/scope reuse the
+  installers' native behavior (no custom code to maintain); one small INI + a first-run guard covers
+  the whole deployment-customization surface and works for MSI, NSIS, and portable copies alike.
+- **Guardrail — no secrets in the deploy file (CLAUDE.md):** the INI must never carry credentials.
+  `deploy.rs` drops any key containing `key`/`token`/`secret`/`credential`/`password` as defense in
+  depth; API keys, OAuth tokens and sync passwords remain in the OS credential store only. Seeding a
+  provider (e.g. `llm_provider=anthropic`) still requires the admin/user to provision its key
+  separately — no new egress path is created by the file.
+- **Negative / care:** the seed runs only when `settings.json` is absent (truly first run); it does
+  **not** re-apply on upgrade, matching "the user's own settings win thereafter." Array config merges
+  in Tauri **replace** rather than append, so `wa-defaults.ini` must be listed in both
+  `tauri.conf.json` and `tauri.vulkan.conf.json` bundle resources (the shipped build uses the latter).
+
+## Revisit if
+
+Admins need per-machine policy that **overrides** user settings on every launch (not just seeds
+first-run), or a signed/locked-down enterprise policy channel beyond a plain INI.

docs/enterprise-deployment.md

@@ -0,0 +1,101 @@
+# Enterprise deployment
+
+How to mass-deploy WhispAssist and preset its defaults with native Windows tooling (Group Policy,
+SCCM, Intune, `msiexec`, silent NSIS). No management console, no phone-home. See ADR-0012.
+
+WhispAssist ships two bundles:
+
+| Bundle | Scope | Admin? |
+|---|---|---|
+| `WhispAssist_<ver>_x64_en-US.msi` | Per-machine (all users) | Yes |
+| `WhispAssist_<ver>_x64-setup.exe` (NSIS) | Current-user **or** all-users (prompts) | Only for all-users |
+
+## Install location
+
+- **MSI:** `msiexec /i WhispAssist_<ver>_x64_en-US.msi INSTALLDIR="D:\Apps\WhispAssist" /qn`
+  (`INSTALLDIR` is Tauri's WiX install-dir property; confirm against the generated `.wxs` if a build
+  changes it.)
+- **NSIS:** `WhispAssist_<ver>_x64-setup.exe /S /D=D:\Apps\WhispAssist`
+  (`/S` = silent, `/D=` = install dir; `/D=` must be **last** and unquoted per NSIS.)
+
+## Install scope (per-user vs all-users)
+
+The NSIS `.exe` shows a "current user / all users" page. **Current user needs no admin** and installs
+under the user profile; **all users** requires elevation. Silent all-users:
+`WhispAssist_<ver>_x64-setup.exe /S`. The MSI is always per-machine (all-users) and requires admin.
+
+## Auto-start at login
+
+Off by default (NFR-RES-4). Turn it on for the user either in-app (Settings ▸ Recording ▸ *Launch
+WhispAssist at login*) or by presetting `auto_start = true` in `wa-defaults.ini` (below). It installs
+a **per-user** `HKCU\Software\Microsoft\Windows\CurrentVersion\Run` entry — no admin, and it does
+**not** start recording on its own.
+
+## Preset default settings — `wa-defaults.ini`
+
+On a machine's **first** launch (before `settings.json` exists), WhispAssist reads an admin-supplied
+INI and seeds that user's `settings.json`. After that the user's own settings win and the file is
+ignored. First location found wins:
+
+1. `%PROGRAMDATA%\WhispAssist\wa-defaults.ini` — machine-wide. Deploy with a GPO/SCCM/Intune file copy.
+2. `<install dir>\wa-defaults.ini` — the template shipped next to the executable.
+
+The shipped template is fully commented out, so a default install behaves as if it were absent.
+Uncomment and edit the keys you want to preset.
+
+### Format
+
+Flat `key = value`, one per line. `;` and `#` comment lines and `[section]` headers are ignored.
+`true`/`false` become switches, plain numbers become numbers, everything else is text. Unknown or
+misspelled keys are ignored.
+
+> **Never put secrets in this file.** API keys, OAuth tokens and sync passwords live only in the OS
+> credential store. Any key containing `key`, `token`, `secret`, `credential` or `password` is
+> dropped on read. Presetting `llm_provider = anthropic` still requires the key to be provisioned
+> separately — the file adds no egress path.
+
+### Keys
+
+| Key | Values | Meaning |
+|---|---|---|
+| `default_record` | true/false | Record every meeting by default (consent notice still applies). |
+| `preferred_backend` | auto\|npu\|nvidia\|amd\|intel\|cpu | Transcription backend. |
+| `whisper_model` | catalog id (e.g. `base.en-q5_1`) | Default transcription model. |
+| `auto_download_model` | true/false | Fetch `whisper_model` in the background on first launch. |
+| `whisper_language` | auto\|ISO-639-1 | Default language (multilingual model only). |
+| `low_overhead` | true/false | CPU + smallest model preset. |
+| `storage_root` | path | Where meetings are stored. |
+| `retention_max_age_days` | number | Delete meetings older than N days. |
+| `retention_max_size_gb` | number | Cap total storage at N GB. |
+| `llm_provider` | ollama\|custom\|anthropic\|off | Summary provider (key provisioned separately). |
+| `llm_endpoint` | url | LLM endpoint. |
+| `llm_model` | text | LLM model name. |
+| `microphone_enabled` | true/false | Capture the user's mic into the transcript. |
+| `auto_record_calendar` | true/false | Auto-start recording on calendar events (app open only). |
+| `theme` | system\|light\|dark | UI theme. |
+| `auto_start` | true/false | Launch WhispAssist at login (per-user Run entry). |
+| `sync_enabled` | true/false | Sync master switch (targets/credentials configured in-app). |
+
+### Example
+
+```ini
+default_record       = true
+preferred_backend    = npu
+whisper_model        = small.en-q5_1
+auto_download_model  = true
+retention_max_age_days = 90
+auto_start           = true
+```
+
+## Silent end-to-end example
+
+```bat
+:: 1. Push machine-wide defaults (as SYSTEM via GPO/SCCM)
+mkdir "%ProgramData%\WhispAssist"
+copy wa-defaults.ini "%ProgramData%\WhispAssist\wa-defaults.ini"
+
+:: 2. Install per-machine, custom location, no UI
+msiexec /i WhispAssist_<ver>_x64_en-US.msi INSTALLDIR="C:\Program Files\WhispAssist" /qn
+```
+
+Each user's first launch then seeds their `settings.json` from the machine-wide file.

docs/plans/2026-07-13-diarization-phase3-per-stream-attribution.md

@@ -0,0 +1,243 @@
+# Phase 3 design sketch: per-stream speaker attribution (FR-SPK)
+
+Status: **signed off & in progress**, 2026-07-13. Deepens Phase 3 of
+[`2026-07-13-diarization-speaker-accuracy.md`](./2026-07-13-diarization-speaker-accuracy.md).
+The §§1–8 below are the original sketch (ADR-0005 review + alternatives); the **finalized
+design** the user signed off on is at the top and supersedes the RMS-dominance heuristic.
+
+---
+
+## Finalized design (signed off 2026-07-13)
+
+Simpler than the sketch's dominance heuristic, because the user records on **headphones** so the
+mic never hears the far side (no bleed):
+
+> **The microphone stream is "You" by definition. Diarization runs on the loopback (speaker)
+> audio only, never the mic. Any loopback speaker is "Speaker N".**
+
+Mechanism, all in **`audio.wav` frame time** (so You-spans and sherpa-spans share one clock):
+
+1. **Mic-activity timeline** — captured in the loopback writer, the one place mic and loopback
+   exist separately in `audio.wav`'s timebase (`capture_loop`, where `bridge.pull()` is folded
+   in). Per 100 ms, flag whether the pulled mic chunk was speech-level (reuse `VOICE_ENERGY_FLOOR`).
+   Held in a shared `MicActivity` (sibling to `VoiceSample`); persisted for reprocess.
+2. **At stop / reprocess:**
+   - Collapse active windows → **"You" spans** (dropping sub-`MIN_SPAN_MS` blips).
+   - **Mask** the You ranges out of `audio.wav` (zero them) and run sherpa on the remainder → far
+     side splits into **"Speaker N"** spans. Clustering never sees the mic, so the mic can't
+     pollute or merge with a far-side cluster.
+   - Merge span lists; `assign_by_overlap` unchanged.
+3. **Voiceprint (Phase 1/2) becomes the fallback** for meetings with no timeline (imports, older
+   recordings). Live-tick voiceprint "You" is retired once the timeline drives labels.
+
+Timebase note: with the capture-silence fix (FR-CAP-7) `audio.wav` is now continuous, so the
+frame-indexed timeline aligns cleanly with what sherpa reads back via `read_wav_mono_16k`.
+
+Implementation status (all landed 2026-07-13/14): **(a)** `MicActivity` struct + capture wiring —
+done; **(b)** stop-path attribution + masked diarize (`phase3_attribute`) — done; **(c)** persist
+`mic_activity.json` + reprocess uses it (imports fall back, no timeline) — done; **(d)** ADR-0005 +
+`docs/02`/`03` amended — done. **Not yet done:** empirical user test; and the *live* 15s tick still
+uses the Phase 2 voiceprint, not the timeline (a later polish so live labels match the stop result).
+
+---
+
+_Original sketch (ADR-0005 review, risks, alternatives) follows._
+
+Phases 0–2 shipped: cluster explosion tamed (0.7 + 700 ms floor), reprocess re-diarizes with a
+persistent `voiceprint.wav`, and live "You" reaches the UI. Those all still treat diarization as
+**blind clustering of a summed mono signal**, then guess which cluster is the mic. Phase 3
+removes the guess for the mic side by using information WA throws away today.
+
+---
+
+## 1. ADR-0005 review — what it commits us to
+
+ADR-0005 (Accepted, 2026-06-30) decides: sherpa-onnx offline diarization (pyannote segmentation
++ ERes2Net embedding + clustering), behind the `diarization::Diarizer` trait, run as a
+**post-processing pass over recorded audio**, aligned to whisper segments by **timestamp
+overlap**. Speaker IDs (`S1`…) are internal/stable; names map in the DB, applied at render time,
+**never rewritten onto segments**.
+
+What Phase 3 must respect vs. what it may change:
+
+- **Keeps (contract-level):** the `Diarizer` trait, post-pass-over-audio model, `SpeakerSpan`
+  → segment alignment by overlap (`assign_by_overlap`), stable internal labels, names-in-DB.
+  Phase 3 produces *more* `SpeakerSpan`s from a better source; it does not rewrite segments or
+  move naming.
+- **Bends (needs the ADR noted/updated):** ADR-0005 assumes one clustering pass over "the
+  recorded audio." Phase 3 introduces a **second, non-clustered source of spans** (the mic
+  activity timeline) and restricts clustering to the **far-side** audio. That is new enough to
+  warrant an ADR amendment or a short ADR-0005a, because a future reader will otherwise expect
+  all spans to come from sherpa.
+- **ADR's "Revisit if"** (joint ASR+diarization model, or whisper.cpp diarization) is unrelated
+  — Phase 3 is orthogonal and does not trigger it.
+
+Conclusion: Phase 3 fits inside the trait and the overlap-alignment contract. The only doc debt
+is recording that "mic-dominant spans bypass clustering," which is a genuine deviation from
+ADR-0005's single-source assumption. **Recommend: amend ADR-0005** (Consequences section) rather
+than a new ADR — same decision, refined.
+
+---
+
+## 2. The asset we currently discard
+
+When the mic is enabled (FR-CAP-7), `start_recording` (`commands.rs:342`) wires two independent
+16 kHz-mono streams into `spawn_mixer` (`audio/mod.rs:1020`). The `Mixer` (`audio/mod.rs:979`)
+holds `loopback` and `mic` **time-aligned in the same buffer** and sums them sample-for-sample
+into the transcript stream — then the per-stream identity is **gone**. Diarization later reads
+`audio.wav` (the summed signal) and has to reverse-engineer which cluster was the mic. That
+reversal is the entire reason `voiceprint.rs` exists.
+
+Two facts make this cheap to exploit:
+
+1. **Per-stream RMS already exists.** `audio_level(mono, mic)` (`audio/mod.rs:522`) runs on every
+   WASAPI chunk for *both* directions (`is_loopback` at :564, emitted at :678). We already know,
+   per ~10 ms chunk, how loud each side is. We just don't persist it against time.
+2. **The two streams are already aligned at the mixer**, in the **same timebase the transcript
+   segments derive from** (the streaming worker consumes the mixer's summed output; segment
+   timestamps are cumulative-samples-fed / 16). So a timeline indexed by cumulative mixer-output
+   samples shares the segments' clock — tighter than sherpa spans, which come from `audio.wav`.
+
+---
+
+## 3. Core idea
+
+> The mic stream is a **known** speaker. Don't cluster to find it — record when it's dominant,
+> call those spans "You", and run sherpa only on the **far-side** audio to split the *other*
+> participants.
+
+Pipeline at stop (mic enabled + retained):
+
+```
+mixer  ─┬─► loopback samples ──► [far-side WAV] ──► sherpa cluster ──► Speaker 2..N spans
+        │                                                                      │
+        └─► mic vs loopback RMS per 100 ms ──► mic-dominant ranges ──► "You" spans
+                                                                               │
+                              merge span lists ──► assign_by_overlap ──► segments
+```
+
+The mic-dominant spans need no embedding, no clustering, no voiceprint — they are attribution by
+construction. Voiceprint (Phases 1–2) stays as the **fallback** for meetings without a timeline
+(imports, pre-Phase-3 recordings, mic-disabled meetings).
+
+---
+
+## 4. The two real risks (why this needs sign-off, not just a ticket)
+
+### 4a. Acoustic bleed — the mic is not purely "You"
+
+Without echo cancellation (WA has none — fully local, minimal), a user on **speakers** (not
+headphones) has the far side playing into the room and back into the mic. Naive "mic has signal →
+You" would attribute the far side to the user whenever the far side is loud.
+
+Mitigation (this is why the parent plan says *mic-**dominant**,* not *mic-active*): a range is
+"You" only when **mic RMS meaningfully exceeds loopback RMS** in that window (e.g.
+`mic_rms > k · loopback_rms` with `k ≈ 2`, plus an absolute mic-VAD floor so silence isn't
+"You"). When both are comparably loud → treat as far-side/ambiguous, let sherpa/overlap decide.
+This degrades gracefully: headphone users get near-perfect mic isolation; speaker users get
+"You" only on clear self-speech and fall back to the old behavior during overlap. **`k` is a
+calibration knob, not a constant to bury** — real rooms differ.
+
+### 4b. Timebase — three clocks, currently reconciled by tolerance
+
+- **Segments** are in *mixer-output* time (streaming worker over the summed stream).
+- **sherpa spans** are in *`audio.wav`* time (loopback thread's byte-accurate WAV, mic bridged in
+  at native rate then resampled to 16 kHz on read).
+- The **new mic-timeline** would be in *mixer-output* time.
+
+Today `assign_by_overlap` already bridges segments↔sherpa-spans across the first two clocks and
+tolerates the small skew. Phase 3 merges a *third* source. The mic-timeline is actually the
+**closest** to segment time (shared origin), so aligning You-spans to segments is *more* reliable
+than sherpa's. The residual risk is only that You-spans (mixer time) and Speaker-2..N spans
+(`audio.wav` time) coexist in one merged list with a small relative skew at the boundaries. In
+practice that costs at most a fraction of a `MIN_SPAN_MS` at each turn boundary — below segment
+granularity. **Verification item, not a blocker.** If it proves visible, the clean fix is to run
+the far-side sherpa pass over the mixer's *loopback lane* buffered to a WAV rather than over
+`audio.wav`, putting everything on one clock; heavier, so deferred until measured.
+
+---
+
+## 5. Concrete design
+
+**Capture (live, near-zero cost).** Add a mic-activity accumulator the mixer writes to. In
+`Mixer::drain_ready` (or the `mixer_loop`), for each emitted window compute `mic_rms` and
+`loopback_rms` over the aligned overlap (samples are already in hand there) and append a coarse
+flag per 100 ms:
+
+```rust
+// audio/mod.rs — new, alongside Mixer
+struct MicActivity { flags: Vec<bool>, /* one per 100 ms of mixer output */ }
+// ponytail: Vec<bool> at 10 Hz = 10 bytes/s ≈ 36 KB/hour. No ring buffer, no config.
+```
+
+Expose it via the same `Arc` sharing pattern `VoiceSample` uses so `stop_recording` can read it.
+Reuse `audio_level`'s RMS math — do **not** add a second RMS implementation.
+
+**Persist (retained meetings).** At stop, alongside `voiceprint.wav`, write the timeline as
+`mic_activity.json` (or pack into an existing artifact) so **reprocess** can reuse it — same
+ADR-0009 retention/consent gate as `voiceprint.wav`, same lives-and-dies-with-`audio.wav` rule
+(delete it everywhere `audio.wav` is deleted). Tiny, plaintext-or-sealed to match.
+
+**Attribute (stop + reprocess).**
+1. Collapse the flag timeline into `You` `SpeakerSpan`s (merge adjacent flags, drop runs
+   `< MIN_SPAN_MS`, reuse Phase 0's floor).
+2. Build a **far-side-only** 16 kHz signal: zero out (or excise) mic-dominant ranges from the
+   loopback lane, run the existing `SherpaDiarizer::diarize` over it → `Speaker 2..N` spans
+   (label numbering starts at 2, mirroring `voiceprint::build_name_map`).
+3. Merge the two span lists; `assign_by_overlap` unchanged.
+4. Fallback chain: no timeline → today's Phase-1/2 voiceprint path; no models → raw `S1`.
+
+**Contract/doc changes.**
+- Amend **ADR-0005** Consequences: "mic-dominant spans are attributed directly from a per-stream
+  activity timeline and bypass clustering; sherpa runs on the far-side residual only."
+- `docs/02-architecture.md` data-flow: note the mixer emits a mic-activity side-channel.
+- `docs/03-data-model.md`: add `mic_activity.json` to the meeting-dir list (like `voiceprint.wav`).
+- No new egress; no new default-on setting; `k` threshold lives in code with a calibration comment.
+
+**Tests.**
+- Unit: flag-run → span collapse (boundaries, sub-`MIN_SPAN_MS` drop) — pure, table-driven.
+- Unit: dominance classifier (`mic`, `loop` RMS pairs → You / far / ambiguous), incl. the bleed
+  case (both loud → not You).
+- Manual: 2-person call on **headphones** → exactly "You" + 1 speaker, near-zero errors; repeat
+  on **speakers** → verify bleed doesn't mislabel far-side as You.
+
+---
+
+## 6. Alternatives considered (and why not)
+
+- **Do nothing / stop at Phase 2.** Legitimate. Phases 0–2 already fix the reported bugs
+  (83-cluster explosion, reprocess collapse, live "You"). Phase 3 is an **accuracy ceiling
+  raise**, not a bug fix. *If 2-person calls are the dominant case and Phase 2's voiceprint "You"
+  tests well in the field, Phase 3 may not clear the cost/benefit bar — measure Phase 2 first.*
+  This is the ponytail-honest recommendation: **verify Phase 2 empirically before building 3.**
+- **"Mic active → You" (no dominance test).** Simpler, but wrong under speaker bleed (4a). The
+  dominance test is the minimum that survives real rooms.
+- **Full source separation / AEC.** Much larger, new deps, defeats "minimal + local." The
+  dominance heuristic gets ~90 % of the benefit for ~5 % of the effort.
+- **Better voiceprint only** (bigger sample, re-match every tick — already done in Phase 2).
+  Still bottlenecked on sherpa's ability to *form* a clean mic cluster in the summed signal;
+  Phase 3's point is to stop relying on that.
+
+---
+
+## 7. Open questions for sign-off
+
+1. **Build it now, or verify Phase 2 first?** Recommendation: run the empirical 2-person
+   re-diarization (outstanding for Phases 0–2) and a Phase-2 live "You" check **before**
+   committing to Phase 3. If Phase 2 accuracy is acceptable, Phase 3 becomes optional.
+2. **Dominance threshold `k` and mic-VAD floor** — accept a code-constant default with a
+   calibration comment (proposed), or expose a hidden setting? Recommendation: constant first.
+3. **Far-side signal construction** — zero-out mic-dominant ranges in the loopback lane (simple,
+   preserves timeline) vs. excise-and-concat (shorter audio, shifts timestamps → needs remap).
+   Recommendation: **zero-out**, keeps one timebase.
+4. **ADR: amend 0005 vs. new ADR-0005a?** Recommendation: amend.
+5. **Scope of `mic_activity.json`** — persist for reprocess (proposed) or compute live-only and
+   accept that reprocess falls back to voiceprint? Persisting is cheap and keeps reprocess at
+   parity.
+
+## 8. Rough effort
+
+Backend-only until the far-side pass; no frontend change (labels already flow via Phase 2). Est.
+~1–1.5 days: mixer side-channel + accumulator (½ day), span collapse + far-side pass + merge
+(½ day), persistence + reprocess wiring + docs/tests (½ day). Contained, reversible, behind the
+existing mic-enabled + retention gates.

docs/plans/2026-07-13-diarization-speaker-accuracy.md

@@ -0,0 +1,137 @@
+# Fix plan: speaker diarization accuracy ("You", live labels, 83-speaker explosion)
+
+Status: planned 2026-07-13 (v0.5.2). Investigation confirmed in code; see memanto memories
+`a9e5f29e` (diagnosis) and the entry referencing this file (fix order). Execute phases **in
+order** — each phase is independently shippable and verifiable, and later phases assume
+earlier ones landed.
+
+## Symptoms (user report, all reproduced against code)
+
+1. Live recording shows `S1` for every line; the user's own speech should show "You".
+2. Post-stop diarization of a 2-person conversation produced **83** speakers.
+3. Re-transcribing (English Medium) collapsed every segment to speaker "You".
+
+## Root causes (verified anchors)
+
+| # | Cause | Anchor |
+|---|-------|--------|
+| A | Live segments always carry the pre-diarization `"S1"` placeholder | `src-tauri/src/transcription/mod.rs:222`, emitted at `src-tauri/src/commands.rs:480` |
+| B | The 15s provisional diarization tick relabels only the backend buffer and emits `diarization://updated`, which **no frontend code listens to** (`src/lib/api.ts` has no listener); live view renders raw labels with no name map (`src/lib/views/TranscriptNotes.svelte:527`) | `src-tauri/src/commands.rs:509-567` |
+| C | "You" (mic voiceprint match) runs only in `stop_recording`, never live | `src-tauri/src/commands.rs:666-695`, `src-tauri/src/diarization/voiceprint.rs` |
+| D | `reprocess_transcript` never diarizes: fresh segments all default to `"S1"`, then the meeting's **stale** name map (`S1 → "You"` from the original run) is reused, so the whole transcript renders "You" | `src-tauri/src/commands.rs:1646-1742` |
+| E | Clustering over the **summed mic+loopback mono** WAV with `num_clusters: -1, threshold: 0.5` over-clusters badly (overlapped speech → mixed embeddings; short pyannote chunks → unstable ERes2Net embeddings) | `src-tauri/src/diarization/mod.rs:68-75` |
+
+---
+
+## Phase 0 — Tame the cluster explosion (do first; everything else is useless at 83 clusters)
+
+Smallest possible change, lands before Phase 1 so re-diarization doesn't reproduce the
+83-speaker mess.
+
+1. In `SherpaDiarizer::new` (`src-tauri/src/diarization/mod.rs:73`) raise `threshold` from
+   `0.5` to `0.7` as a code constant with a comment naming the tuning evidence (83 clusters
+   for 2 speakers at 0.5 on mixed mono audio). sherpa fast-clustering semantics: larger
+   threshold → fewer clusters. Do **not** add a settings knob yet.
+2. In `segment_to_span`'s caller (`SherpaDiarizer::diarize`), drop spans shorter than
+   700 ms before returning — sub-second chunks carry unstable embeddings and only cause
+   label churn in `assign_by_overlap` (which already keeps a segment's prior label when no
+   span overlaps).
+3. Unit test: spans under the minimum are filtered; existing overlap tests still pass.
+4. **Verify empirically**: re-run diarization over the user's retained 2-person recording
+   (the 0.5.2 test meeting) and confirm the cluster count lands near 2–4, not 83. If 0.7
+   still over-clusters, try 0.8 before considering per-meeting configurability.
+
+Acceptance: 2-person retained recording diarizes to ≤4 speakers.
+
+## Phase 1 — `reprocess_transcript` re-diarizes + persistent mic voiceprint (fixes "everything is You")
+
+1. **Persist the voiceprint at stop.** In `stop_recording`
+   (`src-tauri/src/commands.rs` after the transcription worker join), when
+   `session.mic_voice_sample` exists **and** `session.retention` is on, write the mic
+   sample as 16 kHz mono `voiceprint.wav` next to `audio.wav` in `meeting_dir(&meeting_id)`.
+   ADR-0009 gate: it is retained audio of the user's voice, so it lives and dies with
+   `audio.wav` — every code path that deletes `audio.wav` (retention off at finalize,
+   delete-recording command) must delete `voiceprint.wav` too. Update
+   `docs/03-data-model.md` (meeting dir file list).
+2. **Re-diarize on reprocess.** In `reprocess_transcript` (`src-tauri/src/commands.rs:1646`),
+   after `transcribe_file`: build the diarizer via `diarizer_from_installed_models`
+   (inside `spawn_blocking`, same as `stop_recording`), `diarize(&wav_path)`, then
+   `assign(&mut segments, &spans)`. Missing models → skip gracefully (same degradation as
+   live).
+3. **Rebuild the name map instead of reusing the stale one.** Old labels are meaningless
+   after re-clustering. If `voiceprint.wav` exists, run
+   `voiceprint::match_mic_speaker` against the new spans to get a fresh
+   `You`/`Speaker N` map; otherwise use an empty map (raw `S1…` labels).
+   Pass `speaker_infos_from_segments(&segments, &new_names)` to `finalize_meeting`
+   instead of `meeting.speakers` (`src-tauri/src/commands.rs:1713`), and persist the new
+   names via the store the same way `stop_recording` does. Deliberate policy: user-typed
+   names from the original run are dropped on reprocess because they key to dead labels —
+   note this in the command's doc comment.
+4. `import_media` reuses this path (`commands.rs:1825` "same batch path"); confirm imported
+   meetings get diarized speakers too (they have no voiceprint — expect raw labels).
+5. Tests: unit test the fresh-name-map policy (stale map not reused); manual: re-transcribe
+   the 0.5.2 meeting with English Medium → segments show distinct speakers, the user's own
+   lines show "You".
+
+Acceptance: re-transcription yields per-speaker labels again, with "You" on the mic
+speaker when a voiceprint exists; never a single-speaker collapse.
+
+## Phase 2 — Live labels reach the UI, including live "You" (fixes "S1 for everyone")
+
+Backend (`src-tauri/src/commands.rs`, the 15s tick at 509–567):
+
+1. After `diarizer.assign` in the tick, **re-emit every committed segment whose speaker
+   changed** via the existing `transcript://segment` event (ids are stable; the frontend
+   store already replaces by id — `src/lib/stores/recording.svelte.ts:49-53`). No new
+   event needed for relabeling.
+2. In the same tick, run `voiceprint::match_mic_speaker` with
+   `session.mic_voice_sample.samples()` against the fresh spans (clusters re-shuffle every
+   tick, so match every tick; candidate audio is already capped at 10 s per cluster).
+   Merge results into `session.speaker_names` **without overwriting user-set names** —
+   same already-named guard as the post-stop pass (`commands.rs:676-683`). The tick's
+   existing `diarization://updated` emit then carries the "You" display name.
+3. Frontend: add an `onDiarizationUpdated` listener to `src/lib/api.ts` (payload
+   `{ meetingId, speakers: SpeakerInfo[] }`, already documented in
+   `docs/04-api-contracts.md:188`). Recording store gains a `speakers` state updated by
+   it; live rendering at `TranscriptNotes.svelte:527` passes that list to `speakerName`
+   (the finalized path at :344 already does this).
+4. Docs: note in `docs/04-api-contracts.md` that `transcript://segment` may re-emit a
+   committed segment with an updated `speaker` (replace-by-id contract).
+5. Tests: frontend store test — a re-emitted segment with the same id replaces the old
+   one; manual: during a live 2-person call, own speech flips to "You" within ~15–30 s.
+
+Acceptance: during recording, labels differentiate speakers and the mic speaker shows
+"You" while talking (within one tick), not just after stop.
+
+## Phase 3 — Per-stream attribution (accuracy end-game, larger change, needs design sign-off)
+
+Blind clustering of a summed mono signal is the ceiling on accuracy. WA knows which
+samples are mic before `spawn_mixer` sums them (`commands.rs:342-371`,
+`src-tauri/src/audio/mod.rs` MicBridge/mixer). Design sketch — do NOT start without
+reviewing ADR-0005 and getting sign-off, since it changes the diarization contract:
+
+- Record a coarse mic-activity timeline during capture (e.g. per-100 ms mic-RMS-dominant
+  flags, negligible memory).
+- At stop: mic-dominant ranges become "You" spans directly; run sherpa only over the
+  remaining (far-side) ranges to split the *other* participants; merge span lists before
+  `assign_by_overlap`.
+- Voiceprint match remains as fallback for meetings without the timeline (imports, old
+  recordings).
+- Update ADR-0005 (or add a new ADR) + `docs/02-architecture.md` data flow.
+
+Acceptance: 2-person call yields exactly "You" + 1 speaker with near-zero attribution
+errors on non-overlapping speech; overlapping speech attributes to the dominant stream.
+
+---
+
+## Cross-cutting rules for the executing agent
+
+- CLAUDE.md applies in full: `cargo fmt` + `cargo clippy -- -D warnings`, no
+  `unwrap()`/`expect()` on user-reachable paths, conventional commits referencing
+  FR-SPK-*, commit each file right after finishing it (one commit per file), docs updated
+  in the same change when contracts move.
+- Memory discipline: `memanto agent activate whispassist` first; `memanto remember` every
+  decision/tuning result (especially the empirical threshold from Phase 0) with full
+  metadata; record phase completion so the next agent knows where to resume.
+- No new egress, no new settings defaults ON; `voiceprint.wav` is local retained audio
+  under the existing ADR-0009 consent/retention gate.

docs/plans/2026-07-14-stereo-dual-channel-capture.md

@@ -0,0 +1,150 @@
+# Plan: dual-channel capture (mic = left, loopback = right) — FR-SPK / FR-CAP
+
+Status: **implemented** 2026-07-14 (all 3 sign-off answers: DB column; retire MicActivity with a
+code note; export → dual-mono). Migration 0009 + `write_wav_split` + `attribute_split` +
+`fold_wav` + ADR-0005 Phase 3.5. Supersedes the summed-mono + masked-diarize approach
+in [`2026-07-13-diarization-phase3-per-stream-attribution.md`](./2026-07-13-diarization-phase3-per-stream-attribution.md).
+The proposed **ADR-0005 amendment** is at the bottom — apply it on approval.
+
+## Motivation
+
+Every diarization problem this month traces to one root: **`audio.wav` is a summed mono mix** of
+mic + loopback, so the two sources are irrecoverable after the fact. Phase 3 works around that by
+(a) recording a separate `mic_activity.json` "You" timeline and (b) diarizing the mix with the You
+ranges masked out. It works at stop but is fragile on **reprocess** — the persisted timeline has to
+re-align against freshly-transcribed segments over a mono mix, and it currently collapses to a
+single "Speaker 2". The timelines are being written correctly (verified: real `you_spans`), so the
+failure is the masked-mono re-alignment itself, not the capture.
+
+**Fix the root, not the instance:** keep both sources *physically separate in the recording* by
+recording **stereo — left channel = microphone ("You"), right channel = system/loopback
+("Speaker")**. The separation then lives in the file forever; reprocess and diarization always have
+both streams; no sidecar timeline, no masking, no summed-mono clustering.
+
+## Key facts that make this cheap
+
+- `audio.wav` is **already 2-channel** on a normal (stereo) render endpoint — today we *sum the mic
+  into both channels*. Writing mic→L and loopback(downmixed)→R is the **same channel count, same
+  bytes** — no size increase in the common case.
+- Transcription already downmixes to mono: `read_wav_mono_16k` averaging L+R of a split file yields
+  `(mic + loopback)/2` = the same summed signal it gets today, so **transcription is unchanged**.
+- The mic↔loopback alignment machinery (`MicBridge`, the drain fix, the capture-silence fix) carries
+  over unchanged — we just route the aligned samples to two channels instead of summing them.
+
+## Design
+
+### 1. Capture write path (`audio`)
+- Force the recording spec to **2ch, 16-bit, native rate** when the mic is enabled ("split"
+  layout). Loopback-only (mic off) keeps today's behavior.
+- New writer (replacing `write_wav_bytes`'s summing for split recordings): per frame,
+  **L = mic sample** (0 when the mic underran), **R = downmix(loopback channels)**.
+- Idle path (loopback silent): **L = drained mic, R = 0** — same silence-synthesis + device-paced
+  drain we just fixed, now writing to L instead of summing.
+
+### 2. Layout versioning (distinguish old "summed" from new "split")
+Content alone can't tell a summed-stereo file from a split-stereo one, so mark it explicitly:
+- Add `audio_layout TEXT` to the `meetings` row: `'summed'` (default / NULL for all existing rows
+  and imports) or `'split'` (new mic-enabled recordings).
+- Include it in the **`MeetingBundle` manifest** so exported recordings stay interpretable on
+  another machine.
+- Playback and diarization branch on it; old recordings keep every current code path untouched.
+
+### 3. Read helpers (`audio`)
+- `read_wav_mono_16k` — **unchanged** (downmix all channels → mono 16k). Used for transcription of
+  both layouts; on a split file it returns the summed signal, exactly as wanted.
+- New `read_wav_channel_16k(path, channel)` — extract one channel → mono 16k. Diarization reads the
+  **right** channel (far side); "You" detection reads the **left** (mic).
+
+### 4. Diarization / attribution (`commands`) — *simpler* than Phase 3
+For a **split** recording, one path replaces the timeline + masking:
+```
+far_spans = diarizer.diarize_samples( read_wav_channel_16k(wav, RIGHT) )   // clean far side
+you_spans = vad_spans( read_wav_channel_16k(wav, LEFT) )                    // energy-windowed mic
+merge + assign_by_overlap + build_name_map("You")                           // You + Speaker N
+```
+- The mic is **never** in the clustering input — structurally, not procedurally.
+- `vad_spans` is the `MicActivity` windowing logic (energy floor + 100 ms windows) applied to the L
+  channel from the file — so **stop and reprocess share one path** and produce identical results.
+- **This deletes `MicActivity`, `mic_activity.json` (persist + read + delete), and `mask_ranges`.**
+  Net: less code than the current Phase 3.
+- Mic-off (`'summed'`) recordings keep the whole-signal pass + voiceprint fallback.
+
+### 5. Playback (`commands::serve_recording`)
+- A split file plays mic-in-left / computer-in-right — wrong for a listener. On playback, **downmix
+  L+R → mono** (dual-mono) so both sources come out both ears.
+- `serve_recording` already reads+decrypts the whole file into memory; add a downmix-to-mono step
+  for `'split'` layout and serve that mono WAV (range/seek computed over the transcoded bytes).
+  `'summed'` files stream as today.
+
+### 6. Migration & portability
+- **Existing recordings** (`'summed'`): every current path unchanged. Their reprocess falls back to
+  the voiceprint match (acceptable — they predate split). We may retire the `MicActivity`/mask code
+  once split is default; the handful of recent test recordings lose Phase 3 on reprocess, which is
+  fine.
+- **Bundle export**: export **downmixes the canonical split `audio.wav` to mono** so a shared file
+  plays normally in any player; the working file stays split for WA's own re-processing.
+
+## What this removes / simplifies
+
+| Removed | Replaced by |
+|---|---|
+| `MicActivity` struct + capture wiring | L-channel VAD read from the file |
+| `mic_activity.json` (write/read/delete) | nothing — recompute from L channel |
+| `mask_ranges` + masked diarize | diarize the R channel directly |
+| stop vs reprocess divergence | one shared split-attribution path |
+
+The capture-silence fix, the device-paced `drain`, `diarize_samples`, and the `build_name_map`
+naming all **stay**.
+
+## Costs / risks (none fatal)
+- **New code:** split writer, `audio_layout` column + migration + manifest field, channel read
+  helper, playback downmix, `vad_spans`, and the split-vs-summed branch. Bigger than a patch, but
+  it *retires* comparable Phase 3 code.
+- **Loopback stereo image lost** (R = loopback downmixed to mono) — irrelevant for speech.
+- **Mono render endpoint** (rare): 1ch→2ch doubles that recording's size.
+- **Raw file plays split** in a third-party player until exported (export downmixes) — minor.
+- **Clock drift** between mic/loopback now shows as tiny L/R desync instead of dropped summed
+  samples — harmless (diarization tolerates it; playback sums it away).
+
+## Open questions for sign-off
+1. Layout marker: **DB column** (proposed) vs. a self-describing WAV chunk? DB column is simpler and
+   travels via the bundle manifest; recommend it.
+2. Retire `MicActivity`/`mask_ranges` outright, or keep them for existing `'summed'` recordings'
+   reprocess? Recommend **retire** (summed → voiceprint fallback) to avoid two live systems.
+3. Export downmix to mono — agree that's the right default for shared/synced copies?
+
+## Implementation order
+1. `audio_layout` column + `MeetingBundle` field + read on `get_meeting` (no behavior yet).
+2. Split writer + force 2ch when mic on; set `audio_layout='split'`. (`audio` + `start_recording`.)
+3. `read_wav_channel_16k` + `vad_spans`; split-attribution path in stop + reprocess.
+4. Playback downmix for split; bundle-export downmix.
+5. Delete `MicActivity` / `mic_activity.json` / `mask_ranges`; update tests.
+6. Apply the ADR-0005 amendment + update `docs/02`/`03`.
+
+---
+
+## Proposed ADR-0005 amendment (apply on approval)
+
+> ### Phase 3.5 refinement — dual-channel capture (FR-SPK/FR-CAP, 2026-07-14)
+>
+> The Phase 3 per-stream attribution (above) reconstructs the mic/far-side split *after the fact*
+> from a summed-mono `audio.wav` plus a `mic_activity.json` timeline, then masks the mic ranges
+> before clustering. This is reliable at stop but fragile on reprocess (re-aligning a sidecar
+> timeline against a mono mix), and it carries a parallel persistence path.
+>
+> **Decision:** when the microphone is enabled, record `audio.wav` as **stereo with the streams
+> separated — left = microphone ("You"), right = system/loopback ("Speaker")** — instead of summing
+> them. The separation is then intrinsic to the recording:
+> - Diarization runs on the **right channel only** → `Speaker N`; "You" comes from **left-channel**
+>   voice activity. The mic is never clustered, by construction.
+> - Reprocess recomputes both from the file — no sidecar timeline, no masking.
+> - Transcription still downmixes (L+R) to the same summed mono it uses today; playback and bundle
+>   export downmix to mono so shared/played audio is normal.
+> - A `meetings.audio_layout` flag (`summed` | `split`, in the bundle manifest for portability)
+>   distinguishes recordings; existing `summed` recordings keep the original blind-clustering +
+>   voiceprint path. On a stereo render endpoint this is **size-neutral** (the file was already 2ch
+>   with the mic summed into both).
+>
+> This **supersedes** `MicActivity` / `mic_activity.json` / masked diarization from Phase 3; the
+> capture-silence handling, `diarize_samples`, and the `build_name_map` naming are retained. Segment
+> IDs and the names-in-DB rule are unchanged.

package.json

@@ -1,7 +1,7 @@
 {
   "name": "whispassist",
   "private": true,
-  "version": "0.5.2",
+  "version": "0.7.3",
   "type": "module",
   "description": "Privacy-first, fully local Windows meeting assistant.",
   "license": "MIT OR Apache-2.0",

src-tauri/Cargo.lock

@@ -124,6 +124,17 @@ version = "1.1.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1505bd5d3d116872e7271a6d4e16d81d0c8570876c8de68093a09ac269d8aac0"
 
+[[package]]
+name = "auto-launch"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1f012b8cc0c850f34117ec8252a44418f2e34a2cf501de89e29b241ae5f79471"
+dependencies = [
+ "dirs 4.0.0",
+ "thiserror 1.0.69",
+ "winreg 0.10.1",
+]
+
 [[package]]
 name = "autocfg"
 version = "1.5.1"
@@ -848,6 +859,15 @@ dependencies = [
  "subtle",
 ]
 
+[[package]]
+name = "dirs"
+version = "4.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ca3aa72a6f96ea37bbc5aa912f6788242832f75369bdfdadcb0e38423f100059"
+dependencies = [
+ "dirs-sys 0.3.7",
+]
+
 [[package]]
 name = "dirs"
 version = "5.0.1"
@@ -866,6 +886,17 @@ dependencies = [
  "dirs-sys 0.5.0",
 ]
 
+[[package]]
+name = "dirs-sys"
+version = "0.3.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1b1d1d91c932ef41c0f2663aa8b0ca0342d444d842c06914aa0a7e352d0bada6"
+dependencies = [
+ "libc",
+ "redox_users 0.4.6",
+ "winapi",
+]
+
 [[package]]
 name = "dirs-sys"
 version = "0.4.1"
@@ -1043,7 +1074,7 @@ dependencies = [
  "rustc_version",
  "toml 1.1.2+spec-1.1.0",
  "vswhom",
- "winreg",
+ "winreg 0.55.0",
 ]
 
 [[package]]
@@ -4935,6 +4966,20 @@ dependencies = [
  "walkdir",
 ]
 
+[[package]]
+name = "tauri-plugin-autostart"
+version = "2.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "459383cebc193cdd03d1ba4acc40f2c408a7abce419d64bdcd2d745bc2886f70"
+dependencies = [
+ "auto-launch",
+ "serde",
+ "serde_json",
+ "tauri",
+ "tauri-plugin",
+ "thiserror 2.0.18",
+]
+
 [[package]]
 name = "tauri-plugin-dialog"
 version = "2.7.1"
@@ -6043,7 +6088,7 @@ dependencies = [
 
 [[package]]
 name = "whispassist"
-version = "0.5.2"
+version = "0.7.3"
 dependencies = [
  "argon2",
  "async-trait",
@@ -6073,6 +6118,7 @@ dependencies = [
  "sqlx",
  "tauri",
  "tauri-build",
+ "tauri-plugin-autostart",
  "tauri-plugin-dialog",
  "thiserror 1.0.69",
  "tokio",
@@ -6767,6 +6813,15 @@ dependencies = [
  "memchr",
 ]
 
+[[package]]
+name = "winreg"
+version = "0.10.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "80d0f4e272c85def139476380b12f9ac60926689dd2e01d4923222f40580869d"
+dependencies = [
+ "winapi",
+]
+
 [[package]]
 name = "winreg"
 version = "0.55.0"

src-tauri/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "whispassist"
-version = "0.5.2"
+version = "0.7.3"
 description = "Privacy-first, fully local Windows meeting assistant"
 authors = ["WhispAssist contributors"]
 license = "MIT OR Apache-2.0"
@@ -81,6 +81,7 @@ ort = { version = "=2.0.0-rc.10", optional = true, default-features = false, fea
 rustfft = { version = "6", optional = true }
 sherpa-rs = { version = "0.6", optional = true, default-features = false, features = ["download-binaries"] }  # sherpa-onnx bindings (Phase 4, ADR-0005)
 tauri-plugin-dialog = "2"                                   # native Save/choose-folder (Phase 2 export)
+tauri-plugin-autostart = "2"                                # opt-in launch-on-login (per-user HKCU\Run, no admin; NFR-RES-4)
 
 # notes export (Phase 8, FR-NOTE-4) — pure-Rust, no external binary/cloud
 # conversion service, consistent with the fully-local invariant.

src-tauri/gen/schemas/desktop-schema.json

@@ -176,6 +176,48 @@
     "Identifier": {
       "description": "Permission identifier",
       "oneOf": [
+        {
+          "description": "This permission set configures if your\napplication can enable or disable auto\nstarting the application on boot.\n\n#### Granted Permissions\n\nIt allows all to check, enable and\ndisable the automatic start on boot.\n\n\n#### This default permission set includes:\n\n- `allow-enable`\n- `allow-disable`\n- `allow-is-enabled`",
+          "type": "string",
+          "const": "autostart:default",
+          "markdownDescription": "This permission set configures if your\napplication can enable or disable auto\nstarting the application on boot.\n\n#### Granted Permissions\n\nIt allows all to check, enable and\ndisable the automatic start on boot.\n\n\n#### This default permission set includes:\n\n- `allow-enable`\n- `allow-disable`\n- `allow-is-enabled`"
+        },
+        {
+          "description": "Enables the disable command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:allow-disable",
+          "markdownDescription": "Enables the disable command without any pre-configured scope."
+        },
+        {
+          "description": "Enables the enable command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:allow-enable",
+          "markdownDescription": "Enables the enable command without any pre-configured scope."
+        },
+        {
+          "description": "Enables the is_enabled command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:allow-is-enabled",
+          "markdownDescription": "Enables the is_enabled command without any pre-configured scope."
+        },
+        {
+          "description": "Denies the disable command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:deny-disable",
+          "markdownDescription": "Denies the disable command without any pre-configured scope."
+        },
+        {
+          "description": "Denies the enable command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:deny-enable",
+          "markdownDescription": "Denies the enable command without any pre-configured scope."
+        },
+        {
+          "description": "Denies the is_enabled command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:deny-is-enabled",
+          "markdownDescription": "Denies the is_enabled command without any pre-configured scope."
+        },
         {
           "description": "Default core plugins set.\n#### This default permission set includes:\n\n- `core:path:default`\n- `core:event:default`\n- `core:window:default`\n- `core:webview:default`\n- `core:app:default`\n- `core:image:default`\n- `core:resources:default`\n- `core:menu:default`\n- `core:tray:default`",
           "type": "string",

src-tauri/gen/schemas/windows-schema.json

@@ -176,6 +176,48 @@
     "Identifier": {
       "description": "Permission identifier",
       "oneOf": [
+        {
+          "description": "This permission set configures if your\napplication can enable or disable auto\nstarting the application on boot.\n\n#### Granted Permissions\n\nIt allows all to check, enable and\ndisable the automatic start on boot.\n\n\n#### This default permission set includes:\n\n- `allow-enable`\n- `allow-disable`\n- `allow-is-enabled`",
+          "type": "string",
+          "const": "autostart:default",
+          "markdownDescription": "This permission set configures if your\napplication can enable or disable auto\nstarting the application on boot.\n\n#### Granted Permissions\n\nIt allows all to check, enable and\ndisable the automatic start on boot.\n\n\n#### This default permission set includes:\n\n- `allow-enable`\n- `allow-disable`\n- `allow-is-enabled`"
+        },
+        {
+          "description": "Enables the disable command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:allow-disable",
+          "markdownDescription": "Enables the disable command without any pre-configured scope."
+        },
+        {
+          "description": "Enables the enable command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:allow-enable",
+          "markdownDescription": "Enables the enable command without any pre-configured scope."
+        },
+        {
+          "description": "Enables the is_enabled command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:allow-is-enabled",
+          "markdownDescription": "Enables the is_enabled command without any pre-configured scope."
+        },
+        {
+          "description": "Denies the disable command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:deny-disable",
+          "markdownDescription": "Denies the disable command without any pre-configured scope."
+        },
+        {
+          "description": "Denies the enable command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:deny-enable",
+          "markdownDescription": "Denies the enable command without any pre-configured scope."
+        },
+        {
+          "description": "Denies the is_enabled command without any pre-configured scope.",
+          "type": "string",
+          "const": "autostart:deny-is-enabled",
+          "markdownDescription": "Denies the is_enabled command without any pre-configured scope."
+        },
         {
           "description": "Default core plugins set.\n#### This default permission set includes:\n\n- `core:path:default`\n- `core:event:default`\n- `core:window:default`\n- `core:webview:default`\n- `core:app:default`\n- `core:image:default`\n- `core:resources:default`\n- `core:menu:default`\n- `core:tray:default`",
           "type": "string",

src-tauri/migrations/0009_audio_layout.sql

@@ -0,0 +1,12 @@
+-- WhispAssist dual-channel capture (FR-SPK/FR-CAP). Forward-only migration.
+-- Mirrors docs/03-data-model.md.
+PRAGMA foreign_keys = ON;
+
+-- How audio.wav stores its channels:
+--   NULL / 'summed' — mic summed into every loopback channel (all pre-2026-07-14
+--                     recordings and every import): one mixed mono-equivalent.
+--   'split'         — stereo, left = microphone ("You"), right = system/loopback
+--                     ("Speaker"), kept separate so diarization runs on the far
+--                     side alone. New mic-enabled recordings.
+-- Diarization, playback (downmix), and bundle export branch on this.
+ALTER TABLE meetings ADD COLUMN audio_layout TEXT;

src-tauri/src/audio/mod.rs

@@ -45,9 +45,26 @@ pub enum AudioError {
 pub struct CaptureHandle {
     running: Arc<AtomicBool>,
     paused: Arc<AtomicBool>,
+    /// Mic-only (FR-CAP-7): when set, the microphone stream emits silence instead
+    /// of real samples — the recording's mic-left channel and the live transcript
+    /// go quiet, the meter drops to zero, while loopback keeps recording. Toggled
+    /// live via `set_muted` (the "press M to mute" control).
+    muted: Arc<AtomicBool>,
     thread: JoinHandle<Result<CaptureSummary, AudioError>>,
 }
 
+impl CaptureHandle {
+    /// Mute/unmute this stream live. Only meaningful for the microphone capture.
+    pub fn set_muted(&self, muted: bool) {
+        self.muted.store(muted, Ordering::SeqCst);
+    }
+
+    /// Whether this stream is currently muted.
+    pub fn is_muted(&self) -> bool {
+        self.muted.load(Ordering::SeqCst)
+    }
+}
+
 /// Where captured frames are delivered for live transcription: mono f32 @ 16kHz,
 /// bounded so a slow/absent consumer can never stall the capture thread.
 pub type FrameSink = SyncSender<Vec<f32>>;
@@ -201,16 +218,37 @@ impl MicBridge {
             }
         }
     }
+
+    /// Loopback idle path: take everything buffered at once, device-paced. The
+    /// idle writer must NOT meter the mic by wall-clock — that lets the buffer
+    /// back up to `cap` and drop the oldest samples, which sounds like the audio
+    /// skipping/fast-forwarding late in a long silent stretch. Draining fully
+    /// keeps the mic flowing at its own clock.
+    fn drain(&self) -> Vec<f32> {
+        self.buf
+            .lock()
+            .map(|mut buf| buf.drain(..).collect())
+            .unwrap_or_default()
+    }
 }
 
+/// Below this per-chunk RMS the mic is treated as silence/room noise and not
+/// collected into a voiceprint. Without it the voiceprint was just the first
+/// few seconds of the recording regardless of whether the user had spoken —
+/// so a user silent for the first 30s got a noise voiceprint that spuriously
+/// matched a cluster and mislabeled everyone "You". ponytail: fixed floor for
+/// normalized [-1,1] speech; promote to a calibration knob if real mics vary.
+const VOICE_ENERGY_FLOOR: f32 = 0.01;
+
 /// A one-shot, bounded capture of raw mic-only audio (16kHz mono, same format
-/// the transcriber and diarizer both use) taken early in a recording — enough
+/// the transcriber and diarizer both use) taken from the user's early *speech*
-/// to compute a voiceprint that identifies which diarized speaker cluster is
+/// — enough to compute a voiceprint that identifies which diarized speaker
-/// the mic (so it can be labeled "You" instead of a clustered "S1"/"S2"; see
+/// cluster is the mic (so it can be labeled "You" instead of a clustered
-/// `diarization::voiceprint`). Unlike `MicBridge`, this is filled once and
+/// "S1"/"S2"; see `diarization::voiceprint`). Unlike `MicBridge`, this is filled
-/// never drained: the first `cap` samples are kept and everything after is
+/// once and never drained: the first `cap` samples of speech-level audio are
-/// dropped, since a voiceprint only needs a few seconds of real speech, not
+/// kept and everything after is dropped, since a voiceprint only needs a few
-/// the whole meeting.
+/// seconds of real speech, not the whole meeting. Silence is skipped so the
+/// window tracks when the user actually talks, not the wall clock.
 pub struct VoiceSample {
     cap: usize,
     buf: Mutex<Vec<f32>>,
@@ -225,6 +263,18 @@ impl VoiceSample {
     }
 
     fn push(&self, samples: &[f32]) {
+        // Skip silence/room noise so the voiceprint is the user's voice, not
+        // whatever the mic picked up before they first spoke (see
+        // VOICE_ENERGY_FLOOR). RMS inline so VoiceSample stays independent of
+        // the `audio`-gated level helper.
+        let rms = if samples.is_empty() {
+            0.0
+        } else {
+            (samples.iter().map(|s| s * s).sum::<f32>() / samples.len() as f32).sqrt()
+        };
+        if rms < VOICE_ENERGY_FLOOR {
+            return;
+        }
         if let Ok(mut buf) = self.buf.lock() {
             if buf.len() < self.cap {
                 buf.extend_from_slice(samples);
@@ -240,6 +290,12 @@ impl VoiceSample {
     }
 }
 
+// MicActivity (the live mic-speech timeline) was retired 2026-07-14 with the
+// masked-mono diarization it fed — the split (mic-left/loopback-right) layout
+// makes the mic/far-side separation intrinsic to audio.wav, so "You" is
+// recomputed from the left channel via `vad_spans`. See the retirement note in
+// commands.rs (attribute_split).
+
 /// Number of audio frames in a raw WASAPI byte buffer of the given format.
 #[cfg(feature = "audio")]
 fn frame_count(bytes: &[u8], format: &WaveFormat) -> usize {
@@ -273,12 +329,15 @@ impl WasapiCapture {
         event_sink: EventSink,
         bridge: Option<Arc<MicBridge>>,
         voice_sample: Option<Arc<VoiceSample>>,
+        split: bool,
         emit_level: bool,
     ) -> Result<CaptureHandle, AudioError> {
         let running = Arc::new(AtomicBool::new(true));
         let paused = Arc::new(AtomicBool::new(false));
+        let muted = Arc::new(AtomicBool::new(false));
         let running_th = running.clone();
         let paused_th = paused.clone();
+        let muted_th = muted.clone();
         let wav_path = wav_path.map(Path::to_path_buf);
         let device_id = device_id.map(str::to_string);
 
@@ -293,8 +352,10 @@ impl WasapiCapture {
                     &event_sink,
                     &running_th,
                     &paused_th,
+                    &muted_th,
                     bridge.as_ref(),
                     voice_sample.as_ref(),
+                    split,
                     emit_level,
                 )
             })
@@ -303,6 +364,7 @@ impl WasapiCapture {
         Ok(CaptureHandle {
             running,
             paused,
+            muted,
             thread,
         })
     }
@@ -327,7 +389,8 @@ impl WasapiCapture {
             event_sink,
             Some(bridge),
             None,
-            true,
+            true, // split: mic-enabled loopback records mic-left / loopback-right
+            true, // emit_level
         )
     }
 
@@ -352,7 +415,8 @@ impl WasapiCapture {
             event_sink,
             Some(bridge),
             voice_sample,
-            true,
+            false, // split: mic stream has no writer
+            true,  // emit_level
         )
     }
 }
@@ -375,7 +439,8 @@ impl AudioCapture for WasapiCapture {
             event_sink,
             None,
             None,
-            true,
+            false, // split: loopback-only / mic stream — no split writer
+            true,  // emit_level
         )
     }
 
@@ -394,7 +459,8 @@ impl AudioCapture for WasapiCapture {
             event_sink,
             None,
             None,
-            true,
+            false, // split: loopback-only / mic stream — no split writer
+            true,  // emit_level
         )
     }
 
@@ -540,6 +606,18 @@ fn audio_level(mono: &[f32], mic: bool) -> AudioLevel {
 /// without flooding the frontend with an event per ~10ms WASAPI callback.
 const LEVEL_EMIT_INTERVAL: Duration = Duration::from_millis(50);
 
+/// When the loopback (render) endpoint is idle, WASAPI delivers no packets, so
+/// the recording would stall and the bridged mic — the user talking over system
+/// silence — would be lost as the MicBridge overflows. The loopback loop then
+/// synthesizes silence for the elapsed gap and folds the mic in, keeping
+/// `audio.wav` continuous and complete (FR-CAP-7). MIN gates on a *sustained*
+/// gap so brief inter-packet stalls during active playback don't punch silence
+/// into the loopback (audible clicks); it stays well under the MicBridge's
+/// ~0.5s cap so the mic can't back up and drop samples between drains. MAX caps
+/// the silence padding written in one chunk.
+const SILENCE_TICK_MIN: Duration = Duration::from_millis(150);
+const SILENCE_TICK_MAX: Duration = Duration::from_millis(500);
+
 /// Runs on a dedicated OS thread for the lifetime of a `CaptureHandle`. Owns the
 /// WASAPI client and the WAV writer; exits (and finalizes the WAV) once `running`
 /// is cleared.
@@ -553,8 +631,15 @@ fn capture_loop(
     event_sink: &EventSink,
     running: &AtomicBool,
     paused: &AtomicBool,
+    // Mic-only live mute (FR-CAP-7): zeroes the decoded mic samples so the
+    // recording, transcript, and meter all go silent while loopback continues.
+    muted: &AtomicBool,
     bridge: Option<&Arc<MicBridge>>,
     voice_sample: Option<&Arc<VoiceSample>>,
+    // FR-SPK: when true, the loopback WAV is stereo L=mic / R=loopback (the mic
+    // is a separate track, not summed in) so diarization can run on the far side
+    // alone. Only meaningful for the loopback writer.
+    split: bool,
     emit_level: bool,
 ) -> Result<CaptureSummary, AudioError> {
     wasapi::initialize_mta()
@@ -566,14 +651,18 @@ fn capture_loop(
     // UI can overlay them — but only the loopback stream writes a WAV.
     let mut session = open_capture_session(&direction, device_id)?;
     // Loopback publishes its rate so the mic knows what to resample to before
-    // pushing into the shared bridge (mic-into-recording, FR-CAP-7).
+    // pushing into the shared bridge (mic-into-recording, FR-CAP-7), and so the
+    // Phase 3 mic-activity timeline maps frame positions to windows.
     if is_loopback {
         if let Some(b) = bridge {
             b.rate
                 .store(session.format.get_samplespersec(), Ordering::Relaxed);
         }
     }
-    let spec = wav_spec_for(&session.format)?;
+    let mut spec = wav_spec_for(&session.format)?;
+    if split && is_loopback {
+        spec.channels = 2; // L = mic, R = loopback (downmixed)
+    }
     let mut writer = match wav_path {
         Some(path) => Some(WavWriter::create(path, spec).map_err(|e| {
             AudioError::Capture(format!("could not create {}: {e}", path.display()))
@@ -593,6 +682,9 @@ fn capture_loop(
     let mut queue: std::collections::VecDeque<u8> = std::collections::VecDeque::new();
     let mut frames_written: u64 = 0;
     let mut last_level_emit = Instant::now() - LEVEL_EMIT_INTERVAL;
+    // Wall-clock of the last WAV write (real or synthesized), so the loopback
+    // loop can fill idle gaps with silence+mic instead of stalling.
+    let mut last_write = Instant::now();
 
     while running.load(Ordering::Relaxed) {
         // Short timeout so we periodically re-check `running` even with no data.
@@ -636,8 +728,42 @@ fn capture_loop(
             continue;
         }
         if queue.is_empty() {
+            // Loopback idle (nothing playing): keep audio.wav advancing so the
+            // bridged mic (user talking over system silence) is preserved
+            // instead of lost to MicBridge overflow. Only after a *sustained*
+            // gap, so brief inter-packet stalls during active playback don't
+            // punch silence into the loopback (clicks). Loopback + mic-enabled
+            // only; the mic capture stream has no writer.
+            if is_loopback && !paused.load(Ordering::Relaxed) {
+                if let (Some(w), Some(b)) = (writer.as_mut(), bridge) {
+                    let elapsed = last_write.elapsed();
+                    if elapsed >= SILENCE_TICK_MIN {
+                        // Drain the whole mic buffer (device-paced). Metering the
+                        // pull by wall-clock let the buffer back up to its cap and
+                        // drop the oldest samples — the audio skipped/fast-
+                        // forwarded late in a long gap. Draining fully avoids that.
+                        let mic = b.drain();
+                        let rate = session.format.get_samplespersec() as usize;
+                        // Advance the timeline by the wall-clock gap (capped), but
+                        // never fewer frames than the mic we must write.
+                        let gap_frames =
+                            (elapsed.min(SILENCE_TICK_MAX).as_micros() as usize * rate / 1_000_000)
+                                .max(mic.len());
+                        let bpf = session.format.get_nchannels() as usize
+                            * (session.format.get_bitspersample() as usize / 8);
+                        let silence = vec![0u8; gap_frames * bpf];
+                        frames_written += if split {
+                            write_wav_split(w, &silence, &session.format, &mic)?
+                        } else {
+                            write_wav_bytes(w, &silence, &session.format, &mic)?
+                        };
+                        last_write = Instant::now();
+                    }
+                }
+            }
             continue;
         }
+        last_write = Instant::now();
         let bytes: Vec<u8> = queue.drain(..).collect();
 
         // Must keep pulling WASAPI buffers even while paused (required to avoid
@@ -653,9 +779,20 @@ fn capture_loop(
                 Some(b) if is_loopback => b.pull(frame_count(&bytes, &session.format)),
                 _ => Vec::new(),
             };
-            frames_written += write_wav_bytes(w, &bytes, &session.format, &mic)?;
+            frames_written += if split {
+                write_wav_split(w, &bytes, &session.format, &mic)?
+            } else {
+                write_wav_bytes(w, &bytes, &session.format, &mic)?
+            };
+        }
+        let mut mono = decode_mono_f32(&bytes, &session.format)?;
+        // Mic muted: replace the decoded samples with silence before anything
+        // downstream sees them — the recording's mic channel, the bridge, the
+        // transcript feed, the meter, and the voiceprint sample all go quiet.
+        // Loopback (`is_loopback`) is never muted this way.
+        if !is_loopback && muted.load(Ordering::Relaxed) {
+            mono.iter_mut().for_each(|s| *s = 0.0);
         }
-        let mono = decode_mono_f32(&bytes, &session.format)?;
 
         // Mic: feed the shared bridge (resampled to the loopback's rate) so the
         // loopback thread can fold it into the recording.
@@ -783,6 +920,57 @@ fn f32_to_i16(v: f32) -> i16 {
     (v.clamp(-1.0, 1.0) * i16::MAX as f32) as i16
 }
 
+/// Split layout (FR-SPK): write frames as **stereo, L = mic, R = loopback**
+/// (downmixed to mono), so the two sources stay separate in `audio.wav` and
+/// diarization can run on the far side alone. `loopback_bytes` is the raw WASAPI
+/// mix buffer (all-zero for a synthesized-silence chunk); `mic` may be shorter
+/// than the frame count (rest silent) or empty (mic underran). Returns frames.
+fn write_wav_split(
+    writer: &mut WavWriter<BufWriter<File>>,
+    loopback_bytes: &[u8],
+    format: &WaveFormat,
+    mic: &[f32],
+) -> Result<u64, AudioError> {
+    let sample_type = format
+        .get_subformat()
+        .map_err(|e| AudioError::Device(format!("unrecognized mix format: {e}")))?;
+    let channels = format.get_nchannels() as usize;
+    if channels == 0 {
+        return Ok(0);
+    }
+    let (bytes_per_sample, decode): (usize, fn(&[u8]) -> f32) =
+        match (sample_type, format.get_bitspersample()) {
+            (SampleType::Float, 32) => (4, |c| f32::from_le_bytes(c.try_into().unwrap())),
+            (SampleType::Int, 16) => (2, |c| {
+                i16::from_le_bytes(c.try_into().unwrap()) as f32 / i16::MAX as f32
+            }),
+            (st, bits) => {
+                return Err(AudioError::Device(format!(
+                    "unsupported capture format: {st} {bits}-bit"
+                )))
+            }
+        };
+    let mut frames = 0u64;
+    for frame in loopback_bytes.chunks_exact(bytes_per_sample * channels) {
+        // R = loopback downmixed to mono.
+        let loop_mono = frame
+            .chunks_exact(bytes_per_sample)
+            .map(decode)
+            .sum::<f32>()
+            / channels as f32;
+        // L = mic for this frame (silent if the mic underran).
+        let mic_s = mic.get(frames as usize).copied().unwrap_or(0.0);
+        writer
+            .write_sample(f32_to_i16(mic_s))
+            .map_err(|e| AudioError::Capture(format!("wav write: {e}")))?;
+        writer
+            .write_sample(f32_to_i16(loop_mono))
+            .map_err(|e| AudioError::Capture(format!("wav write: {e}")))?;
+        frames += 1;
+    }
+    Ok(frames)
+}
+
 /// Downmix raw WASAPI capture bytes to mono `f32` in `[-1.0, 1.0]`, at the
 /// device's native sample rate (resampling to 16kHz happens separately).
 fn decode_mono_f32(bytes: &[u8], format: &WaveFormat) -> Result<Vec<f32>, AudioError> {
@@ -897,6 +1085,22 @@ impl Resampler {
 /// through hound's size-aware iterator; hound is only used to parse the
 /// format/`fmt ` chunk.
 pub fn read_wav_mono_16k(path: &Path) -> Result<Vec<f32>, AudioError> {
+    read_wav_16k(path, None)
+}
+
+/// Like [`read_wav_mono_16k`] but extracts a single channel instead of
+/// downmixing — used by the Phase 3.5 split path to diarize the right
+/// (loopback) channel and VAD the left (mic) channel (FR-SPK). `channel` is
+/// clamped to the file's last channel, so it's safe on a mono/summed file.
+pub fn read_wav_channel_16k(path: &Path, channel: usize) -> Result<Vec<f32>, AudioError> {
+    read_wav_16k(path, Some(channel))
+}
+
+/// Shared reader: decrypt (T8.8), reduce each frame to one f32 — averaging all
+/// channels (`channel = None`) or picking one (`channel = Some`) — resample to
+/// 16kHz. Reads the `data` payload as raw bytes (an unfinalized WAV's size
+/// fields read back as 0), so it works on a mid-write / crash-left file (FR-REL-1).
+fn read_wav_16k(path: &Path, channel: Option<usize>) -> Result<Vec<f32>, AudioError> {
     let raw = std::fs::read(path).map_err(|e| AudioError::Capture(format!("read wav: {e}")))?;
     // Decrypt if the recording was sealed at rest (T8.8); plaintext passes through.
     let file =
@@ -906,6 +1110,8 @@ pub fn read_wav_mono_16k(path: &Path) -> Result<Vec<f32>, AudioError> {
         .map_err(|e| AudioError::Capture(format!("open wav: {e}")))?
         .spec();
     let channels = spec.channels.max(1) as usize;
+    // Which channel index to pick (clamped), or None to downmix.
+    let pick = channel.map(|c| c.min(channels - 1));
 
     let data_marker = file
         .windows(4)
@@ -918,21 +1124,17 @@ pub fn read_wav_mono_16k(path: &Path) -> Result<Vec<f32>, AudioError> {
         (SampleFormat::Float, 32) => bytes
             .chunks_exact(4 * channels)
             .map(|frame| {
-                frame
+                reduce_frame(frame, 4, pick, channels, |c| {
-                    .chunks_exact(4)
+                    f32::from_le_bytes(c.try_into().unwrap())
-                    .map(|c| f32::from_le_bytes(c.try_into().unwrap()))
+                })
-                    .sum::<f32>()
-                    / channels as f32
             })
             .collect(),
         (SampleFormat::Int, 16) => bytes
             .chunks_exact(2 * channels)
             .map(|frame| {
-                frame
+                reduce_frame(frame, 2, pick, channels, |c| {
-                    .chunks_exact(2)
+                    i16::from_le_bytes(c.try_into().unwrap()) as f32 / i16::MAX as f32
-                    .map(|c| i16::from_le_bytes(c.try_into().unwrap()) as f32 / i16::MAX as f32)
+                })
-                    .sum::<f32>()
-                    / channels as f32
             })
             .collect(),
         (fmt, bits) => {
@@ -946,6 +1148,78 @@ pub fn read_wav_mono_16k(path: &Path) -> Result<Vec<f32>, AudioError> {
     Ok(resampler.process(&mono))
 }
 
+/// Voice-activity "You" spans (FR-SPK): 100ms windows of 16kHz mono audio above
+/// [`VOICE_ENERGY_FLOOR`], merged into `(start_ms, end_ms)` ranges. Run on the
+/// left (mic) channel of a split recording to attribute segments to "You" —
+/// replaces the old live `MicActivity` timeline (recomputed from the file, so
+/// stop and reprocess agree).
+pub fn vad_spans(samples: &[f32]) -> Vec<(u64, u64)> {
+    const WINDOW_MS: u64 = 100;
+    let win = (TARGET_SAMPLE_RATE as u64 * WINDOW_MS / 1000) as usize; // 1600 @16kHz
+    if win == 0 {
+        return Vec::new();
+    }
+    let mut spans = Vec::new();
+    let mut start: Option<u64> = None;
+    let mut w = 0u64;
+    let mut i = 0usize;
+    while i < samples.len() {
+        let chunk = &samples[i..(i + win).min(samples.len())];
+        let rms = (chunk.iter().map(|s| s * s).sum::<f32>() / chunk.len().max(1) as f32).sqrt();
+        match (rms >= VOICE_ENERGY_FLOOR, start) {
+            (true, None) => start = Some(w * WINDOW_MS),
+            (false, Some(s)) => {
+                spans.push((s, w * WINDOW_MS));
+                start = None;
+            }
+            _ => {}
+        }
+        i += win;
+        w += 1;
+    }
+    if let Some(s) = start {
+        spans.push((s, w * WINDOW_MS));
+    }
+    spans
+}
+
+/// Reduce one interleaved frame to a single f32: pick channel `pick` or average
+/// all `channels`. `width` is bytes-per-sample; `decode` turns those bytes → f32.
+fn reduce_frame(
+    frame: &[u8],
+    width: usize,
+    pick: Option<usize>,
+    channels: usize,
+    decode: impl Fn(&[u8]) -> f32,
+) -> f32 {
+    match pick {
+        Some(c) => decode(&frame[c * width..c * width + width]),
+        None => frame.chunks_exact(width).map(decode).sum::<f32>() / channels as f32,
+    }
+}
+
+/// Write mono 16kHz `f32` samples as a Float32 WAV — the format
+/// [`read_wav_mono_16k`] reads back. Used to persist the mic voiceprint sample
+/// next to a retained recording so it can be re-matched on reprocess (FR-SPK).
+pub fn write_wav_mono_16k(path: &Path, samples: &[f32]) -> Result<(), AudioError> {
+    let spec = WavSpec {
+        channels: 1,
+        sample_rate: 16_000,
+        bits_per_sample: 32,
+        sample_format: SampleFormat::Float,
+    };
+    let mut writer = WavWriter::create(path, spec)
+        .map_err(|e| AudioError::Capture(format!("write wav: {e}")))?;
+    for &s in samples {
+        writer
+            .write_sample(s)
+            .map_err(|e| AudioError::Capture(format!("write wav: {e}")))?;
+    }
+    writer
+        .finalize()
+        .map_err(|e| AudioError::Capture(format!("write wav: {e}")))
+}
+
 /// Ceiling on how far the still-flowing stream may run ahead of a stalled
 /// partner before the mixer forwards it alone — so a muted/dead microphone (or a
 /// silent system output) can't hold up the live transcript. 0.5s @ 16kHz.
@@ -1051,6 +1325,34 @@ fn mixer_loop(
 mod tests {
     use super::*;
 
+    #[test]
+    fn vad_spans_ignores_silence() {
+        // 200ms of 16kHz silence → no "You" spans.
+        assert!(vad_spans(&[0.0; 3_200]).is_empty());
+    }
+
+    #[test]
+    fn vad_spans_collapses_speech_into_spans_with_gaps() {
+        // 16kHz: 1600 samples per 100ms window. Speech in windows 0,1 (0-200ms)
+        // and 4,5 (400-600ms), silence between.
+        let mut samples = vec![0.0f32; 9_600]; // 600ms
+        samples[0..3_200].fill(0.2); // windows 0,1
+        samples[6_400..9_600].fill(0.2); // windows 4,5
+        assert_eq!(vad_spans(&samples), vec![(0, 200), (400, 600)]);
+    }
+
+    #[test]
+    fn voice_sample_skips_silence_and_keeps_speech() {
+        let vs = VoiceSample::new(16_000);
+        vs.push(&[0.0; 4_000]); // silence: below the energy floor
+        assert!(
+            vs.samples().is_empty(),
+            "silence must not fill the voiceprint"
+        );
+        vs.push(&[0.2; 4_000]); // speech-level audio
+        assert_eq!(vs.samples().len(), 4_000, "speech-level audio is collected");
+    }
+
     #[test]
     fn mixer_sums_aligned_streams_and_keeps_the_unmatched_tail() {
         let mut m = Mixer::new();
@@ -1127,6 +1429,33 @@ mod tests {
         let _ = std::fs::remove_dir_all(&dir);
     }
 
+    #[test]
+    fn write_wav_split_puts_mic_left_and_loopback_right() {
+        let dir = std::env::temp_dir().join(format!("wa-test-{}", uuid::Uuid::new_v4()));
+        std::fs::create_dir_all(&dir).unwrap();
+        let path = dir.join("audio.wav");
+        // 2ch float loopback input; split output is always 2ch 16-bit.
+        let format = WaveFormat::new(32, 32, &SampleType::Float, 48_000, 2, None);
+        let spec = WavSpec {
+            channels: 2,
+            sample_rate: 48_000,
+            bits_per_sample: 16,
+            sample_format: SampleFormat::Int,
+        };
+        let mut writer = WavWriter::create(&path, spec).unwrap();
+        // One 2ch loopback frame [0.4, 0.6] → R = downmix 0.5; mic 0.2 → L.
+        let loop_bytes: Vec<u8> = [0.4f32, 0.6].iter().flat_map(|s| s.to_le_bytes()).collect();
+        let n = write_wav_split(&mut writer, &loop_bytes, &format, &[0.2]).unwrap();
+        writer.finalize().unwrap();
+        assert_eq!(n, 1);
+
+        let mut reader = hound::WavReader::open(&path).unwrap();
+        let s: Vec<i16> = reader.samples::<i16>().map(|x| x.unwrap()).collect();
+        assert!((s[0] as f32 / i16::MAX as f32 - 0.2).abs() < 0.001); // L = mic
+        assert!((s[1] as f32 / i16::MAX as f32 - 0.5).abs() < 0.001); // R = loopback downmix
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+
     #[test]
     fn f32_to_i16_clamps_and_scales() {
         assert_eq!(f32_to_i16(0.0), 0);

src-tauri/src/deploy.rs

@@ -0,0 +1,183 @@
+//! Enterprise deployment: seed default settings from an admin-supplied `.ini`
+//! on **first run only** (before any `settings.json` exists).
+//!
+//! An admin mass-deploying WhispAssist (GPO / SCCM / Intune) drops a
+//! `wa-defaults.ini` and every fresh install picks it up once, seeding
+//! `settings.json` with their chosen defaults (record-by-default, preferred
+//! backend, retention, model to auto-download, …) — all via native Windows file
+//! deployment, no WiX custom actions. See `docs/enterprise-deployment.md`.
+//!
+//! **Guardrail (CLAUDE.md):** the file must never carry secrets. Keys that look
+//! like credentials are ignored here as defense in depth — API keys / OAuth
+//! tokens live only in the OS credential store.
+
+use crate::models::Settings;
+use serde_json::{Map, Value};
+use std::path::PathBuf;
+
+/// Special (non-`Settings`) INI key: when truthy, the first-run seed also fetches
+/// the configured `whisper_model` in the background so the machine is ready
+/// offline. Stripped before the settings merge.
+const AUTO_DOWNLOAD_KEY: &str = "auto_download_model";
+
+/// Candidate locations, first found wins:
+/// 1. `%PROGRAMDATA%\WhispAssist\wa-defaults.ini` — machine-wide enterprise path.
+/// 2. `<exe dir>\wa-defaults.ini` — the bundled template / per-install override.
+fn candidate_paths() -> Vec<PathBuf> {
+    let mut paths = Vec::new();
+    if let Ok(program_data) = std::env::var("ProgramData") {
+        paths.push(
+            PathBuf::from(program_data)
+                .join("WhispAssist")
+                .join("wa-defaults.ini"),
+        );
+    }
+    if let Ok(exe) = std::env::current_exe() {
+        if let Some(dir) = exe.parent() {
+            paths.push(dir.join("wa-defaults.ini"));
+        }
+    }
+    paths
+}
+
+/// Reads the first existing defaults file and produces the seeded settings plus
+/// the whisper model id to auto-download (if `auto_download_model` was set).
+/// `None` when no file exists or it contains no overrides (the shipped template
+/// is fully commented, so normal installs get exactly today's behavior).
+pub fn seed_settings_from_defaults() -> Option<(Settings, Option<String>)> {
+    let text = candidate_paths()
+        .into_iter()
+        .find_map(|p| std::fs::read_to_string(p).ok())?;
+    seed_from_ini(&text)
+}
+
+/// Testable core: parse INI text → merge onto the built-in defaults.
+fn seed_from_ini(text: &str) -> Option<(Settings, Option<String>)> {
+    let mut overrides = parse_ini(text);
+    if overrides.is_empty() {
+        return None;
+    }
+
+    // Pull the non-Settings auto-download flag out before the merge.
+    let auto_download = overrides
+        .remove(AUTO_DOWNLOAD_KEY)
+        .map(|v| truthy(&v))
+        .unwrap_or(false);
+
+    // Merge overrides onto the default settings' JSON form, then deserialize.
+    // Unknown keys (typos) are ignored — `Settings` has no deny_unknown_fields.
+    let mut base = match serde_json::to_value(crate::commands::default_settings()) {
+        Ok(Value::Object(map)) => map,
+        _ => return None,
+    };
+    for (k, v) in overrides {
+        base.insert(k, v);
+    }
+
+    let settings: Settings = serde_json::from_value(Value::Object(base)).ok()?;
+    let model = if auto_download {
+        Some(settings.whisper_model.clone())
+    } else {
+        None
+    };
+    Some((settings, model))
+}
+
+/// Minimal INI reader: skips blanks, `;`/`#` comments and `[section]` headers;
+/// splits each `key = value` on the first `=`; coerces values to bool / integer /
+/// string so serde lands them on the typed `Settings` fields. Silently drops any
+/// key that looks like a secret (guardrail — no credentials in the deploy file).
+fn parse_ini(text: &str) -> Map<String, Value> {
+    let mut map = Map::new();
+    for line in text.lines() {
+        let line = line.trim();
+        if line.is_empty()
+            || line.starts_with(';')
+            || line.starts_with('#')
+            || line.starts_with('[')
+        {
+            continue;
+        }
+        let Some((key, value)) = line.split_once('=') else {
+            continue;
+        };
+        let key = key.trim().to_string();
+        let value = value.trim();
+        if key.is_empty() || looks_like_secret(&key) {
+            continue;
+        }
+        map.insert(key, coerce(value));
+    }
+    map
+}
+
+/// `true`/`false` → bool, all-integer → number, everything else → string.
+fn coerce(value: &str) -> Value {
+    match value.to_ascii_lowercase().as_str() {
+        "true" => return Value::Bool(true),
+        "false" => return Value::Bool(false),
+        _ => {}
+    }
+    if let Ok(n) = value.parse::<i64>() {
+        return Value::Number(n.into());
+    }
+    Value::String(value.to_string())
+}
+
+fn truthy(v: &Value) -> bool {
+    matches!(v, Value::Bool(true))
+        || matches!(v, Value::String(s) if s.eq_ignore_ascii_case("true"))
+}
+
+/// Defense in depth: never seed anything that smells like a credential.
+fn looks_like_secret(key: &str) -> bool {
+    let k = key.to_ascii_lowercase();
+    ["key", "token", "secret", "credential", "password"]
+        .iter()
+        .any(|needle| k.contains(needle))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn fully_commented_file_is_a_noop() {
+        let ini = "; default_record = true\n# preferred_backend = cpu\n[general]\n\n";
+        assert!(seed_from_ini(ini).is_none());
+    }
+
+    #[test]
+    fn coerces_bool_int_and_string_fields() {
+        let ini = "default_record = true\nretention_max_age_days = 90\npreferred_backend = cpu\n";
+        let (settings, model) = seed_from_ini(ini).expect("overrides present");
+        assert!(settings.default_record);
+        assert_eq!(settings.retention_max_age_days, Some(90));
+        assert_eq!(settings.preferred_backend, "cpu");
+        assert!(model.is_none());
+    }
+
+    #[test]
+    fn auto_download_returns_the_configured_model() {
+        let ini = "whisper_model = base.en-q5_1\nauto_download_model = true\n";
+        let (_settings, model) = seed_from_ini(ini).expect("overrides present");
+        assert_eq!(model.as_deref(), Some("base.en-q5_1"));
+    }
+
+    #[test]
+    fn unset_fields_keep_their_defaults() {
+        let ini = "default_record = true\n";
+        let (settings, _) = seed_from_ini(ini).unwrap();
+        // microphone stays on, auto_start stays off — only the named key changed.
+        assert!(settings.microphone_enabled);
+        assert!(!settings.auto_start);
+    }
+
+    #[test]
+    fn secret_keys_are_ignored() {
+        let ini = "anthropic_api_key = sk-should-be-dropped\ndefault_record = true\n";
+        let map = parse_ini(ini);
+        assert!(!map.contains_key("anthropic_api_key"));
+        assert!(map.contains_key("default_record"));
+    }
+}

src-tauri/src/diarization/mod.rs

@@ -23,6 +23,11 @@ pub enum DiarError {
 pub trait Diarizer: Send + Sync {
     /// Partition audio into speaker spans.
     fn diarize(&self, wav: &Path) -> Result<Vec<SpeakerSpan>, DiarError>;
+    /// Partition already-loaded 16kHz mono samples into speaker spans. Phase 3
+    /// (FR-SPK) diarizes a *masked* far-side signal (the mic's "You" ranges
+    /// zeroed), so the caller loads and masks the samples itself rather than
+    /// pointing at the raw WAV.
+    fn diarize_samples(&self, samples: Vec<f32>) -> Result<Vec<SpeakerSpan>, DiarError>;
     /// Assign speaker labels to transcript segments by overlap with spans.
     fn assign(&self, segments: &mut [TranscriptSegment], spans: &[SpeakerSpan]);
 }
@@ -53,6 +58,60 @@ pub fn assign_by_overlap(segments: &mut [TranscriptSegment], spans: &[SpeakerSpa
     }
 }
 
+/// Split-layout attribution (FR-SPK): decides per segment between "You" (mic
+/// channel voice activity) and the far side's diarized speakers by comparing
+/// the *total* voiced overlap on each channel, not by picking the single
+/// longest span — a long far-side diarizer span could otherwise swallow a
+/// segment the user spoke most of, showing their words under "Speaker N".
+/// The mic channel is physically the user's voice alone, so channel evidence
+/// outranks cluster evidence; ties go to "You" (mislabeling the user's own
+/// words as someone else is the worse failure). A segment with no voiced
+/// overlap on either channel keeps its prior label rather than guessing.
+// ponytail: whole-segment labels — a segment genuinely containing both sides
+// still gets one speaker; the upgrade path is transcribing each channel
+// separately so segments can never mix voices.
+pub fn assign_split(
+    segments: &mut [TranscriptSegment],
+    you_spans: &[(u64, u64)],
+    far_vad: &[(u64, u64)],
+    far_spans: &[SpeakerSpan],
+) {
+    fn overlap(a0: u64, a1: u64, b0: u64, b1: u64) -> u64 {
+        a1.min(b1).saturating_sub(a0.max(b0))
+    }
+    for seg in segments.iter_mut() {
+        let mic_ms: u64 = you_spans
+            .iter()
+            .map(|&(s, e)| overlap(seg.start_ms, seg.end_ms, s, e))
+            .sum();
+        let far_ms: u64 = far_vad
+            .iter()
+            .map(|&(s, e)| overlap(seg.start_ms, seg.end_ms, s, e))
+            .sum();
+        if mic_ms == 0 && far_ms == 0 {
+            continue;
+        }
+        if mic_ms >= far_ms {
+            seg.speaker = "You".to_string();
+        } else if let Some(span) = far_spans
+            .iter()
+            .map(|sp| {
+                (
+                    overlap(seg.start_ms, seg.end_ms, sp.start_ms, sp.end_ms),
+                    sp,
+                )
+            })
+            .filter(|(o, _)| *o > 0)
+            .max_by_key(|(o, _)| *o)
+            .map(|(_, sp)| sp)
+        {
+            seg.speaker = span.speaker.clone();
+        }
+        // Far side voiced but no diarizer span overlaps (e.g. a sub-700ms span
+        // was filtered): keep the prior label rather than guess.
+    }
+}
+
 /// sherpa-onnx-backed diarizer: pyannote segmentation + speaker-embedding +
 /// fast clustering (ADR-0005, T4.1). `Diarize::compute` needs `&mut self`; it's
 /// wrapped in a `Mutex` to satisfy `Diarizer: Sync` — diarization is a
@@ -70,7 +129,13 @@ impl SherpaDiarizer {
             // sherpa-onnx to pick the cluster count itself from `threshold`
             // instead of forcing a fixed number of speakers.
             num_clusters: Some(-1),
-            threshold: Some(0.5),
+            // Tuning evidence: 0.5 over the summed mic+loopback mono WAV
+            // exploded a 2-speaker conversation into 83 clusters (overlapped
+            // speech + short pyannote chunks yield unstable embeddings). sherpa
+            // fast-clustering: larger threshold → fewer clusters. 0.7 lands a
+            // 2-person meeting near 2–4. ponytail: code constant, no settings
+            // knob until 0.7 proves insufficient in the field.
+            threshold: Some(0.7),
             ..Default::default()
         };
         let engine = sherpa_rs::diarize::Diarize::new(segmentation_model, embedding_model, config)
@@ -86,11 +151,28 @@ impl Diarizer for SherpaDiarizer {
     fn diarize(&self, wav: &Path) -> Result<Vec<SpeakerSpan>, DiarError> {
         let samples =
             crate::audio::read_wav_mono_16k(wav).map_err(|e| DiarError::Run(e.to_string()))?;
+        self.diarize_samples(samples)
+    }
+
+    fn diarize_samples(&self, samples: Vec<f32>) -> Result<Vec<SpeakerSpan>, DiarError> {
+        // Guard the native FFI against degenerate input (a fully-masked far side
+        // is all zeros; a tiny buffer has no usable window). sherpa can crash
+        // rather than error on these, so skip cleanly instead.
+        if samples.len() < MIN_DIARIZE_SAMPLES {
+            return Ok(Vec::new());
+        }
         let mut engine = self.engine.lock().unwrap_or_else(|e| e.into_inner());
         let segments = engine
             .compute(samples, None)
             .map_err(|e| DiarError::Run(e.to_string()))?;
-        Ok(segments.into_iter().map(segment_to_span).collect())
+        // Drop sub-700ms spans: their embeddings are unstable and only cause
+        // label churn in assign_by_overlap (which keeps the prior label when no
+        // span overlaps). See MIN_SPAN_MS.
+        Ok(segments
+            .into_iter()
+            .map(segment_to_span)
+            .filter(span_long_enough)
+            .collect())
     }
 
     fn assign(&self, segments: &mut [TranscriptSegment], spans: &[SpeakerSpan]) {
@@ -98,6 +180,23 @@ impl Diarizer for SherpaDiarizer {
     }
 }
 
+/// Minimum span duration kept from a diarization pass. Sub-second pyannote
+/// chunks carry unstable ERes2Net embeddings and only churn labels.
+#[cfg(feature = "diarization")]
+const MIN_SPAN_MS: u64 = 700;
+
+/// Below this many 16kHz samples (~1s) the diarizer skips the sherpa FFI: a
+/// fully-masked far side is all zeros and a tiny buffer has no usable
+/// segmentation window, either of which can crash the native code.
+#[cfg(feature = "diarization")]
+const MIN_DIARIZE_SAMPLES: usize = 16_000;
+
+/// A span is kept only if it lasts at least [`MIN_SPAN_MS`].
+#[cfg(feature = "diarization")]
+fn span_long_enough(span: &SpeakerSpan) -> bool {
+    span.end_ms.saturating_sub(span.start_ms) >= MIN_SPAN_MS
+}
+
 /// sherpa-onnx speaker indices are 0-based; WA's internal labels are 1-based ("S1"…).
 #[cfg(feature = "diarization")]
 fn segment_to_span(seg: sherpa_rs::diarize::Segment) -> SpeakerSpan {
@@ -165,6 +264,63 @@ mod overlap_tests {
         assign_by_overlap(&mut segments, &[]);
         assert_eq!(segments[0].speaker, "S1");
     }
+
+    #[test]
+    fn split_labels_a_mic_dominant_segment_you_even_against_a_longer_far_span() {
+        // The user spoke 0-4000ms; the far side 4000-6000ms — but the far
+        // cluster span covers the whole window, so the old merged max-overlap
+        // pick handed the entire segment (the user's words included) to the
+        // far speaker. Channel totals must side with the mic instead.
+        let mut segments = vec![segment(0, 6000)];
+        let you = vec![(0u64, 4000u64)];
+        let far_vad = vec![(4000u64, 6000u64)];
+        let far_spans = vec![span(0, 6000, "S1")]; // long far cluster span
+        assign_split(&mut segments, &you, &far_vad, &far_spans);
+        assert_eq!(segments[0].speaker, "You");
+    }
+
+    #[test]
+    fn split_ties_go_to_you() {
+        let mut segments = vec![segment(0, 2000)];
+        let you = vec![(0u64, 1000u64)];
+        let far_vad = vec![(1000u64, 2000u64)];
+        let far_spans = vec![span(1000, 2000, "S1")];
+        assign_split(&mut segments, &you, &far_vad, &far_spans);
+        assert_eq!(segments[0].speaker, "You");
+    }
+
+    #[test]
+    fn split_assigns_the_best_far_span_when_the_far_side_dominates() {
+        let mut segments = vec![segment(0, 3000)];
+        let you = vec![(0u64, 500u64)];
+        let far_vad = vec![(500u64, 3000u64)];
+        let far_spans = vec![span(500, 1000, "S1"), span(1000, 3000, "S2")];
+        assign_split(&mut segments, &you, &far_vad, &far_spans);
+        assert_eq!(segments[0].speaker, "S2");
+    }
+
+    #[test]
+    fn split_keeps_the_prior_label_when_both_channels_are_silent() {
+        let mut segments = vec![segment(5000, 6000)];
+        segments[0].speaker = "S9".to_string();
+        assign_split(
+            &mut segments,
+            &[(0, 1000)],
+            &[(0, 1000)],
+            &[span(0, 1000, "S1")],
+        );
+        assert_eq!(segments[0].speaker, "S9");
+    }
+
+    #[test]
+    fn split_keeps_the_prior_label_when_far_is_voiced_but_no_far_span_overlaps() {
+        // Far VAD hears speech but every diarizer span was filtered (sub-700ms):
+        // don't guess a label.
+        let mut segments = vec![segment(0, 1000)];
+        segments[0].speaker = "S3".to_string();
+        assign_split(&mut segments, &[], &[(0, 1000)], &[span(2000, 3000, "S1")]);
+        assert_eq!(segments[0].speaker, "S3");
+    }
 }
 
 #[cfg(all(test, feature = "diarization"))]
@@ -184,6 +340,22 @@ mod tests {
         assert_eq!(span.speaker, "S1");
     }
 
+    #[test]
+    fn span_long_enough_filters_sub_minimum_spans() {
+        let short = SpeakerSpan {
+            start_ms: 0,
+            end_ms: MIN_SPAN_MS - 1,
+            speaker: "S1".to_string(),
+        };
+        let exact = SpeakerSpan {
+            start_ms: 0,
+            end_ms: MIN_SPAN_MS,
+            speaker: "S1".to_string(),
+        };
+        assert!(!span_long_enough(&short));
+        assert!(span_long_enough(&exact));
+    }
+
     #[test]
     fn new_surfaces_a_load_error_for_missing_models_instead_of_panicking() {
         let result =

src-tauri/src/diarization/voiceprint.rs

@@ -110,8 +110,9 @@ pub fn match_mic_speaker(
 }
 
 /// Distinct speaker labels in first-appearance order — spans come back from
-/// the diarizer already sorted by start time.
+/// the diarizer already sorted by start time. `pub(crate)` so the Phase 3
-fn first_appearance_order(spans: &[SpeakerSpan]) -> Vec<String> {
+/// per-stream path can reuse it to name its merged You + far-side spans.
+pub(crate) fn first_appearance_order(spans: &[SpeakerSpan]) -> Vec<String> {
     let mut seen = std::collections::HashSet::new();
     spans
         .iter()
@@ -152,8 +153,12 @@ fn cosine_similarity(a: &[f32], b: &[f32]) -> f32 {
 
 /// `mic_label` -> "You"; every other label, in first-appearance order ->
 /// "Speaker 2", "Speaker 3", … (numbering starts at 2 — "You" stands in for
-/// "Speaker 1" without ever being called that).
+/// "Speaker 1" without ever being called that). `pub(crate)` so Phase 3 reuses
-fn build_name_map(labels_in_order: &[String], mic_label: &str) -> HashMap<String, String> {
+/// it for its merged You + far-side spans, keeping naming uniform app-wide.
+pub(crate) fn build_name_map(
+    labels_in_order: &[String],
+    mic_label: &str,
+) -> HashMap<String, String> {
     let mut names = HashMap::new();
     let mut next_speaker_number = 2;
     for label in labels_in_order {

src-tauri/src/lib.rs

@@ -9,6 +9,7 @@ pub mod audio;
 pub mod briefs;
 pub mod calendar;
 pub mod commands;
+pub mod deploy;
 pub mod diarization;
 pub mod error;
 pub mod hardware;
@@ -27,7 +28,8 @@ pub mod vault;
 use std::path::PathBuf;
 use std::sync::{Arc, Mutex as StdMutex};
 use std::thread::JoinHandle;
-use tauri::tray::TrayIcon;
+use tauri::menu::{Menu, MenuItem};
+use tauri::tray::{MouseButton, MouseButtonState, TrayIcon, TrayIconBuilder, TrayIconEvent};
 use tauri::Manager;
 use tokio::sync::Mutex;
 
@@ -85,6 +87,10 @@ pub struct RecordingSession {
     /// auto-labeled "You" instead of a clustered "S1"/"S2". `None` when the
     /// mic is disabled (same conditions as `mic_capture`).
     pub mic_voice_sample: Option<Arc<audio::VoiceSample>>,
+    /// `audio.wav` channel layout for this recording (FR-SPK): `"split"` (mic
+    /// enabled → stereo mic-left/loopback-right) or `"summed"` (mic off). Passed
+    /// to `finalize_meeting` so playback/diarization/export can branch on it.
+    pub audio_layout: &'static str,
     /// Live notes redesign: raw user-authored notes accumulated *during* the
     /// recording (freeform text + per-moment annotations) — see
     /// `models::ManualNotes`. Mutated by `update_live_notes`/`set_segment_note`
@@ -111,7 +117,11 @@ fn fatal_startup_error(body: &str) -> ! {
             MessageBoxW, MB_ICONERROR, MB_OK, MB_SYSTEMMODAL,
         };
         // Null-terminated UTF-16 for the Win32 wide API.
-        let to_wide = |s: &str| s.encode_utf16().chain(std::iter::once(0)).collect::<Vec<u16>>();
+        let to_wide = |s: &str| {
+            s.encode_utf16()
+                .chain(std::iter::once(0))
+                .collect::<Vec<u16>>()
+        };
         let text = to_wide(body);
         let caption = to_wide("WhispAssist can't start");
         // SAFETY: both buffers are null-terminated and outlive the call.
@@ -151,6 +161,14 @@ pub fn run() {
 
     tauri::Builder::default()
         .plugin(tauri_plugin_dialog::init())
+        // Opt-in launch-at-login (NFR-RES-4). The macOS launcher arg is required
+        // by the signature but unused on Windows, where enable/disable writes a
+        // per-user HKCU\...\Run entry (no admin). Off until the user (or an
+        // enterprise deploy file) turns `auto_start` on.
+        .plugin(tauri_plugin_autostart::init(
+            tauri_plugin_autostart::MacosLauncher::LaunchAgent,
+            None,
+        ))
         // In-memory streaming of recordings for the player (FR-REC-5): decrypts
         // on the fly so no plaintext audio is ever written to disk.
         .register_uri_scheme_protocol("waaudio", |_ctx, request| {
@@ -161,13 +179,88 @@ pub fn run() {
             session: Mutex::new(None),
         })
         .setup(move |app| {
+            // Single tray icon (the `trayIcon` in tauri.conf.json was removed so
+            // this is the only one). It carries a Show/Quit menu and, on
+            // left-click, restores the window — the always-available way back
+            // from "close to tray".
             let icon = tauri::image::Image::from_bytes(include_bytes!("../icons/tray.png"))?;
-            let tray = tauri::tray::TrayIconBuilder::new()
+            let show_item = MenuItem::with_id(app, "show", "Show WhispAssist", true, None::<&str>)?;
+            let quit_item = MenuItem::with_id(app, "quit", "Quit", true, None::<&str>)?;
+            let menu = Menu::with_items(app, &[&show_item, &quit_item])?;
+            let tray = TrayIconBuilder::new()
                 .icon(icon)
                 .tooltip("WhispAssist — idle")
+                .menu(&menu)
+                .show_menu_on_left_click(false)
+                .on_menu_event(|app, event| match event.id.as_ref() {
+                    "show" => show_main_window(app),
+                    "quit" => app.exit(0),
+                    _ => {}
+                })
+                .on_tray_icon_event(|tray, event| {
+                    if let TrayIconEvent::Click {
+                        button: MouseButton::Left,
+                        button_state: MouseButtonState::Up,
+                        ..
+                    } = event
+                    {
+                        show_main_window(tray.app_handle());
+                    }
+                })
                 .build(app)?;
             app.manage(TrayHandle(tray));
 
+            // First-run enterprise deploy seeding (deploy.rs): if no settings.json
+            // exists yet and an admin dropped a wa-defaults.ini, seed settings once
+            // and optionally fetch the configured model in the background. One-shot
+            // — guarded by the settings file's absence, so it never re-runs and adds
+            // nothing to idle cost (NFR-RES-1).
+            if !crate::paths::settings_path().exists() {
+                if let Some((seeded, model_to_download)) = deploy::seed_settings_from_defaults() {
+                    match commands::save_settings(&seeded) {
+                        Ok(()) => {
+                            tracing::info!("seeded settings.json from wa-defaults.ini");
+                            if let Some(id) = model_to_download {
+                                let app_handle = app.handle().clone();
+                                tauri::async_runtime::spawn(async move {
+                                    if let Err(e) = commands::download_model(
+                                        app_handle,
+                                        commands::DownloadModelArgs {
+                                            kind: "whisper".into(),
+                                            id,
+                                        },
+                                    )
+                                    .await
+                                    {
+                                        tracing::warn!(
+                                            "deploy auto-download of model failed: {e:?}"
+                                        );
+                                    }
+                                });
+                            }
+                        }
+                        Err(e) => {
+                            tracing::error!(
+                                "first-run deploy seeding failed to write settings: {e:?}"
+                            )
+                        }
+                    }
+                }
+            }
+
+            // Reconcile launch-at-login with the persisted preference (NFR-RES-4):
+            // if the user opted in but the OS entry is missing (e.g. after a
+            // reinstall or a deploy file that set auto_start), restore it. One-shot.
+            {
+                use tauri_plugin_autostart::ManagerExt;
+                let manager = app.autolaunch();
+                if commands::load_settings().auto_start && !manager.is_enabled().unwrap_or(false) {
+                    if let Err(e) = manager.enable() {
+                        tracing::warn!("failed to restore auto-start entry: {e}");
+                    }
+                }
+            }
+
             // Reminders (Phase 8, T8.6, FR-CAL-5) are Windows-scheduled toasts, not
             // an app-side timer — Windows itself is what's "polling", so this stays
             // within NFR-RES-1. init() just registers the AppUserModelID.
@@ -261,6 +354,17 @@ pub fn run() {
             });
             Ok(())
         })
+        // Close to tray (keep running in background): when the setting is on,
+        // the window X hides instead of quitting; the tray "Quit" is the real
+        // exit. Off → default behavior (closing the window quits the app).
+        .on_window_event(|window, event| {
+            if let tauri::WindowEvent::CloseRequested { api, .. } = event {
+                if commands::load_settings().close_to_tray {
+                    api.prevent_close();
+                    let _ = window.hide();
+                }
+            }
+        })
         .invoke_handler(tauri::generate_handler![
             commands::start_recording,
             commands::stop_recording,
@@ -268,6 +372,7 @@ pub fn run() {
             commands::recording_playback_path,
             commands::pause_recording,
             commands::resume_recording,
+            commands::toggle_microphone_mute,
             commands::set_recording_retention,
             commands::acknowledge_recording_consent,
             commands::update_live_notes,
@@ -279,6 +384,9 @@ pub fn run() {
             commands::list_audio_devices,
             commands::list_input_devices,
             commands::set_preferred_backend,
+            commands::set_auto_start,
+            commands::monitor_audio_level,
+            commands::stress_test_hardware,
             commands::list_models,
             commands::list_whisper_languages,
             commands::download_npu_package,
@@ -307,6 +415,7 @@ pub fn run() {
             commands::generate_summary,
             commands::confirm_action_items,
             commands::generate_tags,
+            commands::enhance_notes,
             commands::llm_setup_suggestions,
             commands::pull_ollama_model,
             commands::import_pst,
@@ -351,6 +460,16 @@ pub fn run() {
         .expect("error while running WhispAssist");
 }
 
+/// Restore the main window from the tray (show + unminimize + focus). Shared by
+/// the tray left-click and the "Show WhispAssist" menu item.
+fn show_main_window(app: &tauri::AppHandle) {
+    if let Some(w) = app.get_webview_window("main") {
+        let _ = w.show();
+        let _ = w.unminimize();
+        let _ = w.set_focus();
+    }
+}
+
 /// Used by `commands.rs` to keep the tray tooltip honest about capture state (FR-CAP-4).
 pub(crate) fn update_tray_tooltip(app: &tauri::AppHandle, text: &str) {
     if let Some(tray) = app.try_state::<TrayHandle>() {

src-tauri/src/media.rs

@@ -21,10 +21,7 @@ pub enum MediaError {
     #[error("{0} isn't installed or on PATH — install it and try again")]
     ToolMissing(&'static str),
     #[error("{tool} failed: {message}")]
-    Failed {
+    Failed { tool: &'static str, message: String },
-        tool: &'static str,
-        message: String,
-    },
     #[error("io error: {0}")]
     Io(#[from] std::io::Error),
 }
@@ -60,7 +57,11 @@ fn run(program: &'static str, cmd: &mut Command) -> Result<(), MediaError> {
     if !output.status.success() {
         let stderr = String::from_utf8_lossy(&output.stderr);
         let stdout = String::from_utf8_lossy(&output.stdout);
-        let text = if stderr.trim().is_empty() { stdout } else { stderr };
+        let text = if stderr.trim().is_empty() {
+            stdout
+        } else {
+            stderr
+        };
         let tail: Vec<&str> = text.lines().filter(|l| !l.trim().is_empty()).collect();
         let start = tail.len().saturating_sub(6);
         return Err(MediaError::Failed {
@@ -145,12 +146,7 @@ mod tests {
     fn import_reports_a_missing_local_file_without_touching_a_tool() {
         let dir = std::env::temp_dir().join(format!("wa-media-{}", uuid::Uuid::new_v4()));
         std::fs::create_dir_all(&dir).unwrap();
-        let err = import_to_wav(
+        let err = import_to_wav("no-such-file.mp4", &dir.join("out.wav"), &dir).unwrap_err();
-            "no-such-file.mp4",
-            &dir.join("out.wav"),
-            &dir,
-        )
-        .unwrap_err();
         assert!(matches!(err, MediaError::Failed { tool: "import", .. }));
         let _ = std::fs::remove_dir_all(&dir);
     }

src-tauri/src/models.rs

@@ -214,6 +214,11 @@ pub struct MeetingBundle {
     pub tags: Vec<String>,
     pub speakers: Vec<SpeakerInfo>,
     pub action_items: Vec<ActionItem>,
+    /// `audio.wav` channel layout (FR-SPK/FR-CAP): `"split"` or `"summed"`.
+    /// `default` so bundles exported before this field deserialize as `None`
+    /// (treated as `"summed"`).
+    #[serde(default)]
+    pub audio_layout: Option<String>,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -349,6 +354,17 @@ pub struct Settings {
     pub mcp_expose: String, // none|selected|all
     #[serde(default)]
     pub mcp_expose_recordings: bool,
+    /// Launch WhispAssist automatically at login (opt-in, NFR-RES-4). OFF by
+    /// default; toggled via `set_auto_start`, which writes a per-user
+    /// `HKCU\...\Run` entry through `tauri-plugin-autostart` (no admin). An
+    /// enterprise deploy file may set this to `true` (see `deploy.rs`).
+    #[serde(default)]
+    pub auto_start: bool,
+    /// Closing the window hides WhispAssist to the system tray instead of
+    /// quitting, so it keeps running in the background (tray "Quit" really
+    /// exits). ON by default; the tray icon is the always-available way back.
+    #[serde(default = "default_true")]
+    pub close_to_tray: bool,
 }
 
 fn default_mcp_transport() -> String {

src-tauri/src/storage/mod.rs

@@ -74,6 +74,10 @@ pub struct FinalizeMeeting {
     pub language: Option<String>,
     pub backend_used: Option<String>,
     pub model_used: Option<String>,
+    /// How `audio.wav`'s channels are laid out (FR-SPK/FR-CAP): `"split"` =
+    /// stereo mic-left/loopback-right, `"summed"`/`None` = mic mixed into every
+    /// channel. `None` leaves the stored value unchanged (e.g. on reprocess).
+    pub audio_layout: Option<String>,
 }
 
 /// Full meeting detail: DB row + transcript + speakers + notes (`get_meeting`'s
@@ -91,6 +95,10 @@ pub struct Meeting {
     pub language: Option<String>,
     pub backend_used: Option<String>,
     pub model_used: Option<String>,
+    /// `audio.wav` channel layout (FR-SPK/FR-CAP): `"split"` (mic-left /
+    /// loopback-right) or `"summed"`/`None` (mic mixed into every channel).
+    /// Diarization, playback downmix, and bundle export branch on it.
+    pub audio_layout: Option<String>,
     pub segments: Vec<TranscriptSegment>,
     pub speakers: Vec<SpeakerInfo>,
     pub notes_markdown: String,
@@ -268,6 +276,10 @@ pub trait Store: Send + Sync {
     ) -> Result<Vec<MeetingListItem>, StoreError>;
     async fn get_meeting(&self, id: &MeetingId) -> Result<Meeting, StoreError>;
     async fn delete_meeting(&self, id: &MeetingId) -> Result<(), StoreError>;
+    /// Overwrite a meeting's lifecycle `status` (e.g. mark a background import
+    /// `transcribing` while it runs, or `error` if it fails). `finalize_meeting`
+    /// is still the only path to `ready`.
+    async fn set_meeting_status(&self, id: &MeetingId, status: &str) -> Result<(), StoreError>;
     async fn update_notes(&self, id: &MeetingId, markdown: &str) -> Result<(), StoreError>;
     /// (Re)builds this meeting's FTS index row from the current title and
     /// whatever's on disk/in the DB for transcript/notes/summary/tags (Phase
@@ -284,6 +296,14 @@ pub trait Store: Send + Sync {
         label: &str,
         name: &str,
     ) -> Result<(), StoreError>;
+    /// Delete every `speakers` row for a meeting (labels, names, participant
+    /// links, merges). Used before a full re-diarization rebuild
+    /// (`reprocess_transcript`, FR-SPK): the old labels key to the previous
+    /// clustering and are meaningless once the audio is re-clustered, so the
+    /// caller re-inserts the fresh set via `finalize_meeting`. Without this,
+    /// stale labels from an over-split run linger in the DB (and the
+    /// Participants pane) even though the transcript no longer references them.
+    async fn clear_speakers(&self, id: &MeetingId) -> Result<(), StoreError>;
     /// Fold over-split speaker labels into one canonical label (T4.5,
     /// FR-SPK-3). Segment speaker IDs in storage are never rewritten
     /// (FR-SPK-5) — `get_meeting` resolves `from` labels to `into` when it
@@ -734,7 +754,8 @@ impl Store for SqliteStore {
         let now = now_unix();
         sqlx::query(
             "UPDATE meetings SET status = 'ready', ended_at = ?, duration_secs = ?, recorded = ?,
-                language = ?, backend_used = ?, model_used = ?, updated_at = ? WHERE id = ?",
+                language = ?, backend_used = ?, model_used = ?,
+                audio_layout = COALESCE(?, audio_layout), updated_at = ? WHERE id = ?",
         )
         .bind(now)
         .bind(s.duration_secs)
@@ -742,6 +763,7 @@ impl Store for SqliteStore {
         .bind(&s.language)
         .bind(&s.backend_used)
         .bind(&s.model_used)
+        .bind(&s.audio_layout)
         .bind(now)
         .bind(id)
         .execute(&self.pool)
@@ -857,7 +879,7 @@ impl Store for SqliteStore {
 
     async fn get_meeting(&self, id: &MeetingId) -> Result<Meeting, StoreError> {
         let row = sqlx::query(
-            "SELECT id, title, started_at, ended_at, duration_secs, status, recorded, language, backend_used, model_used, calendar_event_id, template_id
+            "SELECT id, title, started_at, ended_at, duration_secs, status, recorded, language, backend_used, model_used, calendar_event_id, template_id, audio_layout
              FROM meetings WHERE id = ?",
         )
         .bind(id)
@@ -928,6 +950,7 @@ impl Store for SqliteStore {
             language: row.get("language"),
             backend_used: row.get("backend_used"),
             model_used: row.get("model_used"),
+            audio_layout: row.get("audio_layout"),
             segments,
             speakers,
             notes_markdown,
@@ -957,6 +980,16 @@ impl Store for SqliteStore {
         Ok(())
     }
 
+    async fn set_meeting_status(&self, id: &MeetingId, status: &str) -> Result<(), StoreError> {
+        sqlx::query("UPDATE meetings SET status = ?, updated_at = ? WHERE id = ?")
+            .bind(status)
+            .bind(now_unix())
+            .bind(id)
+            .execute(&self.pool)
+            .await?;
+        Ok(())
+    }
+
     async fn update_notes(&self, id: &MeetingId, markdown: &str) -> Result<(), StoreError> {
         write_artifact(
             &paths::meeting_dir(id).join("notes.md"),
@@ -1227,13 +1260,15 @@ impl Store for SqliteStore {
         started_at: i64,
         ended_at: Option<i64>,
     ) -> Result<(), StoreError> {
-        sqlx::query("UPDATE meetings SET started_at = ?, ended_at = ?, updated_at = ? WHERE id = ?")
+        sqlx::query(
-            .bind(started_at)
+            "UPDATE meetings SET started_at = ?, ended_at = ?, updated_at = ? WHERE id = ?",
-            .bind(ended_at)
+        )
-            .bind(now_unix())
+        .bind(started_at)
-            .bind(meeting_id)
+        .bind(ended_at)
-            .execute(&self.pool)
+        .bind(now_unix())
-            .await?;
+        .bind(meeting_id)
+        .execute(&self.pool)
+        .await?;
         Ok(())
     }
 
@@ -1252,6 +1287,14 @@ impl Store for SqliteStore {
             .await
     }
 
+    async fn clear_speakers(&self, id: &MeetingId) -> Result<(), StoreError> {
+        sqlx::query("DELETE FROM speakers WHERE meeting_id = ?")
+            .bind(id)
+            .execute(&self.pool)
+            .await?;
+        Ok(())
+    }
+
     async fn merge_speakers(
         &self,
         id: &MeetingId,
@@ -2015,6 +2058,7 @@ mod tests {
                     language: Some("fr".to_string()), // what auto-detect resolved to
                     backend_used: Some("cpu".to_string()),
                     model_used: Some("small-q5_1".to_string()),
+                    audio_layout: None,
                 },
             )
             .await
@@ -2058,6 +2102,7 @@ mod tests {
                     language: Some("en".to_string()),
                     backend_used: Some("cpu".to_string()),
                     model_used: Some("small-q5_1".to_string()),
+                    audio_layout: None,
                 },
             )
             .await

src-tauri/src/transcription/mod.rs

@@ -588,7 +588,10 @@ mod tests {
         feed_into(&mut s, &step, &d2, &mut out);
         assert!(out[3].interim && out[3].text == "next sentence");
         assert_ne!(out[3].id, out[2].id);
-        assert!(out[3].start_ms >= 3000, "starts after the 3 committed steps");
+        assert!(
+            out[3].start_ms >= 3000,
+            "starts after the 3 committed steps"
+        );
     }
 
     #[test]
@@ -612,7 +615,10 @@ mod tests {
         assert!(out.last().unwrap().interim, "still growing before flush");
         flush_into(&mut s, &decode, &mut out);
         let last = out.last().unwrap();
-        assert!(!last.interim && last.text == "partial", "flush finalizes it");
+        assert!(
+            !last.interim && last.text == "partial",
+            "flush finalizes it"
+        );
     }
 
     #[test]

src-tauri/src/transcription/npu.rs

@@ -7,21 +7,24 @@
 //!
 //! Split of work, validated by the T3.4 spike (encoder ~3.6× faster on NPU):
 //! - **encoder** (fixed `[1,80,3000]` shape) runs on the **accelerator** (NPU or
-//!   GPU) — the expensive graph, and the fixed shape accelerators want;
+//!   DirectML GPU) — the expensive graph, and the fixed shape accelerators want;
-//! - **decoder** (dynamic, autoregressive) runs greedy on the **CPU** EP — a
+//! - **decoder** (dynamic, autoregressive) decodes greedily **with a KV cache**
-//!   dynamic KV loop is a poor accelerator fit and the cheap half anyway.
+//!   (Optimum's merged export: `use_cache_branch` + `past_key_values`), so each
-//!
+//!   step feeds one token instead of re-running the whole prefix. On the NPU
-//! ponytail: greedy, no-KV-cache decode (re-feeds the full token prefix each
+//!   path the decoder session itself goes to the **OpenVINO GPU EP** (Intel
-//! step). Correct and simple; windows are short so the token count is small.
+//!   iGPU, same runtime bundle) so sustained CPU stays near-idle; it falls back
-//! Add `decoder_with_past` only if the decoder shows up in a profile.
+//!   to the CPU EP when no iGPU is usable, and `WA_ONNX_DECODER_DEVICE=cpu|gpu`
+//!   overrides the choice. DirectML backends keep the CPU decoder — the DML EP
+//!   handles per-step-growing KV shapes poorly.
 
 use super::{AudioWindow, SegmentSink, Transcriber, TrxError};
 use crate::models::{BackendId, TranscriptSegment};
 use ort::execution_providers::{
     DirectMLExecutionProvider, ExecutionProviderDispatch, OpenVINOExecutionProvider,
 };
-use ort::session::Session;
+use ort::session::{Session, SessionInputValue};
-use ort::value::Tensor;
+use ort::value::{Tensor, ValueType};
+use std::borrow::Cow;
 use std::collections::HashMap;
 use std::path::Path;
 use std::sync::atomic::{AtomicU64, Ordering};
@@ -30,6 +33,36 @@ use std::sync::Mutex;
 /// Safety cap on generated tokens per 30 s window (Whisper's own max is 448).
 const MAX_NEW_TOKENS: usize = 224;
 
+/// One KV-cache tensor pair of the merged decoder, discovered from the graph
+/// at load: the `past_key_values.*` input, its matching `present.*` output,
+/// whether it's decoder self-attention (refreshed every step) or encoder
+/// cross-attention (computed once on the first step, then passed through
+/// untouched by the cache branch), and the dummy shape to feed before any
+/// past exists (see [`dummy_past_dims`]).
+struct KvSlot {
+    past: String,
+    present: String,
+    self_attention: bool,
+    dummy_dims: Vec<usize>,
+}
+
+/// `"past_key_values.3.decoder.key"` → `"present.3.decoder.key"` — Optimum's
+/// fixed naming convention for the merged decoder's cache I/O.
+fn present_name_for(past: &str) -> String {
+    format!("present.{}", past.trim_start_matches("past_key_values."))
+}
+
+/// The "no past yet" shape for a KV input: batch → 1, any other dynamic (−1)
+/// dim → **1** (a dummy length-1 past — ORT's raw-data tensor creation rejects
+/// zero-length dims, and the no-cache branch never reads the values anyway;
+/// same dummy Optimum's own runner feeds), static dims kept.
+fn dummy_past_dims(dims: &[i64]) -> Vec<usize> {
+    dims.iter()
+        .enumerate()
+        .map(|(i, &d)| if i == 0 || d < 0 { 1 } else { d as usize })
+        .collect()
+}
+
 pub struct OnnxTranscriber {
     encoder: Mutex<Session>,
     decoder: Mutex<Session>,
@@ -38,6 +71,13 @@ pub struct OnnxTranscriber {
     dec_ids: String,
     dec_hidden: String,
     dec_logits: String,
+    /// The merged decoder's KV-cache plumbing (see [`KvSlot`]) plus its
+    /// branch selector input.
+    kv_slots: Vec<KvSlot>,
+    dec_use_cache: String,
+    /// Which EP the decoder session actually landed on ("OpenVINO/GPU" or
+    /// "CPU") — surfaced so `load_transcriber`'s engine log tells the truth.
+    decoder_ep: &'static str,
     detok: Detok,
     /// `[decoder_start_token_id, ...forced_decoder_ids]` — the fixed prompt
     /// prefix before generation begins.
@@ -120,29 +160,76 @@ impl OnnxTranscriber {
         let hidden = Tensor::from_array((hidden_shape, hidden_data))
             .map_err(|e| TrxError::Inference(e.to_string()))?;
 
-        // Greedy decode on the CPU.
+        // Greedy KV-cache decode. Step 1 runs the merged graph's no-cache
+        // branch over the full prompt and yields every present KV; later steps
+        // feed a single token plus the cache, refreshing only the decoder
+        // self-attention slots — the cache branch passes encoder
+        // cross-attention KVs through untouched, so the step-1 tensors stay
+        // authoritative (same contract as Optimum's own runner).
+        // ponytail: KVs round-trip host memory each step; wire ort IoBinding
+        // to pin them on-device if GPU-decoder profiling shows the copies matter.
+        let mut kv_cache: Vec<Tensor<f32>> = self
+            .kv_slots
+            .iter()
+            .map(|slot| {
+                let len = slot.dummy_dims.iter().product::<usize>();
+                Tensor::from_array((slot.dummy_dims.clone(), vec![0.0f32; len]))
+            })
+            .collect::<Result<_, _>>()
+            .map_err(|e| TrxError::Inference(e.to_string()))?;
+
         let mut tokens = self.initial_tokens.clone();
+        let mut cached = false; // flips true once step 1 has filled kv_cache
         for _ in 0..MAX_NEW_TOKENS {
-            let ids: Vec<i64> = tokens.clone();
+            let ids: Vec<i64> = if cached {
+                vec![*tokens.last().unwrap_or(&self.eot)]
+            } else {
+                tokens.clone()
+            };
             let ids_tensor = Tensor::from_array(([1usize, ids.len()], ids))
                 .map_err(|e| TrxError::Inference(e.to_string()))?;
+            let branch = Tensor::from_array(([1usize], vec![cached]))
+                .map_err(|e| TrxError::Inference(e.to_string()))?;
 
-            let next = {
+            // Presents are extracted (copied) inside the outputs' borrow scope,
+            // then written back into kv_cache after it ends.
+            let (next, fresh_kvs) = {
                 let mut dec = self
                     .decoder
                     .lock()
                     .map_err(|_| TrxError::Inference("decoder mutex poisoned".into()))?;
+                let mut inputs: Vec<(Cow<'_, str>, SessionInputValue<'_>)> = vec![
+                    (Cow::from(self.dec_ids.as_str()), ids_tensor.view().into()),
+                    (Cow::from(self.dec_hidden.as_str()), hidden.view().into()),
+                    (Cow::from(self.dec_use_cache.as_str()), branch.view().into()),
+                ];
+                for (slot, kv) in self.kv_slots.iter().zip(kv_cache.iter()) {
+                    inputs.push((Cow::from(slot.past.as_str()), kv.view().into()));
+                }
                 let outputs = dec
-                    .run(ort::inputs![
+                    .run(inputs)
-                        self.dec_ids.as_str() => ids_tensor.view(),
-                        self.dec_hidden.as_str() => hidden.view(),
-                    ])
                     .map_err(|e| TrxError::Inference(e.to_string()))?;
                 let (shape, data) = outputs[self.dec_logits.as_str()]
                     .try_extract_tensor::<f32>()
                     .map_err(|e| TrxError::Inference(e.to_string()))?;
-                self.argmax_last(shape, data)
+                let next = self.argmax_last(shape, data);
+
+                let mut fresh: Vec<(usize, Vec<usize>, Vec<f32>)> = Vec::new();
+                for (i, slot) in self.kv_slots.iter().enumerate() {
+                    if slot.self_attention || !cached {
+                        let (s, d) = outputs[slot.present.as_str()]
+                            .try_extract_tensor::<f32>()
+                            .map_err(|e| TrxError::Inference(e.to_string()))?;
+                        fresh.push((i, s.iter().map(|d| *d as usize).collect(), d.to_vec()));
+                    }
+                }
+                (next, fresh)
             };
+            for (i, dims, data) in fresh_kvs {
+                kv_cache[i] = Tensor::from_array((dims, data))
+                    .map_err(|e| TrxError::Inference(e.to_string()))?;
+            }
+            cached = true;
 
             if next == self.eot {
                 break;
@@ -180,6 +267,11 @@ impl OnnxTranscriber {
         best
     }
 
+    /// Which EP the decoder session landed on — "OpenVINO/GPU" or "CPU".
+    pub fn decoder_ep(&self) -> &'static str {
+        self.decoder_ep
+    }
+
     fn segment(&self, text: String, start_ms: u64, end_ms: u64) -> Option<TranscriptSegment> {
         if text.is_empty() {
             return None;
@@ -252,7 +344,7 @@ impl Transcriber for OnnxTranscriber {
         Self::ensure_runtime_env(runtime_dll);
 
         let encoder_path = model.join("encoder_model.onnx");
-        let decoder_path = model.join("decoder_model.onnx");
+        let decoder_path = model.join("decoder_model_merged.onnx");
         let tokenizer_path = model.join("tokenizer.json");
         let gen_config_path = model.join("generation_config.json");
 
@@ -263,11 +355,46 @@ impl Transcriber for OnnxTranscriber {
             .commit_from_file(&encoder_path)
             .map_err(|e| TrxError::Load(e.to_string()))?;
 
-        // Decoder on the default CPU EP (dynamic shapes; poor NPU fit).
+        // Decoder session. The KV-cache loop is dynamic-shaped and
-        let decoder = Session::builder()
+        // autoregressive — a poor NPU fit — but it needn't burn CPU either: on
-            .map_err(|e| TrxError::Load(e.to_string()))?
+        // the NPU path, try the same OpenVINO runtime's GPU plugin (Intel
-            .commit_from_file(&decoder_path)
+        // iGPU) first and fall back to the default CPU EP if no GPU is usable.
-            .map_err(|e| TrxError::Load(e.to_string()))?;
+        // `WA_ONNX_DECODER_DEVICE=cpu|gpu` overrides (debug escape hatch, like
+        // WA_NPU_RUNTIME_URL). DirectML backends keep the CPU decoder — the
+        // DML EP handles per-step-growing KV shapes poorly.
+        let want_gpu = match std::env::var("WA_ONNX_DECODER_DEVICE").as_deref() {
+            Ok("cpu") => false,
+            Ok("gpu") => true,
+            _ => matches!(backend, BackendId::Npu),
+        };
+        let mut decoder_ep = "CPU";
+        let mut decoder = None;
+        if want_gpu {
+            let gpu = Session::builder()
+                .and_then(|b| {
+                    b.with_execution_providers([OpenVINOExecutionProvider::default()
+                        .with_device_type("GPU")
+                        .build()
+                        .error_on_failure()])
+                })
+                .and_then(|b| b.commit_from_file(&decoder_path));
+            match gpu {
+                Ok(s) => {
+                    decoder_ep = "OpenVINO/GPU";
+                    decoder = Some(s);
+                }
+                Err(e) => tracing::warn!(
+                    "OpenVINO GPU EP unavailable for the decoder ({e}); using the CPU EP"
+                ),
+            }
+        }
+        let decoder = match decoder {
+            Some(s) => s,
+            None => Session::builder()
+                .map_err(|e| TrxError::Load(e.to_string()))?
+                .commit_from_file(&decoder_path)
+                .map_err(|e| TrxError::Load(e.to_string()))?,
+        };
 
         let enc_input = Self::find_input(&encoder, "input_features")?;
         let enc_output = Self::find_output(&encoder, "last_hidden_state")
@@ -275,6 +402,53 @@ impl Transcriber for OnnxTranscriber {
         let dec_ids = Self::find_input(&decoder, "input_ids")?;
         let dec_hidden = Self::find_input(&decoder, "encoder_hidden")?;
         let dec_logits = Self::find_output(&decoder, "logits")?;
+        let dec_use_cache = Self::find_input(&decoder, "use_cache_branch")?;
+
+        // KV-cache plumbing, discovered rather than hardcoded so the layer
+        // count comes from the graph (whisper-base: 6 layers → 24 slots).
+        let mut kv_slots: Vec<KvSlot> = Vec::new();
+        for input in &decoder.inputs {
+            if !input.name.starts_with("past_key_values.") {
+                continue;
+            }
+            let dims: Vec<i64> = match &input.input_type {
+                ValueType::Tensor { shape, .. } => shape.to_vec(),
+                other => {
+                    return Err(TrxError::Load(format!(
+                        "KV input '{}' has non-tensor type {other:?}",
+                        input.name
+                    )))
+                }
+            };
+            let present = present_name_for(&input.name);
+            if !decoder.outputs.iter().any(|o| o.name == present) {
+                return Err(TrxError::Load(format!(
+                    "decoder missing present output '{present}'"
+                )));
+            }
+            kv_slots.push(KvSlot {
+                self_attention: input.name.contains(".decoder."),
+                dummy_dims: dummy_past_dims(&dims),
+                past: input.name.clone(),
+                present,
+            });
+        }
+        if kv_slots.is_empty() {
+            return Err(TrxError::Load(
+                "decoder has no past_key_values inputs — expected the merged (KV-cache) export"
+                    .into(),
+            ));
+        }
+
+        // A pre-KV-cache install leaves the old plain decoder behind — 208 MB
+        // of dead weight once the merged graph loads. Best-effort removal.
+        let legacy = model.join("decoder_model.onnx");
+        if legacy.exists() {
+            match std::fs::remove_file(&legacy) {
+                Ok(()) => tracing::info!("removed superseded decoder_model.onnx"),
+                Err(e) => tracing::warn!("couldn't remove superseded decoder_model.onnx: {e}"),
+            }
+        }
 
         let detok = Detok::from_tokenizer_json(&tokenizer_path)?;
         let gen = GenConfig::from_file(&gen_config_path)?;
@@ -290,6 +464,9 @@ impl Transcriber for OnnxTranscriber {
             dec_ids,
             dec_hidden,
             dec_logits,
+            kv_slots,
+            dec_use_cache,
+            decoder_ep,
             detok,
             initial_tokens,
             special_floor: gen.decoder_start,
@@ -457,6 +634,28 @@ mod tests {
         assert_eq!(d.get(&'A'), Some(&b'A'));
     }
 
+    #[test]
+    fn present_name_follows_optimum_convention() {
+        assert_eq!(
+            present_name_for("past_key_values.3.decoder.key"),
+            "present.3.decoder.key"
+        );
+        assert_eq!(
+            present_name_for("past_key_values.0.encoder.value"),
+            "present.0.encoder.value"
+        );
+    }
+
+    #[test]
+    fn dummy_past_dims_pins_dynamic_axes_to_one_and_keeps_static_ones() {
+        // whisper-base decoder self-attention past: [batch, 8, past_seq, 64]
+        // with batch/past_seq dynamic — both pin to 1 (ORT rejects 0-length
+        // dims for raw-data tensors; the no-cache branch ignores the values).
+        assert_eq!(dummy_past_dims(&[-1, 8, -1, 64]), vec![1, 8, 1, 64]);
+        // A static trailing shape is preserved (batch still pins to 1).
+        assert_eq!(dummy_past_dims(&[-1, 8, 1500, 64]), vec![1, 8, 1500, 64]);
+    }
+
     #[test]
     fn byte_decoder_covers_all_256_bytes() {
         let d = byte_decoder();
@@ -481,10 +680,14 @@ mod tests {
             return;
         };
         let wav = std::env::var("WA_NPU_TEST_WAV").expect("set WA_NPU_TEST_WAV");
+        // Surface the engine's own EP-selection logs (GPU-decoder fallback
+        // warnings etc.) — the test harness has no subscriber otherwise.
+        let _ = tracing_subscriber::fmt().with_env_filter("info").try_init();
         let t0 = std::time::Instant::now();
         let t = OnnxTranscriber::load(Path::new(&model_dir), BackendId::Npu, None)
             .expect("load NPU transcriber");
         let load_ms = t0.elapsed().as_millis();
+        eprintln!("[spike] decoder EP: {}", t.decoder_ep());
         let t1 = std::time::Instant::now();
         let segs = t.transcribe_file(Path::new(&wav)).expect("transcribe");
         let infer_ms = t1.elapsed().as_millis();

src-tauri/src/transcription/onnx_models.rs

@@ -15,9 +15,19 @@ const REPO: &str = "onnx-community/whisper-base.en";
 pub const DEFAULT_ONNX_MODEL: &str = "base.en";
 
 /// The four artifacts the `OnnxTranscriber` needs, relative to the repo root.
+/// The decoder is Optimum's *merged* export — both the no-past and with-past
+/// branches in one graph, selected by a `use_cache_branch` input — which is
+/// what enables the KV-cache decode loop (the plain `decoder_model.onnx`
+/// re-ran the whole prefix every step). Existing installs migrate for free:
+/// swapping this entry flips `is_installed` to false, the startup background
+/// fetch re-runs `download`, and `download` skips the three files already on
+/// disk, fetching only the merged decoder.
 const FILES: &[(&str, &str)] = &[
     ("encoder_model.onnx", "onnx/encoder_model.onnx"),
-    ("decoder_model.onnx", "onnx/decoder_model.onnx"),
+    (
+        "decoder_model_merged.onnx",
+        "onnx/decoder_model_merged.onnx",
+    ),
     ("tokenizer.json", "tokenizer.json"),
     ("generation_config.json", "generation_config.json"),
 ];
@@ -45,7 +55,7 @@ pub fn model_files(id: &str) -> OnnxModelFiles {
     let dir = onnx_dir(id);
     OnnxModelFiles {
         encoder: dir.join("encoder_model.onnx"),
-        decoder: dir.join("decoder_model.onnx"),
+        decoder: dir.join("decoder_model_merged.onnx"),
         tokenizer: dir.join("tokenizer.json"),
         generation_config: dir.join("generation_config.json"),
     }
@@ -125,6 +135,6 @@ mod tests {
     fn catalog_lists_the_four_required_artifacts() {
         assert_eq!(FILES.len(), 4);
         assert!(FILES.iter().any(|(n, _)| *n == "encoder_model.onnx"));
-        assert!(FILES.iter().any(|(n, _)| *n == "decoder_model.onnx"));
+        assert!(FILES.iter().any(|(n, _)| *n == "decoder_model_merged.onnx"));
     }
 }

src-tauri/tauri.conf.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "WhispAssist",
-  "version": "0.5.2",
+  "version": "0.7.3",
   "identifier": "bet.dou.whispassist",
   "build": {
     "frontendDist": "../dist",
@@ -22,18 +22,18 @@
     ],
     "security": {
       "csp": "default-src 'self'; connect-src 'self' http://localhost:* http://127.0.0.1:*; img-src 'self' data:; media-src 'self' http://waaudio.localhost; style-src 'self' 'unsafe-inline'"
-    },
-    "trayIcon": {
-      "iconPath": "icons/tray.png",
-      "tooltip": "WhispAssist"
     }
   },
   "bundle": {
     "active": true,
     "targets": ["msi", "nsis"],
     "icon": ["icons/icon.ico"],
+    "resources": ["wa-defaults.ini"],
     "windows": {
-      "webviewInstallMode": { "type": "downloadBootstrapper" }
+      "webviewInstallMode": { "type": "downloadBootstrapper" },
+      "nsis": {
+        "installMode": "both"
+      }
     }
   }
 }

src-tauri/tauri.vulkan.conf.json

@@ -1,6 +1,6 @@
 {
   "$schema": "gen/schemas/desktop-schema.json",
   "bundle": {
-    "resources": ["vulkan-1.dll"]
+    "resources": ["vulkan-1.dll", "wa-defaults.ini"]
   }
 }

src-tauri/wa-defaults.ini

@@ -0,0 +1,60 @@
+; ============================================================================
+;  WhispAssist enterprise deployment defaults  (wa-defaults.ini)
+; ============================================================================
+;  Read ONCE, on a machine's FIRST launch (before settings.json exists), to
+;  seed the app's default settings. After that the user's own settings.json
+;  wins and this file is ignored. Deploy it with native Windows tooling
+;  (Group Policy / SCCM / Intune file copy) to either location — the first
+;  one found wins:
+;
+;    1.  %PROGRAMDATA%\WhispAssist\wa-defaults.ini   (machine-wide)
+;    2.  <install dir>\wa-defaults.ini               (this bundled template)
+;
+;  As shipped, every setting below is COMMENTED OUT, so a normal install
+;  behaves exactly as if this file were absent. Uncomment and edit the lines
+;  you want to preset. Format is flat "key = value" — no sections required;
+;  [section] headers, ';' and '#' comment lines are ignored.
+;
+;  Values:  true / false  for switches;  a plain number for numeric fields;
+;           text otherwise.  Unknown / misspelled keys are ignored.
+;
+;  SECURITY: never put secrets here. API keys, OAuth tokens and sync
+;  passwords live only in the OS credential store; any key containing
+;  "key", "token", "secret", "credential" or "password" is dropped on read.
+; ----------------------------------------------------------------------------
+
+; ---- Recording (ADR-0009) --------------------------------------------------
+; Record every meeting to disk by default (consent notice still applies).
+; default_record = false
+
+; ---- Transcription backend & model -----------------------------------------
+; preferred_backend = auto        ; auto | npu | nvidia | amd | intel | cpu
+; whisper_model = base.en-q5_1     ; a catalog model id
+; whisper_language = auto          ; auto | ISO-639-1 code (multilingual model only)
+; low_overhead = false
+
+; Download whisper_model in the background on first launch so the machine is
+; ready offline. Requires network at first run.
+; auto_download_model = false
+
+; ---- Storage & retention (FR-STORE-2) --------------------------------------
+; storage_root = C:\ProgramData\WhispAssist\data
+; retention_max_age_days = 90
+; retention_max_size_gb = 20
+
+; ---- Local LLM / summaries (ADR-0007) --------------------------------------
+; llm_provider = ollama            ; ollama | custom | anthropic | off
+; llm_endpoint = http://localhost:11434
+; llm_model = llama3
+
+; ---- Capture & UX ----------------------------------------------------------
+; microphone_enabled = true
+; auto_record_calendar = false
+; theme = system                   ; system | light | dark
+
+; ---- Startup (NFR-RES-4) ---------------------------------------------------
+; Launch WhispAssist automatically at login for the user.
+; auto_start = false
+
+; ---- Sync master switch (ADR-0010; targets/creds configured in-app) --------
+; sync_enabled = false

src/App.svelte

@@ -12,6 +12,7 @@
   import { recording } from "./lib/stores/recording.svelte";
   import { settings } from "./lib/stores/settings.svelte";
   import { meetings } from "./lib/stores/meetings.svelte";
+  import { imports } from "./lib/stores/imports.svelte";
   import { calendar } from "./lib/stores/calendar.svelte";
   import { api, type NoteTemplate, type CalendarEvent } from "./lib/api";
   import { onMount } from "svelte";
@@ -24,6 +25,8 @@
     Square,
     Trash2,
     FilePlus,
+    Mic,
+    MicOff,
     Settings as SettingsIcon,
     AlertTriangle,
     PanelLeftClose,
@@ -131,6 +134,7 @@
     recording.init();
     settings.load();
     meetings.init();
+    imports.init(); // live background-import progress for the tracker
     calendar.load(); // events power the auto-record timer above
     checkVault();
     api
@@ -207,6 +211,18 @@
     } else if (e.ctrlKey && e.key === ",") {
       e.preventDefault();
       showSettings = !showSettings;
+    } else if (
+      // Press "M" to mute/unmute the mic mid-meeting (FR-CAP-7). Bare key (no
+      // modifiers) and only while recording with the mic on.
+      e.key.toLowerCase() === "m" &&
+      !e.ctrlKey &&
+      !e.metaKey &&
+      !e.altKey &&
+      recording.state !== "idle" &&
+      settings.settings.microphone_enabled
+    ) {
+      e.preventDefault();
+      recording.toggleMute();
     }
   }
 
@@ -232,10 +248,16 @@
 <div class="app" data-theme={resolvedTheme}>
   <div class="sr-only" role="status" aria-live="polite">{recordingAnnouncement}</div>
   <header class="bar">
-    <strong>WhispAssist</strong>
-    <span class="muted">{t("app.tagline")}</span>
-    <div class="spacer"></div>
     {#if recording.state === "idle"}
+      <button
+        class="record-btn"
+        onclick={startRecording}
+        title={t("app.record_title")}
+        aria-keyshortcuts="Control+Shift+R"
+      >
+        <Circle size={11} fill="currentColor" aria-hidden="true" />
+        {t("app.record")}
+      </button>
       <select
         class="theme-select"
         bind:value={selectedTemplateId}
@@ -247,23 +269,6 @@
           <option value={tpl.id}>{tpl.name}</option>
         {/each}
       </select>
-      <button
-        class="import-btn"
-        onclick={() => (showImport = true)}
-        title={t("app.add_meeting_title")}
-      >
-        <FilePlus size={13} aria-hidden="true" />
-        {t("app.add_meeting")}
-      </button>
-      <button
-        class="record-btn"
-        onclick={startRecording}
-        title={t("app.record_title")}
-        aria-keyshortcuts="Control+Shift+R"
-      >
-        <Circle size={11} fill="currentColor" aria-hidden="true" />
-        {t("app.record")}
-      </button>
     {:else}
       <button
         class="stop-btn"
@@ -278,6 +283,18 @@
         <Trash2 size={12} aria-hidden="true" />
         {t("app.cancel")}
       </button>
+    {/if}
+    <div class="spacer"></div>
+    {#if recording.state === "idle"}
+      <button
+        class="import-btn"
+        onclick={() => (showImport = true)}
+        title={t("app.add_meeting_title")}
+      >
+        <FilePlus size={13} aria-hidden="true" />
+        {t("app.add_meeting")}
+      </button>
+    {:else}
       <span class="rec">
         <span class="rec-dot" aria-hidden="true"></span>
         {t("app.recording")}
@@ -289,6 +306,23 @@
         micPeak={recording.levelPeakMic}
         showMic={settings.settings.microphone_enabled}
       />
+      {#if settings.settings.microphone_enabled}
+        <button
+          class="mute-btn"
+          class:muted={recording.micMuted}
+          onclick={() => recording.toggleMute()}
+          aria-pressed={recording.micMuted}
+          aria-keyshortcuts="M"
+          title={recording.micMuted ? t("app.unmute_title") : t("app.mute_title")}
+        >
+          {#if recording.micMuted}
+            <MicOff size={14} aria-hidden="true" />
+          {:else}
+            <Mic size={14} aria-hidden="true" />
+          {/if}
+          <span class="sr-only">{recording.micMuted ? t("app.unmute") : t("app.mute")}</span>
+        </button>
+      {/if}
       {#if settings.hardware}
         <span class="backend" title={t("app.backend_title")}>{settings.hardware.active}</span>
       {/if}
@@ -588,10 +622,6 @@
     border-bottom: 1px solid var(--border);
     background: var(--bg-elevated);
   }
-  .bar strong {
-    font-size: 0.95rem;
-    letter-spacing: -0.01em;
-  }
   .spacer {
     flex: 1;
   }
@@ -718,6 +748,26 @@
     border-radius: var(--radius-full);
     padding: 0.15rem 0.5rem;
   }
+  .mute-btn {
+    display: inline-flex;
+    align-items: center;
+    justify-content: center;
+    width: 30px;
+    height: 30px;
+    color: var(--fg);
+    background: var(--bg);
+    border: 1px solid var(--border);
+    border-radius: var(--radius-full);
+    cursor: pointer;
+  }
+  .mute-btn:hover {
+    background: var(--bg-hover);
+  }
+  .mute-btn.muted {
+    color: var(--danger, #d33);
+    border-color: var(--danger, #d33);
+    background: color-mix(in srgb, var(--danger, #d33) 12%, transparent);
+  }
   .retention {
     display: flex;
     align-items: center;

src/lib/api.ts

@@ -56,6 +56,19 @@ export interface AudioDeviceInfo {
   name: string;
 }
 
+// Quick hardware stress test (Settings ▸ Hardware): per-(backend, model)
+// real-time factor, plus the recommended real-time-capable pairing.
+export interface StressResult {
+  backend: string;
+  model: string;
+  rtf: number;
+  realtime: boolean;
+}
+export interface StressTestResult {
+  results: StressResult[];
+  recommended: { backend: string; model: string } | null;
+}
+
 export interface LlmStatus {
   provider: string; // ollama|custom|anthropic|off (ADR-0011; "openai" not yet wired)
   reachable: boolean;
@@ -83,6 +96,19 @@ export interface LanguageOption {
 
 export type MeetingStatus = "recording" | "transcribing" | "ready" | "recovering" | "error";
 
+// The four ordered phases of a background media import (import://progress).
+export type ImportPhase = "prepare" | "transcribe" | "diarize" | "finalize";
+
+// One `import://progress` tick. `state` is active (running), done (finished,
+// `elapsedMs` set) or error (`error` message set) for the given `phase`.
+export interface ImportProgress {
+  meetingId: MeetingId;
+  phase: ImportPhase;
+  state: "active" | "done" | "error";
+  elapsedMs: number | null;
+  error: string | null;
+}
+
 export interface MeetingListItem {
   id: MeetingId;
   title: string;
@@ -336,6 +362,12 @@ export interface AppSettings {
   audio_output_device: string | null;
   microphone_enabled: boolean;
   audio_input_device: string | null;
+  /** Launch WhispAssist at login (opt-in, off by default; NFR-RES-4). Toggled
+   * via setAutoStart, which writes a per-user Run entry (no admin). */
+  auto_start: boolean;
+  /** Closing the window hides to the tray (keep running in background) instead
+   * of quitting; on by default. Tray "Quit" is the real exit. */
+  close_to_tray: boolean;
 }
 
 // Feature brief — agent-ready spec distilled from a meeting (ADR-0011).
@@ -398,6 +430,10 @@ export const api = {
     invoke<string>("recording_playback_path", { meetingId }),
   pauseRecording: (meetingId: MeetingId) => invoke<void>("pause_recording", { meetingId }),
   resumeRecording: (meetingId: MeetingId) => invoke<void>("resume_recording", { meetingId }),
+  // Toggle mic mute for the active recording (FR-CAP-7); returns the new muted
+  // state. Errors if the meeting was started with the mic off.
+  toggleMicrophoneMute: (meetingId: MeetingId) =>
+    invoke<boolean>("toggle_microphone_mute", { meetingId }),
   setRecordingRetention: (meetingId: MeetingId, record: boolean) =>
     invoke<void>("set_recording_retention", { meetingId, record }),
   acknowledgeRecordingConsent: () => invoke<void>("acknowledge_recording_consent"),
@@ -417,6 +453,11 @@ export const api = {
   listInputDevices: () => invoke<AudioDeviceInfo[]>("list_input_devices"),
   setPreferredBackend: (backend: BackendId | "auto") =>
     invoke<void>("set_preferred_backend", { args: { backend } }),
+  setAutoStart: (enabled: boolean) => invoke<void>("set_auto_start", { enabled }),
+  // Test a device: stream device://level for a few seconds. Resolves when done.
+  monitorAudioLevel: (kind: "input" | "loopback", deviceId: string | null, durationMs = 6000) =>
+    invoke<void>("monitor_audio_level", { kind, deviceId, durationMs }),
+  stressTestHardware: () => invoke<StressTestResult>("stress_test_hardware"),
   downloadNpuPackage: () => invoke<void>("download_npu_package"),
   downloadDirectmlPackage: () => invoke<void>("download_directml_package"),
   listModels: () => invoke<ModelInfo[]>("list_models"),
@@ -437,10 +478,12 @@ export const api = {
     invoke<void>("reprocess_transcript", { meetingId, model, language }),
   // Manually add a meeting from an existing recording — a local audio/video
   // file path or a URL (YouTube/streaming page or direct media URL). Requires
-  // ffmpeg (and yt-dlp for URLs) on PATH; neither is bundled. Returns the new
+  // ffmpeg (and yt-dlp for URLs) on PATH; neither is bundled. `model` overrides
-  // meeting's id once transcription + diarization have finished.
+  // the Settings whisper model for this one import. Returns the new meeting's id
-  importMedia: (source: string, title?: string) =>
+  // *immediately*; transcode/transcribe/diarize run in the background and stream
-    invoke<MeetingId>("import_media", { source, title }),
+  // `import://progress` ticks, finishing with `transcript://finalized`.
+  importMedia: (source: string, title?: string, model?: string) =>
+    invoke<MeetingId>("import_media", { source, title, model }),
   resumeTranscription: (meetingId: MeetingId) =>
     invoke<void>("resume_transcription", { meetingId }),
   listMeetings: (filter?: MeetingFilter) =>
@@ -460,6 +503,10 @@ export const api = {
   deleteMeeting: (meetingId: MeetingId) => invoke<void>("delete_meeting", { meetingId }),
   updateNotes: (meetingId: MeetingId, markdown: string) =>
     invoke<void>("update_notes", { meetingId, markdown }),
+  // AI-enhance rough notes into structured Markdown grounded in the transcript
+  // (Granola-style). Returns the enhanced text; the caller decides to keep it.
+  enhanceNotes: (meetingId: MeetingId, notes: string) =>
+    invoke<string>("enhance_notes", { meetingId, notes }),
   // dest is a file path for md/pdf/docx/obsidian, a folder for bundle.
   // "obsidian" writes one self-contained vault note (no audio) — FR-STORE-4.
   exportMeeting: (
@@ -588,12 +635,24 @@ export const events = {
   onDeviceChanged: (
     cb: (p: { meetingId: string; recovered: boolean; message: string }) => void,
   ): Promise<UnlistenFn> => listen("recording://device", (e) => cb(e.payload as never)),
+  // Mic mute toggled for the active recording (FR-CAP-7).
+  onMicMuted: (
+    cb: (p: { meetingId: string; muted: boolean }) => void,
+  ): Promise<UnlistenFn> => listen("recording://mic", (e) => cb(e.payload as never)),
   onSegment: (
     cb: (p: { meetingId: string; segment: TranscriptSegment }) => void,
   ): Promise<UnlistenFn> => listen("transcript://segment", (e) => cb(e.payload as never)),
   onFinalized: (
     cb: (p: { meetingId: string; segmentCount: number }) => void,
   ): Promise<UnlistenFn> => listen("transcript://finalized", (e) => cb(e.payload as never)),
+  // Per-phase progress of a background media import (feeds the import tracker).
+  onImportProgress: (cb: (p: ImportProgress) => void): Promise<UnlistenFn> =>
+    listen("import://progress", (e) => cb(e.payload as never)),
+  // Live diarization refined the speaker list mid-recording (FR-SPK): updated
+  // labels/display names, including the mic speaker resolved to "You".
+  onDiarizationUpdated: (
+    cb: (p: { meetingId: string; speakers: SpeakerInfo[] }) => void,
+  ): Promise<UnlistenFn> => listen("diarization://updated", (e) => cb(e.payload as never)),
   onLlmToken: (cb: (p: { meetingId: string; text: string }) => void): Promise<UnlistenFn> =>
     listen("llm://token", (e) => cb(e.payload as never)),
   onLlmDone: (cb: (p: { meetingId: string; summary: SummaryFile }) => void): Promise<UnlistenFn> =>
@@ -604,6 +663,15 @@ export const events = {
   onHardwareChanged: (
     cb: (p: { active: BackendId; reason: string }) => void,
   ): Promise<UnlistenFn> => listen("hardware://changed", (e) => cb(e.payload as never)),
+  // Live level meter for a device test (Settings ▸ Hardware). `done` marks the
+  // end of the monitor window.
+  onDeviceLevel: (
+    cb: (p: { kind: string; rms?: number; peak?: number; done?: boolean }) => void,
+  ): Promise<UnlistenFn> => listen("device://level", (e) => cb(e.payload as never)),
+  // Per-(backend, model) progress ticks during the quick stress test.
+  onStressProgress: (
+    cb: (p: { backend: string; model: string }) => void,
+  ): Promise<UnlistenFn> => listen("stress://progress", (e) => cb(e.payload as never)),
   onNpuDownload: (
     cb: (p: {
       stage: "model" | "runtime" | "done";

src/lib/components/ImportMeeting.svelte

@@ -1,22 +1,43 @@
 <script lang="ts">
   // Manually add a meeting from an existing recording (feature: "add a meeting
   // + upload a video URL or audio file"). Transcoding is done by the backend
-  // via ffmpeg (+ yt-dlp for URLs) — both external, not bundled — so this is
+  // via ffmpeg (+ yt-dlp for URLs) — both external, not bundled — so this is a
-  // just a small form: pick a local file or paste a URL, optional title, go.
+  // small form: pick a file or URL, choose the transcription model, go. Import
-  import { api, errorMessage } from "../api";
+  // runs in the background (import_media returns as soon as the meeting row
+  // exists), so this closes immediately and the meetings list shows progress.
+  import { api, errorMessage, type ModelInfo } from "../api";
   import { open } from "@tauri-apps/plugin-dialog";
+  import { onMount } from "svelte";
   import { trapFocus } from "../actions/trapFocus";
   import { t } from "../i18n/index.svelte";
-  import { X, FileUp, Link as LinkIcon } from "@lucide/svelte";
+  import { X, FileUp, Download } from "@lucide/svelte";
 
   let { onClose, onImported }: { onClose: () => void; onImported: (id: string) => void } = $props();
 
+  // External download pages for the two tools this feature shells out to.
+  const FFMPEG_URL = "https://github.com/BtbN/FFmpeg-Builds/releases/latest";
+  const YTDLP_URL = "https://github.com/yt-dlp/yt-dlp/releases/latest";
+
   // `source` is either a local file path (set via Browse) or a URL (typed).
   let source = $state("");
   let title = $state("");
+  let model = $state("");
+  let models = $state<ModelInfo[]>([]);
   let busy = $state(false);
   let error = $state<string | null>(null);
 
+  // Only installed whisper models are selectable; default to the active one so
+  // the pick matches the user's Settings default unless they change it here.
+  onMount(async () => {
+    try {
+      const all = await api.listModels();
+      models = all.filter((m) => m.installed);
+      model = models.find((m) => m.active)?.id ?? models[0]?.id ?? "";
+    } catch {
+      models = [];
+    }
+  });
+
   async function browse() {
     const path = await open({
       multiple: false,
@@ -51,7 +72,7 @@
     busy = true;
     error = null;
     try {
-      const id = await api.importMedia(source.trim(), title.trim() || undefined);
+      const id = await api.importMedia(source.trim(), title.trim() || undefined, model || undefined);
       onImported(id);
       onClose();
     } catch (e) {
@@ -101,17 +122,32 @@
       </div>
     </label>
 
-    <label class="wide">
+    <div class="grid">
-      {t("import.title_label")} <em>({t("import.optional")})</em>
+      <label>
-      <input bind:value={title} placeholder={t("import.title_placeholder")} disabled={busy} />
+        {t("import.model_label")}
-    </label>
+        <select bind:value={model} disabled={busy || models.length === 0}>
+          {#each models as m (m.id)}
+            <option value={m.id}>{m.label}</option>
+          {/each}
+        </select>
+        <span class="hint">{t("import.model_hint")}</span>
+      </label>
 
-    <p class="muted small">
+      <label>
-      <LinkIcon size={12} aria-hidden="true" />
+        {t("import.title_label")} <em>({t("import.optional")})</em>
-      {t("import.requires_1")} <code>ffmpeg</code>
+        <input bind:value={title} placeholder={t("import.title_placeholder")} disabled={busy} />
-      {t("import.requires_2")} <code>yt-dlp</code>
+      </label>
-      {t("import.requires_3")}
+    </div>
-    </p>
+
+    <div class="tools">
+      <span class="muted small">{t("import.requires")}</span>
+      <button class="tool" type="button" onclick={() => api.openUrl(FFMPEG_URL)}>
+        <Download size={12} aria-hidden="true" /> ffmpeg
+      </button>
+      <button class="tool" type="button" onclick={() => api.openUrl(YTDLP_URL)}>
+        <Download size={12} aria-hidden="true" /> yt-dlp
+      </button>
+    </div>
 
     {#if error}
       <p class="error">{error}</p>
@@ -121,6 +157,7 @@
       <button class="primary" onclick={doImport} disabled={!source.trim() || busy}>
         {busy ? t("import.importing") : t("import.import")}
       </button>
+      <span class="muted small note">{t("import.background_note")}</span>
       <button class="link" onclick={onClose} disabled={busy}>{t("import.cancel")}</button>
     </div>
   </div>
@@ -143,7 +180,7 @@
     border: 1px solid var(--border);
     border-radius: var(--radius-lg);
     padding: 1.25rem;
-    width: min(520px, 100%);
+    width: min(540px, 100%);
     max-height: 90vh;
     overflow: auto;
     box-shadow: 0 12px 40px rgba(0, 0, 0, 0.3);
@@ -179,7 +216,8 @@
     font-weight: 400;
     color: var(--muted);
   }
-  input {
+  input,
+  select {
     width: 100%;
     box-sizing: border-box;
     padding: 0.4rem 0.55rem;
@@ -189,10 +227,18 @@
     color: var(--fg);
     font: inherit;
   }
-  input:focus-visible {
+  input:focus-visible,
+  select:focus-visible {
     border-color: var(--accent);
     outline: none;
   }
+  .hint {
+    display: block;
+    margin-top: 0.25rem;
+    font-size: 0.75rem;
+    font-weight: 400;
+    color: var(--muted);
+  }
   .row {
     display: flex;
     gap: 0.4rem;
@@ -213,18 +259,49 @@
     padding: 0.4rem 0.6rem;
     cursor: pointer;
   }
+  /* Model + title side by side on wide panels, stacked when cramped. */
+  .grid {
+    display: grid;
+    grid-template-columns: 1fr 1fr;
+    gap: 0 0.75rem;
+  }
+  @media (max-width: 460px) {
+    .grid {
+      grid-template-columns: 1fr;
+    }
+  }
+  .tools {
+    display: flex;
+    align-items: center;
+    flex-wrap: wrap;
+    gap: 0.4rem;
+    margin-top: 0.9rem;
+  }
+  .tool {
+    display: inline-flex;
+    align-items: center;
+    gap: 0.25rem;
+    background: var(--bg-hover, transparent);
+    color: var(--accent);
+    border: 1px solid var(--border);
+    border-radius: var(--radius-sm);
+    padding: 0.2rem 0.5rem;
+    font-size: 0.78rem;
+    cursor: pointer;
+  }
+  .tool:hover {
+    border-color: var(--accent);
+  }
   .muted {
     color: var(--muted);
   }
   .small {
     font-size: 0.8rem;
-    display: flex;
-    align-items: center;
-    gap: 0.3rem;
   }
   .error {
     color: var(--danger, #d33);
     font-size: 0.85rem;
+    margin-top: 0.75rem;
   }
   .actions {
     display: flex;
@@ -232,6 +309,10 @@
     gap: 0.6rem;
     margin-top: 1rem;
   }
+  .actions .note {
+    flex: 1;
+    line-height: 1.2;
+  }
   .actions .primary {
     background: var(--accent);
     color: var(--accent-fg, #fff);

src/lib/components/ImportTracker.svelte

@@ -0,0 +1,214 @@
+<script lang="ts">
+  // Domino's-pizza-tracker-style progress for a background media import: four
+  // ordered steps, the running one pulses, finished ones show how long they
+  // took. Fed by the `imports` store (import://progress events). Renders nothing
+  // until the first tick arrives. Design per ui-ux-pro-max: color is never the
+  // only signal (icon + label + time), tabular figures for the timers, and the
+  // pulse is dropped under prefers-reduced-motion.
+  import { imports } from "../stores/imports.svelte";
+  import { t } from "../i18n/index.svelte";
+  import { AudioLines, Captions, Users, FileCheck2, Check, X } from "@lucide/svelte";
+
+  let { meetingId }: { meetingId: string } = $props();
+
+  const run = $derived(imports.get(meetingId));
+
+  const ICONS = {
+    prepare: AudioLines,
+    transcribe: Captions,
+    diarize: Users,
+    finalize: FileCheck2,
+  } as const;
+
+  // ms → compact, human duration for a finished step ("820 ms", "4.3s", "2m 05s").
+  function fmtDur(ms: number | null): string {
+    if (ms == null) return "";
+    if (ms < 1000) return `${ms} ms`;
+    const s = ms / 1000;
+    if (s < 60) return `${s.toFixed(1)}s`;
+    const m = Math.floor(s / 60);
+    const rem = Math.round(s % 60);
+    return `${m}m ${String(rem).padStart(2, "0")}s`;
+  }
+</script>
+
+{#if run}
+  <section class="tracker" aria-label={t("import.tracker.label")}>
+    <header>
+      {#if run.error}
+        <span class="head err">{t("import.tracker.failed")}</span>
+      {:else if run.done}
+        <span class="head ok">{t("import.tracker.done")}</span>
+      {:else}
+        <span class="head">{t("import.tracker.running")}</span>
+      {/if}
+    </header>
+
+    <ol class="steps" aria-live="polite">
+      {#each run.phases as p (p.phase)}
+        {@const Icon = ICONS[p.phase]}
+        <li class="step {p.state}">
+          <div class="node">
+            {#if p.state === "done"}
+              <Check size={18} aria-hidden="true" />
+            {:else if p.state === "error"}
+              <X size={18} aria-hidden="true" />
+            {:else}
+              <Icon size={18} aria-hidden="true" />
+            {/if}
+          </div>
+          <div class="meta">
+            <span class="name">{t(`import.phase.${p.phase}`)}</span>
+            <span class="time">
+              {#if p.state === "done"}{fmtDur(p.elapsedMs)}
+              {:else if p.state === "active"}{t("import.tracker.working")}
+              {:else if p.state === "error"}{t("import.tracker.stopped")}
+              {/if}
+            </span>
+          </div>
+        </li>
+      {/each}
+    </ol>
+
+    {#if run.error}
+      <p class="msg">{run.error}</p>
+    {/if}
+  </section>
+{/if}
+
+<style>
+  .tracker {
+    border: 1px solid var(--border);
+    border-radius: var(--radius-sm);
+    background: var(--panel, var(--bg-elevated));
+    padding: 0.85rem 1rem 1rem;
+  }
+  header {
+    margin-bottom: 0.9rem;
+  }
+  .head {
+    font-size: 0.85rem;
+    font-weight: 600;
+    color: var(--fg);
+  }
+  .head.ok {
+    color: var(--success);
+  }
+  .head.err {
+    color: var(--danger);
+  }
+
+  .steps {
+    display: flex;
+    list-style: none;
+    margin: 0;
+    padding: 0;
+  }
+  .step {
+    flex: 1;
+    position: relative;
+    text-align: center;
+    min-width: 0;
+  }
+  /* Connector from the previous node's center to this one's (each step is the
+     same width, so -50%→+50% spans center to center), sitting behind the node. */
+  .step::before {
+    content: "";
+    position: absolute;
+    top: 17px;
+    left: -50%;
+    width: 100%;
+    height: 2px;
+    background: var(--border);
+    z-index: 0;
+  }
+  .step:first-child::before {
+    display: none;
+  }
+  .step.done::before,
+  .step.active::before,
+  .step.error::before {
+    background: var(--accent);
+  }
+
+  .node {
+    position: relative;
+    z-index: 1;
+    width: 36px;
+    height: 36px;
+    margin: 0 auto 0.45rem;
+    display: grid;
+    place-items: center;
+    border-radius: var(--radius-full);
+    border: 2px solid var(--border);
+    background: var(--bg);
+    color: var(--muted);
+  }
+  .step.active .node {
+    border-color: var(--accent);
+    background: var(--accent-soft, transparent);
+    color: var(--accent);
+    animation: pulse 1.4s ease-out infinite;
+  }
+  .step.done .node {
+    border-color: var(--success);
+    background: var(--success);
+    color: #fff;
+  }
+  .step.error .node {
+    border-color: var(--danger);
+    background: var(--danger);
+    color: #fff;
+  }
+
+  .meta {
+    display: flex;
+    flex-direction: column;
+    gap: 0.1rem;
+    padding: 0 0.2rem;
+  }
+  .name {
+    font-size: 0.78rem;
+    font-weight: 500;
+    color: var(--muted);
+    line-height: 1.2;
+  }
+  .step.active .name,
+  .step.done .name {
+    color: var(--fg);
+  }
+  .time {
+    font-size: 0.72rem;
+    color: var(--muted);
+    font-variant-numeric: tabular-nums;
+    min-height: 1em;
+  }
+  .step.active .time {
+    color: var(--accent);
+  }
+
+  .msg {
+    margin: 0.85rem 0 0;
+    font-size: 0.8rem;
+    color: var(--danger);
+    word-break: break-word;
+  }
+
+  @keyframes pulse {
+    0% {
+      box-shadow: 0 0 0 0 color-mix(in srgb, var(--accent) 45%, transparent);
+    }
+    70% {
+      box-shadow: 0 0 0 8px color-mix(in srgb, var(--accent) 0%, transparent);
+    }
+    100% {
+      box-shadow: 0 0 0 0 color-mix(in srgb, var(--accent) 0%, transparent);
+    }
+  }
+  @media (prefers-reduced-motion: reduce) {
+    .step.active .node {
+      animation: none;
+      box-shadow: 0 0 0 3px var(--accent-soft, transparent);
+    }
+  }
+</style>

src/lib/i18n/en.json

@@ -24,6 +24,25 @@
   "settings.transcription.switch_multilingual": "Switch to a multilingual model above to choose a language.",
 
   "settings.hardware.title": "Hardware",
+  "settings.hardware.refresh": "Refresh",
+  "settings.hardware.test_output": "Test",
+  "settings.hardware.test_mic": "Test",
+  "settings.hardware.testing": "Listening…",
+  "settings.hardware.play_tone": "Play tone",
+  "settings.hardware.stress_title": "Quick stress test",
+  "settings.hardware.stress_hint": "Benchmarks your installed models on each available backend and recommends the most accurate one that still keeps up with live speech. Takes a moment.",
+  "settings.hardware.stress_run": "Run stress test",
+  "settings.hardware.stress_running": "Running…",
+  "settings.hardware.stress_progress": "Benchmarking {pair}…",
+  "settings.hardware.stress_recommend": "Recommended: {backend} + {model}",
+  "settings.hardware.stress_apply": "Apply",
+  "settings.hardware.stress_none": "No installed model keeps up with live speech on this hardware — try a smaller model.",
+  "settings.hardware.stress_backend": "Backend",
+  "settings.hardware.stress_model": "Model",
+  "settings.hardware.stress_rtf": "Speed (×real-time)",
+  "settings.hardware.stress_realtime": "Live?",
+  "settings.hardware.stress_yes": "Yes",
+  "settings.hardware.stress_no": "No",
   "settings.hardware.active_backend": "Active backend",
   "settings.hardware.model_meta": "· model {size}",
   "settings.hardware.preferred_backend": "Preferred backend",
@@ -88,6 +107,7 @@
   "settings.calendar.title": "Calendar & Outlook .pst",
   "settings.calendar.intro_1": "Import events and attendees from a local Outlook",
   "settings.calendar.intro_2": "backup — read-only, nothing is written back to the file. Nothing leaves this device.",
+  "settings.calendar.requires": "Needs this on your PATH (not bundled):",
   "settings.calendar.pst_file": ".pst file",
   "settings.calendar.no_file": "No file selected",
   "settings.calendar.browse": "Browse…",
@@ -278,6 +298,7 @@
   "settings.privacy.locked_word": "locked",
   "settings.privacy.vault_locked_2": ". Unlock to read encrypted meetings.",
   "settings.privacy.password": "Password",
+  "settings.privacy.show_password": "Show password",
   "settings.privacy.unlock": "Unlock",
   "settings.privacy.vault_unlocked_1": "Vault is ",
   "settings.privacy.unlocked_word": "unlocked",
@@ -320,12 +341,23 @@
   "import.optional": "optional",
   "import.title_placeholder": "Defaults to the file name",
   "import.filter_av": "Audio / video",
-  "import.requires_1": "Requires",
+  "import.model_label": "Transcription model",
-  "import.requires_2": "installed and on your PATH (plus",
+  "import.model_hint": "Recorded with the meeting so you can see how it was transcribed.",
-  "import.requires_3": "for URLs). WhispAssist doesn't bundle them.",
+  "import.requires": "Needs these on your PATH (not bundled):",
-  "import.importing": "Importing… this can take a while",
+  "import.background_note": "Runs in the background — track it in the list.",
+  "import.importing": "Starting…",
   "import.import": "Import",
   "import.cancel": "Cancel",
+  "import.tracker.label": "Import progress",
+  "import.tracker.running": "Importing…",
+  "import.tracker.done": "Import complete",
+  "import.tracker.failed": "Import failed",
+  "import.tracker.working": "working…",
+  "import.tracker.stopped": "stopped",
+  "import.phase.prepare": "Transcode",
+  "import.phase.transcribe": "Transcribe",
+  "import.phase.diarize": "Identify speakers",
+  "import.phase.finalize": "Finalize",
 
   "tagchip.filter": "Filter meetings tagged \"{tag}\"",
   "tagchip.remove": "Remove tag {tag}",
@@ -345,6 +377,10 @@
   "app.cancel": "Cancel",
   "app.cancel_title": "Discard this recording and delete it",
   "app.recording": "Recording…",
+  "app.mute": "Mute microphone",
+  "app.unmute": "Unmute microphone",
+  "app.mute_title": "Mute microphone (M)",
+  "app.unmute_title": "Unmute microphone (M)",
   "app.backend_title": "Active transcription backend",
   "app.retention_title": "Save audio as .wav for this meeting",
   "app.saving": "saving",
@@ -397,6 +433,7 @@
 
   "transcript.heading": "Transcript",
   "transcript.title_aria": "Meeting title",
+  "transcript.transcribed_with": "Transcribed with",
   "transcript.lang_title": "Transcription language",
   "transcript.lang_auto": "auto-detecting…",
   "transcript.show": "Show transcript",
@@ -428,9 +465,18 @@
   "notes.italic": "Italic",
   "notes.h1": "Heading 1",
   "notes.h2": "Heading 2",
+  "notes.h3": "Heading 3",
   "notes.bullet": "Bullet list",
+  "notes.numbered": "Numbered list",
+  "notes.quote": "Quote",
+  "notes.divider": "Divider",
   "notes.checkbox_title": "Checkbox",
   "notes.checkbox_aria": "Checkbox list item",
+  "notes.enhance": "Enhance",
+  "notes.enhancing": "Enhancing…",
+  "notes.enhance_title": "Expand these notes into structured Markdown using the transcript (AI)",
+  "notes.enhanced_note": "Notes enhanced from the transcript.",
+  "notes.undo_enhance": "Undo",
   "notes.edit_raw": "Edit the raw markdown",
   "notes.render": "Render the markdown",
   "notes.editor": "Editor",
@@ -527,5 +573,9 @@
   "settings.recording.consent_ack": "acknowledged",
   "settings.recording.consent_not": "not yet acknowledged",
   "settings.recording.auto_label": "Auto-start recording when a calendar event begins",
-  "settings.recording.auto_hint": "Only while WhispAssist is open. When an imported calendar event's start time arrives, a recording begins automatically (using your default retention setting above). Nothing runs in the background — the timer is armed only while the app is running. Import events under Settings → Calendar."
+  "settings.recording.auto_hint": "Only while WhispAssist is open. When an imported calendar event's start time arrives, a recording begins automatically (using your default retention setting above). Nothing runs in the background — the timer is armed only while the app is running. Import events under Settings → Calendar.",
+  "settings.recording.autostart_label": "Launch WhispAssist at login",
+  "settings.recording.autostart_hint": "Starts WhispAssist automatically when you sign in to Windows. Off by default; installs a per-user startup entry (no admin required) and does not begin recording on its own.",
+  "settings.recording.close_tray_label": "Close to system tray",
+  "settings.recording.close_tray_hint": "Closing the window keeps WhispAssist running in the background instead of quitting. Reopen it from the tray icon; use the tray's Quit to exit fully. On by default."
 }

src/lib/stores/imports.svelte.ts

@@ -0,0 +1,63 @@
+// Live per-meeting progress of background media imports (feeds ImportTracker).
+// Fed entirely by `import://progress` events emitted by `import_media`; kept in
+// memory only (the meeting's `status` badge is the persistent story after a
+// restart). See commands.rs `run_import_pipeline`.
+
+import { events, type ImportPhase, type ImportProgress, type MeetingId } from "../api";
+
+export type PhaseState = "pending" | "active" | "done" | "error";
+
+// The four phases in the order the backend runs (and the tracker renders) them.
+export const IMPORT_PHASES: ImportPhase[] = ["prepare", "transcribe", "diarize", "finalize"];
+
+export interface PhaseInfo {
+  phase: ImportPhase;
+  state: PhaseState;
+  elapsedMs: number | null;
+}
+
+export interface ImportRun {
+  meetingId: MeetingId;
+  phases: PhaseInfo[];
+  error: string | null;
+  done: boolean;
+}
+
+function freshRun(meetingId: MeetingId): ImportRun {
+  return {
+    meetingId,
+    phases: IMPORT_PHASES.map((phase) => ({ phase, state: "pending", elapsedMs: null })),
+    error: null,
+    done: false,
+  };
+}
+
+class ImportsStore {
+  runs = $state<Record<MeetingId, ImportRun>>({});
+
+  get(meetingId: MeetingId): ImportRun | undefined {
+    return this.runs[meetingId];
+  }
+
+  async init() {
+    await events.onImportProgress((p) => this.apply(p));
+  }
+
+  private apply(p: ImportProgress) {
+    // Re-read through the record after inserting so we mutate the $state proxy,
+    // not the raw object (Svelte 5 deep reactivity only tracks the proxy).
+    if (!this.runs[p.meetingId]) this.runs[p.meetingId] = freshRun(p.meetingId);
+    const run = this.runs[p.meetingId];
+    const info = run.phases.find((x) => x.phase === p.phase);
+    if (!info) return;
+    info.state = p.state;
+    if (p.state === "done") info.elapsedMs = p.elapsedMs;
+    if (p.state === "error") {
+      run.error = p.error;
+      run.done = true;
+    }
+    if (p.phase === "finalize" && p.state === "done") run.done = true;
+  }
+}
+
+export const imports = new ImportsStore();

src/lib/stores/recording.svelte.ts

@@ -1,7 +1,7 @@
 // Recording state store (Svelte 5 runes-friendly via a small class).
 // Subscribes to recording/transcript events and exposes reactive state.
 
-import { api, events, type TranscriptSegment, type MeetingId } from "../api";
+import { api, events, type TranscriptSegment, type SpeakerInfo, type MeetingId } from "../api";
 import { settings } from "./settings.svelte";
 import { SvelteMap } from "svelte/reactivity";
 
@@ -10,6 +10,9 @@ class RecordingStore {
   state = $state<"idle" | "recording" | "paused">("idle");
   elapsedMs = $state(0);
   segments = $state<TranscriptSegment[]>([]);
+  /** Live speaker list from provisional diarization (FR-SPK): resolves labels
+   * to display names ("You" for the mic speaker) for the live transcript. */
+  speakers = $state<SpeakerInfo[]>([]);
   /** Whether audio is being retained as .wav for the in-flight meeting (ADR-0009). */
   retention = $state(false);
   /** Live system/loopback level for the waveform/meter (FR-CAP-5); 0 when not recording. */
@@ -19,6 +22,9 @@ class RecordingStore {
    * (FR-CAP-7); stays 0 when the mic is disabled or not recording. */
   levelRmsMic = $state(0);
   levelPeakMic = $state(0);
+  /** Mic muted for the in-flight meeting (FR-CAP-7): mic channel goes silent
+   * while loopback keeps recording. Toggled by the "M" key / mute button. */
+  micMuted = $state(false);
   /** Set while a capture-device reconnect is in progress; cleared on recovery (FR-CAP-6). */
   deviceNotice = $state<string | null>(null);
   /** Live notes redesign: freeform text typed in the Notes pane while recording. */
@@ -47,11 +53,16 @@ class RecordingStore {
       this.retention = p.record;
     });
     await events.onSegment(({ segment }) => {
-      // Replace an interim segment with the same id, else append.
+      // Replace an interim segment with the same id, else append. Live
+      // diarization also re-emits committed segments here with a refined
+      // speaker (FR-SPK) — same replace-by-id path applies.
       const i = this.segments.findIndex((s) => s.id === segment.id);
       if (i >= 0) this.segments[i] = segment;
       else this.segments.push(segment);
     });
+    await events.onDiarizationUpdated(({ speakers }) => {
+      this.speakers = speakers;
+    });
     await events.onLevel(({ rms, peak, mic }) => {
       if (mic) {
         this.levelRmsMic = rms;
@@ -64,12 +75,30 @@ class RecordingStore {
     await events.onDeviceChanged(({ recovered, message }) => {
       this.deviceNotice = recovered ? null : message;
     });
+    // Keep mute state in sync even if it was toggled elsewhere (e.g. a future
+    // tray control), not just from this store's toggleMute().
+    await events.onMicMuted(({ muted }) => {
+      this.micMuted = muted;
+    });
+  }
+
+  /** Toggle mic mute for the active recording (FR-CAP-7); no-op if not
+   * recording. Optimistically flips, then reconciles with the backend result. */
+  async toggleMute() {
+    if (!this.meetingId || this.state === "idle") return;
+    try {
+      this.micMuted = await api.toggleMicrophoneMute(this.meetingId);
+    } catch {
+      // Mic off for this meeting (or capture gone) — nothing to mute.
+    }
   }
 
   async start(title?: string, record = false, templateId?: string, calendarEventId?: string) {
     this.segments = [];
+    this.speakers = [];
     this.retention = record;
     this.deviceNotice = null;
+    this.micMuted = false;
     this.notesText = "";
     this.segmentNotes.clear();
     // T8.7/FR-TRX-4: whatever language is currently configured in Settings
@@ -89,6 +118,7 @@ class RecordingStore {
     this.levelPeak = 0;
     this.levelRmsMic = 0;
     this.levelPeakMic = 0;
+    this.micMuted = false;
     this.deviceNotice = null;
   }
 
@@ -101,11 +131,13 @@ class RecordingStore {
     if (this.meetingId) await api.cancelRecording(this.meetingId);
     this.meetingId = null;
     this.segments = [];
+    this.speakers = [];
     this.state = "idle";
     this.levelRms = 0;
     this.levelPeak = 0;
     this.levelRmsMic = 0;
     this.levelPeakMic = 0;
+    this.micMuted = false;
     this.deviceNotice = null;
     this.notesText = "";
     this.segmentNotes.clear();

src/lib/stores/settings.svelte.ts

@@ -48,6 +48,8 @@ const DEFAULT_SETTINGS: AppSettings = {
   audio_output_device: null, // system default render device (FR-CAP-1)
   microphone_enabled: true, // capture the user's mic into the transcript (FR-CAP-7)
   audio_input_device: null, // system default capture device
+  auto_start: false, // launch at login — opt-in, off by default (NFR-RES-4)
+  close_to_tray: true, // closing the window hides to tray; on by default
 };
 
 class SettingsStore {
@@ -383,6 +385,18 @@ class SettingsStore {
     return this.patch({ default_record: on });
   }
 
+  /** Launch-at-login toggle (NFR-RES-4). Goes through its own command (not
+   * patch) since the backend also writes the per-user OS Run entry; that
+   * command persists auto_start itself, so we just mirror it locally. */
+  async setAutoStart(on: boolean) {
+    this.settings = { ...this.settings, auto_start: on };
+    try {
+      await api.setAutoStart(on);
+    } catch {
+      this.backendStub = true;
+    }
+  }
+
   setRetentionPolicy(maxAgeDays: number | null, maxSizeGb: number | null) {
     return this.patch({ retention_max_age_days: maxAgeDays, retention_max_size_gb: maxSizeGb });
   }

src/lib/views/Settings.svelte

@@ -9,9 +9,16 @@
   import { t, i18n, LOCALES } from "../i18n/index.svelte";
   import ConsentNotice from "../components/ConsentNotice.svelte";
   import HostedAiBanner from "../components/HostedAiBanner.svelte";
+  import LevelMeter from "../components/LevelMeter.svelte";
   import { open } from "@tauri-apps/plugin-dialog";
   import { api, errorMessage, events } from "../api";
-  import type { BackendId, SyncKind, SyncTargetConfig, SyncTargetInfo } from "../api";
+  import type {
+    BackendId,
+    StressTestResult,
+    SyncKind,
+    SyncTargetConfig,
+    SyncTargetInfo,
+  } from "../api";
   import { trapFocus } from "../actions/trapFocus";
   import {
     X,
@@ -23,7 +30,11 @@
     CalendarDays,
     UploadCloud,
     ShieldCheck,
+    Lock,
+    LockOpen,
     Sparkles,
+    Volume2,
+    Zap,
     RefreshCw,
     ChevronRight,
     RotateCcw,
@@ -275,6 +286,11 @@
   }
 
   // ---- Calendar / .pst import (T6.1/T6.2/T6.3, FR-CAL-1/2) ----
+  // .pst import shells out to readpst (libpst, ADR-0008) — external like
+  // ffmpeg/yt-dlp, so link the same kind of download button as the Import
+  // dialog's. ezwinports is the maintained Windows build of libpst.
+  const LIBPST_URL =
+    "https://sourceforge.net/projects/ezwinports/files/libpst-0.6.63-w32-bin.zip/download";
   let pstPath = $state("");
   let pstPassword = $state("");
   let eventTitleFilter = $state("");
@@ -381,11 +397,98 @@
   let showConsent = $state(false);
   let testResult = $state<{ ok: boolean; message: string } | null>(null);
 
+  // ---- Audio device test (live level meter) ----
+  let monitorKind = $state<"input" | "loopback" | null>(null);
+  let monitorRms = $state(0);
+  let monitorPeak = $state(0);
+  let monitorUnlisten: (() => void) | null = null;
+  function stopMonitor() {
+    monitorUnlisten?.();
+    monitorUnlisten = null;
+    monitorKind = null;
+    monitorRms = 0;
+    monitorPeak = 0;
+  }
+  async function testDevice(kind: "input" | "loopback") {
+    if (monitorKind) return;
+    monitorKind = kind;
+    monitorRms = 0;
+    monitorPeak = 0;
+    monitorUnlisten = await events.onDeviceLevel((p) => {
+      if (p.kind !== kind) return;
+      if (p.done) {
+        stopMonitor();
+        return;
+      }
+      monitorRms = p.rms ?? 0;
+      monitorPeak = p.peak ?? 0;
+    });
+    const deviceId =
+      kind === "input"
+        ? (settings.settings.audio_input_device ?? null)
+        : (settings.settings.audio_output_device ?? null);
+    try {
+      await api.monitorAudioLevel(kind, deviceId, 6000);
+    } catch (e) {
+      testResult = { ok: false, message: errorMessage(e) };
+    } finally {
+      stopMonitor();
+    }
+  }
+  // A 440Hz beep to the default output so the user can confirm speakers work.
+  function playTone() {
+    try {
+      const ctx = new AudioContext();
+      const osc = ctx.createOscillator();
+      const gain = ctx.createGain();
+      osc.frequency.value = 440;
+      gain.gain.value = 0.15;
+      osc.connect(gain).connect(ctx.destination);
+      osc.start();
+      osc.stop(ctx.currentTime + 0.5);
+      osc.onended = () => ctx.close();
+    } catch {
+      /* no Web Audio available */
+    }
+  }
+
+  // ---- Quick hardware stress test ----
+  let stressRunning = $state(false);
+  let stressProgress = $state<string | null>(null);
+  let stressResult = $state<StressTestResult | null>(null);
+  let stressError = $state<string | null>(null);
+  async function runStressTest() {
+    if (stressRunning) return;
+    stressRunning = true;
+    stressError = null;
+    stressResult = null;
+    const un = await events.onStressProgress((p) => {
+      stressProgress = `${p.backend} · ${p.model}`;
+    });
+    try {
+      stressResult = await api.stressTestHardware();
+    } catch (e) {
+      stressError = errorMessage(e);
+    } finally {
+      un();
+      stressProgress = null;
+      stressRunning = false;
+    }
+  }
+  async function applyRecommendation() {
+    const r = stressResult?.recommended;
+    if (!r) return;
+    await settings.setPreferredBackend(r.backend as BackendId | "auto");
+    await settings.patch({ whisper_model: r.model });
+  }
+
   // ---- At-rest encryption vault (T8.8, FR-SEC-3) ----
   let vault = $state<{ enabled: boolean; unlocked: boolean } | null>(null);
   let vaultPw = $state("");
   let vaultPw2 = $state("");
   let vaultMsg = $state<string | null>(null);
+  let vaultMsgError = $state(false);
+  let showVaultPw = $state(false);
   async function loadVault() {
     try {
       vault = await api.vaultStatus();
@@ -394,42 +497,46 @@
     }
   }
   onMount(loadVault);
+  function setVaultMsg(msg: string | null, isError = false) {
+    vaultMsg = msg;
+    vaultMsgError = isError;
+  }
   async function enableVault() {
-    vaultMsg = null;
+    setVaultMsg(null);
     try {
       await api.enableVault(vaultPw);
       vaultPw = "";
-      vaultMsg = "Vault enabled and unlocked.";
+      setVaultMsg("Vault enabled and unlocked.");
       await loadVault();
     } catch (e) {
-      vaultMsg = errorMessage(e);
+      setVaultMsg(errorMessage(e), true);
     }
   }
   async function unlockVault() {
-    vaultMsg = null;
+    setVaultMsg(null);
     try {
       await api.unlockVault(vaultPw);
       vaultPw = "";
-      vaultMsg = "Unlocked.";
+      setVaultMsg("Unlocked.");
       await loadVault();
     } catch (e) {
-      vaultMsg = errorMessage(e);
+      setVaultMsg(errorMessage(e), true);
     }
   }
   async function lockVault() {
     await api.lockVault();
-    vaultMsg = "Locked.";
+    setVaultMsg("Locked.");
     await loadVault();
   }
   async function changeVaultPassword() {
-    vaultMsg = null;
+    setVaultMsg(null);
     try {
       await api.changeVaultPassword(vaultPw, vaultPw2);
       vaultPw = "";
       vaultPw2 = "";
-      vaultMsg = "Password changed.";
+      setVaultMsg("Password changed.");
     } catch (e) {
-      vaultMsg = errorMessage(e);
+      setVaultMsg(errorMessage(e), true);
     }
   }
 
@@ -754,13 +861,40 @@
         </label>
         <p class="muted">{t("settings.recording.auto_hint")}</p>
 
+        <label class="row">
+          <input
+            type="checkbox"
+            checked={settings.settings.auto_start}
+            onchange={(e) => settings.setAutoStart((e.target as HTMLInputElement).checked)}
+          />
+          <span>{t("settings.recording.autostart_label")}</span>
+        </label>
+        <p class="muted">{t("settings.recording.autostart_hint")}</p>
+
+        <label class="row">
+          <input
+            type="checkbox"
+            checked={settings.settings.close_to_tray}
+            onchange={(e) =>
+              settings.patch({ close_to_tray: (e.target as HTMLInputElement).checked })}
+          />
+          <span>{t("settings.recording.close_tray_label")}</span>
+        </label>
+        <p class="muted">{t("settings.recording.close_tray_hint")}</p>
+
         {#if showConsent}
           <ConsentNotice onAccept={acceptConsent} onCancel={() => (showConsent = false)} />
         {/if}
       </section>
     {:else if section === "hardware"}
       <section>
-        <h3>{t("settings.hardware.title")}</h3>
+        <div class="actions">
+          <h3>{t("settings.hardware.title")}</h3>
+          <button class="ghost" onclick={() => settings.loadHardware()}>
+            <RefreshCw size={14} aria-hidden="true" />
+            {t("settings.hardware.refresh")}
+          </button>
+        </div>
         {#if settings.hardware}
           <div class="row">
             {t("settings.hardware.active_backend")}<code>{settings.hardware.active}</code>
@@ -804,6 +938,20 @@
             </select>
           </label>
           <p class="muted">{t("settings.hardware.recording_device_hint")}</p>
+          <div class="device-test">
+            <button onclick={() => testDevice("loopback")} disabled={monitorKind !== null}>
+              <Volume2 size={13} aria-hidden="true" />
+              {monitorKind === "loopback"
+                ? t("settings.hardware.testing")
+                : t("settings.hardware.test_output")}
+            </button>
+            <button onclick={playTone} disabled={monitorKind !== null}>
+              {t("settings.hardware.play_tone")}
+            </button>
+            {#if monitorKind === "loopback"}
+              <LevelMeter rms={monitorRms} peak={monitorPeak} />
+            {/if}
+          </div>
 
           <label
             >{t("settings.hardware.microphone")}
@@ -825,6 +973,77 @@
             </select>
           </label>
           <p class="muted">{t("settings.hardware.mic_hint")}</p>
+          <div class="device-test">
+            <button
+              onclick={() => testDevice("input")}
+              disabled={monitorKind !== null || !settings.settings.microphone_enabled}
+            >
+              <Mic size={13} aria-hidden="true" />
+              {monitorKind === "input"
+                ? t("settings.hardware.testing")
+                : t("settings.hardware.test_mic")}
+            </button>
+            {#if monitorKind === "input"}
+              <LevelMeter rms={monitorRms} peak={monitorPeak} />
+            {/if}
+          </div>
+
+          <h4>{t("settings.hardware.stress_title")}</h4>
+          <p class="muted">{t("settings.hardware.stress_hint")}</p>
+          <button onclick={runStressTest} disabled={stressRunning}>
+            <Zap size={13} aria-hidden="true" />
+            {stressRunning
+              ? t("settings.hardware.stress_running")
+              : t("settings.hardware.stress_run")}
+          </button>
+          {#if stressProgress}
+            <p class="muted">{t("settings.hardware.stress_progress", { pair: stressProgress })}</p>
+          {/if}
+          {#if stressError}<p class="muted err">{stressError}</p>{/if}
+          {#if stressResult}
+            {#if stressResult.recommended}
+              <div class="stress-rec">
+                <ShieldCheck size={14} aria-hidden="true" />
+                <span
+                  >{t("settings.hardware.stress_recommend", {
+                    backend: stressResult.recommended.backend,
+                    model: stressResult.recommended.model,
+                  })}</span
+                >
+                <button class="primary" onclick={applyRecommendation}
+                  >{t("settings.hardware.stress_apply")}</button
+                >
+              </div>
+            {:else}
+              <p class="muted">{t("settings.hardware.stress_none")}</p>
+            {/if}
+            <table class="stress-table">
+              <thead>
+                <tr>
+                  <th>{t("settings.hardware.stress_backend")}</th>
+                  <th>{t("settings.hardware.stress_model")}</th>
+                  <th>{t("settings.hardware.stress_rtf")}</th>
+                  <th>{t("settings.hardware.stress_realtime")}</th>
+                </tr>
+              </thead>
+              <tbody>
+                {#each stressResult.results as r (r.backend + r.model)}
+                  <tr>
+                    <td>{r.backend}</td>
+                    <td>{r.model}</td>
+                    <td class="num">{r.rtf.toFixed(2)}×</td>
+                    <td>
+                      {#if r.realtime}
+                        <Check size={13} aria-hidden="true" /> {t("settings.hardware.stress_yes")}
+                      {:else}
+                        {t("settings.hardware.stress_no")}
+                      {/if}
+                    </td>
+                  </tr>
+                {/each}
+              </tbody>
+            </table>
+          {/if}
 
           {#if settings.hardware.npu?.present}
             {@const npu = settings.hardware.npu}
@@ -1132,6 +1351,12 @@
           {t("settings.calendar.intro_1")} <code>.pst</code>
           {t("settings.calendar.intro_2")}
         </p>
+        <div class="tools">
+          <span class="muted small">{t("settings.calendar.requires")}</span>
+          <button class="tool" type="button" onclick={() => api.openUrl(LIBPST_URL)}>
+            <Download size={12} aria-hidden="true" /> libpst (readpst)
+          </button>
+        </div>
         <div class="grid">
           <label class="wide"
             >{t("settings.calendar.pst_file")}
@@ -1918,67 +2143,118 @@
         {/if}
 
         {#if vault}
-          <h4>{t("settings.privacy.vault_title")}</h4>
+          <div
-          {#if !vault.enabled}
+            class="vault-card"
-            <p class="muted">{t("settings.privacy.vault_intro")}</p>
+            class:locked={vault.enabled && !vault.unlocked}
-            <div class="grid">
+            class:unlocked={vault.enabled && vault.unlocked}
-              <label class="wide"
+          >
-                >{t("settings.privacy.vault_password")}<input
+            <div class="vault-head">
-                  type="password"
+              {#if !vault.enabled}
-                  bind:value={vaultPw}
+                <ShieldCheck size={18} aria-hidden="true" />
-                /></label
+              {:else if !vault.unlocked}
-              >
+                <Lock size={18} aria-hidden="true" />
+              {:else}
+                <LockOpen size={18} aria-hidden="true" />
+              {/if}
+              <h4>{t("settings.privacy.vault_title")}</h4>
+              {#if vault.enabled}
+                <span class="badge" class:busy={!vault.unlocked}>
+                  {vault.unlocked
+                    ? t("settings.privacy.unlocked_word")
+                    : t("settings.privacy.locked_word")}
+                </span>
+              {/if}
             </div>
-            <button class="primary" onclick={enableVault} disabled={vaultPw.length < 8}
+
-              >{t("settings.privacy.enable_vault")}</button
+            {#if !vault.enabled}
-            >
+              <p class="muted">{t("settings.privacy.vault_intro")}</p>
-            <p class="muted">{t("settings.privacy.vault_pw_hint")}</p>
+              <div class="pw-row">
-          {:else if !vault.unlocked}
+                <input
-            <p class="muted">
+                  type={showVaultPw ? "text" : "password"}
-              {t("settings.privacy.vault_locked_1")}<strong
-                >{t("settings.privacy.locked_word")}</strong
-              >{t("settings.privacy.vault_locked_2")}
-            </p>
-            <div class="grid">
-              <label class="wide"
-                >{t("settings.privacy.password")}<input
-                  type="password"
                   bind:value={vaultPw}
-                /></label
+                  placeholder={t("settings.privacy.vault_password")}
-              >
+                  aria-label={t("settings.privacy.vault_password")}
-            </div>
+                />
-            <button class="primary" onclick={unlockVault} disabled={!vaultPw}
+                <button
-              >{t("settings.privacy.unlock")}</button
+                  type="button"
-            >
+                  class="icon pw-toggle"
-          {:else}
+                  onclick={() => (showVaultPw = !showVaultPw)}
-            <p class="muted">
+                  aria-label={t("settings.privacy.show_password")}
-              {t("settings.privacy.vault_unlocked_1")}<strong
+                  title={t("settings.privacy.show_password")}
-                >{t("settings.privacy.unlocked_word")}</strong
-              >{t("settings.privacy.vault_unlocked_2")}
-            </p>
-            <button onclick={lockVault}>{t("settings.privacy.lock_now")}</button>
-            <details>
-              <summary>{t("settings.privacy.change_password")}</summary>
-              <div class="grid">
-                <label class="wide"
-                  >{t("settings.privacy.current_password")}<input
-                    type="password"
-                    bind:value={vaultPw}
-                  /></label
-                >
-                <label class="wide"
-                  >{t("settings.privacy.new_password")}<input
-                    type="password"
-                    bind:value={vaultPw2}
-                  /></label
                 >
+                  {#if showVaultPw}<EyeOff size={14} aria-hidden="true" />{:else}<Eye
+                      size={14}
+                      aria-hidden="true"
+                    />{/if}
+                </button>
               </div>
-              <button onclick={changeVaultPassword} disabled={!vaultPw || vaultPw2.length < 8}
+              <button class="primary" onclick={enableVault} disabled={vaultPw.length < 8}
-                >{t("settings.privacy.change_password")}</button
+                >{t("settings.privacy.enable_vault")}</button
               >
-            </details>
+              <p class="muted hint">{t("settings.privacy.vault_pw_hint")}</p>
-          {/if}
+            {:else if !vault.unlocked}
-          {#if vaultMsg}<p class="muted">{vaultMsg}</p>{/if}
+              <p class="muted">
+                {t("settings.privacy.vault_locked_1")}<strong
+                  >{t("settings.privacy.locked_word")}</strong
+                >{t("settings.privacy.vault_locked_2")}
+              </p>
+              <div class="pw-row">
+                <input
+                  type={showVaultPw ? "text" : "password"}
+                  bind:value={vaultPw}
+                  placeholder={t("settings.privacy.password")}
+                  aria-label={t("settings.privacy.password")}
+                  onkeydown={(e) => e.key === "Enter" && vaultPw && unlockVault()}
+                />
+                <button
+                  type="button"
+                  class="icon pw-toggle"
+                  onclick={() => (showVaultPw = !showVaultPw)}
+                  aria-label={t("settings.privacy.show_password")}
+                  title={t("settings.privacy.show_password")}
+                >
+                  {#if showVaultPw}<EyeOff size={14} aria-hidden="true" />{:else}<Eye
+                      size={14}
+                      aria-hidden="true"
+                    />{/if}
+                </button>
+              </div>
+              <button class="primary" onclick={unlockVault} disabled={!vaultPw}
+                >{t("settings.privacy.unlock")}</button
+              >
+            {:else}
+              <p class="muted">
+                {t("settings.privacy.vault_unlocked_1")}<strong
+                  >{t("settings.privacy.unlocked_word")}</strong
+                >{t("settings.privacy.vault_unlocked_2")}
+              </p>
+              <button onclick={lockVault}>
+                <Lock size={14} aria-hidden="true" />
+                {t("settings.privacy.lock_now")}
+              </button>
+              <details>
+                <summary>{t("settings.privacy.change_password")}</summary>
+                <div class="grid">
+                  <label class="wide"
+                    >{t("settings.privacy.current_password")}<input
+                      type="password"
+                      bind:value={vaultPw}
+                    /></label
+                  >
+                  <label class="wide"
+                    >{t("settings.privacy.new_password")}<input
+                      type="password"
+                      bind:value={vaultPw2}
+                    /></label
+                  >
+                </div>
+                <button onclick={changeVaultPassword} disabled={!vaultPw || vaultPw2.length < 8}
+                  >{t("settings.privacy.change_password")}</button
+                >
+              </details>
+            {/if}
+            {#if vaultMsg}<p class="vault-msg" class:error={vaultMsgError}>{vaultMsg}</p>{/if}
+          </div>
         {/if}
       </section>
     {:else if section === "language"}
@@ -2121,6 +2397,33 @@
     font-size: 0.85rem;
     line-height: 1.5;
   }
+  /* External-tool download chip — identical to the Import dialog's
+     ffmpeg/yt-dlp buttons (ImportMeeting.svelte). */
+  .tools {
+    display: flex;
+    align-items: center;
+    flex-wrap: wrap;
+    gap: 0.4rem;
+    margin-top: 0.9rem;
+  }
+  .tool {
+    display: inline-flex;
+    align-items: center;
+    gap: 0.25rem;
+    background: var(--bg-hover, transparent);
+    color: var(--accent);
+    border: 1px solid var(--border);
+    border-radius: var(--radius-sm);
+    padding: 0.2rem 0.5rem;
+    font-size: 0.78rem;
+    cursor: pointer;
+  }
+  .tool:hover {
+    border-color: var(--accent);
+  }
+  .tools .small {
+    font-size: 0.8rem;
+  }
   .footnote {
     margin-top: 1rem;
     border-top: 1px solid var(--border);
@@ -2429,6 +2732,115 @@
     color: var(--accent, #2563eb);
     border-color: currentColor;
   }
+  .vault-card {
+    margin-top: 0.6rem;
+    padding: 0.85rem 1rem;
+    border: 1px solid var(--border);
+    border-radius: var(--radius-md, 8px);
+    background: var(--bg-elevated);
+    display: flex;
+    flex-direction: column;
+    gap: 0.6rem;
+  }
+  .vault-card.locked {
+    border-color: color-mix(in srgb, var(--accent) 45%, var(--border));
+  }
+  .vault-card.unlocked {
+    border-color: color-mix(in srgb, var(--success, #16a34a) 45%, var(--border));
+  }
+  .vault-head {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+  }
+  .vault-head h4 {
+    margin: 0;
+  }
+  .vault-head :global(svg) {
+    color: var(--muted);
+  }
+  .vault-card.locked .vault-head :global(svg) {
+    color: var(--accent);
+  }
+  .vault-card.unlocked .vault-head :global(svg) {
+    color: var(--success, #16a34a);
+  }
+  .vault-head .badge {
+    margin-left: auto;
+  }
+  .pw-row {
+    display: flex;
+    align-items: center;
+    gap: 0.4rem;
+    max-width: 22rem;
+  }
+  .pw-row input {
+    flex: 1;
+  }
+  .pw-toggle {
+    flex: none;
+  }
+  .vault-card .hint {
+    margin: 0;
+  }
+  .vault-msg {
+    margin: 0;
+    font-size: 0.85rem;
+    color: var(--success, #16a34a);
+  }
+  .vault-msg.error {
+    color: var(--danger);
+  }
+  .device-test {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    flex-wrap: wrap;
+    margin: 0.35rem 0 0.6rem;
+  }
+  .device-test :global(.meter) {
+    flex: 1;
+    min-width: 8rem;
+  }
+  .stress-rec {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    margin: 0.6rem 0;
+    padding: 0.55rem 0.75rem;
+    border: 1px solid color-mix(in srgb, var(--accent) 40%, var(--border));
+    border-radius: var(--radius-md, 8px);
+    background: color-mix(in srgb, var(--accent) 10%, var(--bg));
+  }
+  .stress-rec :global(svg) {
+    color: var(--accent);
+  }
+  .stress-rec span {
+    flex: 1;
+  }
+  .stress-table {
+    width: 100%;
+    border-collapse: collapse;
+    margin-top: 0.5rem;
+    font-size: 0.82rem;
+  }
+  .stress-table th,
+  .stress-table td {
+    text-align: left;
+    padding: 0.3rem 0.5rem;
+    border-bottom: 1px solid var(--border);
+  }
+  .stress-table th {
+    color: var(--muted);
+    font-weight: 600;
+  }
+  .stress-table .num {
+    font-variant-numeric: tabular-nums;
+    text-align: right;
+  }
+  .err {
+    color: var(--danger);
+  }
   .npu-package {
     margin-top: 0.6rem;
     padding: 0.6rem 0.75rem;

src/lib/views/SummaryPanel.svelte

@@ -64,7 +64,7 @@
   // nonce so re-clicking the same segment still jumps). Reading seekNonce is
   // what makes this effect re-run.
   $effect(() => {
-    player.seekNonce;
+    void player.seekNonce;
     const ms = player.seekMs;
     if (ms == null || !audioEl) return;
     audioEl.currentTime = ms / 1000;
@@ -770,47 +770,54 @@
     {#if editableItems.length === 0}
       <p class="muted">{t("summary.ai_empty")}</p>
     {:else}
+      <!-- Two-row card per item: the action text owns the full first row (it
+           was unreadable when six controls shared one row in this narrow
+           pane); owner/due/reminder are a secondary meta row beneath it. -->
       <ul class="action-items">
         {#each editableItems as item, i (i)}
           <li>
-            <input
+            <div class="ai-main">
-              type="checkbox"
+              <input
-              bind:checked={item.confirmed}
+                type="checkbox"
-              aria-label={t("summary.confirmed")}
+                bind:checked={item.confirmed}
-              title={t("summary.confirmed")}
+                aria-label={t("summary.confirmed")}
-            />
+                title={t("summary.confirmed")}
-            <input
+              />
-              class="ai-text"
+              <input
-              bind:value={item.text}
+                class="ai-text"
-              placeholder={t("summary.ai_text_placeholder")}
+                bind:value={item.text}
-              aria-label={t("summary.ai_text_aria")}
+                placeholder={t("summary.ai_text_placeholder")}
-            />
+                aria-label={t("summary.ai_text_aria")}
-            <input
+              />
-              class="ai-owner"
+              <button
-              value={item.owner ?? ""}
+                class="ai-del"
-              oninput={(e) => (item.owner = (e.target as HTMLInputElement).value || null)}
+                onclick={() => removeActionItem(i)}
-              placeholder={t("summary.owner")}
+                title={t("summary.ai_delete_title")}
-              aria-label={t("summary.owner")}
+                aria-label={t("summary.ai_delete_aria")}
-            />
+              >
-            <input
+                <X size={14} aria-hidden="true" />
-              type="date"
+              </button>
-              class="due-date"
+            </div>
-              aria-label={t("summary.due_date")}
+            <div class="ai-meta">
-              value={dueDateInput(item.due_at)}
+              <input
-              onchange={(e) => onDueDateChange(item, (e.target as HTMLInputElement).value)}
+                class="ai-owner"
-            />
+                value={item.owner ?? ""}
-            <label class="remind" title={t("summary.reminder_title")}>
+                oninput={(e) => (item.owner = (e.target as HTMLInputElement).value || null)}
-              <input type="checkbox" bind:checked={item.reminder_set} disabled={!item.due_at} />
+                placeholder={t("summary.owner")}
-              <Bell size={14} aria-hidden="true" />
+                aria-label={t("summary.owner")}
-            </label>
+              />
-            <button
+              <input
-              class="ai-del"
+                type="date"
-              onclick={() => removeActionItem(i)}
+                class="due-date"
-              title={t("summary.ai_delete_title")}
+                aria-label={t("summary.due_date")}
-              aria-label={t("summary.ai_delete_aria")}
+                value={dueDateInput(item.due_at)}
-            >
+                onchange={(e) => onDueDateChange(item, (e.target as HTMLInputElement).value)}
-              <X size={14} aria-hidden="true" />
+              />
-            </button>
+              <label class="remind" title={t("summary.reminder_title")}>
+                <input type="checkbox" bind:checked={item.reminder_set} disabled={!item.due_at} />
+                <Bell size={14} aria-hidden="true" />
+              </label>
+            </div>
           </li>
         {/each}
       </ul>
@@ -1080,9 +1087,9 @@
   }
   ul.action-items li {
     display: flex;
-    align-items: center;
+    flex-direction: column;
     gap: 0.35rem;
-    padding: 0.25rem 0;
+    padding: 0.45rem 0;
     border-bottom: 1px solid var(--border);
   }
   ul.action-items label {
@@ -1090,14 +1097,28 @@
     align-items: center;
     gap: 0.4rem;
   }
+  .ai-main {
+    display: flex;
+    align-items: center;
+    gap: 0.35rem;
+  }
   .ai-text {
     flex: 1;
     min-width: 0;
-    font-size: 0.82rem;
+    font-size: 0.85rem;
+  }
+  /* Meta row indented under the text (past the confirm checkbox), wrapping
+     rather than crushing its inputs when the pane is narrow. */
+  .ai-meta {
+    display: flex;
+    align-items: center;
+    flex-wrap: wrap;
+    gap: 0.35rem;
+    padding-left: 1.4rem;
   }
   .ai-owner {
-    flex: none;
+    flex: 1;
-    width: 5rem;
+    min-width: 5rem;
     font-size: 0.75rem;
   }
   .ai-del {

src/lib/views/TranscriptNotes.svelte

@@ -6,19 +6,27 @@
   import { meetings } from "../stores/meetings.svelte";
   import { settings } from "../stores/settings.svelte";
   import { player } from "../stores/player.svelte";
-  import { api, type SpeakerInfo } from "../api";
+  import { api, errorMessage, type SpeakerInfo } from "../api";
   import { t } from "../i18n/index.svelte";
   import { renderMarkdown } from "../markdown";
   import { save, open } from "@tauri-apps/plugin-dialog";
   import { layout, clamp } from "../stores/layout.svelte";
+  import { imports } from "../stores/imports.svelte";
   import Splitter from "../components/Splitter.svelte";
+  import ImportTracker from "../components/ImportTracker.svelte";
   import {
     Bold,
     Italic,
     Heading1,
     Heading2,
+    Heading3,
     List,
+    ListOrdered,
     ListChecks,
+    Quote,
+    Minus,
+    Sparkles,
+    Undo2,
     FileDown,
     FileText,
     FolderOutput,
@@ -66,6 +74,10 @@
   let editorEl: HTMLTextAreaElement | undefined = $state();
   let saveTimer: ReturnType<typeof setTimeout> | undefined;
   let loadedForId: string | null = null;
+  // The server copy the buffer was last synced against — lets the effect below
+  // tell a server-side notes change (speaker rename, reprocess) apart from the
+  // user's own unsaved edits.
+  let lastServerNotes: string | null = null;
 
   // Transcript/notes split (FR-UX-1): resizable (drag the Splitter) and
   // each side independently hideable, shared across the finalized-meeting
@@ -103,14 +115,23 @@
     selectedSegmentMs = null;
   });
 
-  // Sync the editor buffer whenever a different meeting is selected.
+  // Sync the editor buffer whenever a different meeting is selected — and when
+  // the *server* copy of the same meeting's notes changes underneath us (a
+  // speaker rename rewrites notes.md's dialogue tags, reprocess regenerates it).
+  // A buffer with unsaved local edits is never clobbered: it only adopts the
+  // server copy when it still equals the last-synced one.
   $effect(() => {
     const m = meetings.selected;
     if (m && m.id !== loadedForId) {
       notesText = m.notes_markdown;
+      lastServerNotes = m.notes_markdown;
       loadedForId = m.id;
+    } else if (m && m.id === loadedForId && m.notes_markdown !== lastServerNotes) {
+      if (notesText === lastServerNotes) notesText = m.notes_markdown;
+      lastServerNotes = m.notes_markdown;
     } else if (!m) {
       loadedForId = null;
+      lastServerNotes = null;
     }
   });
 
@@ -158,6 +179,70 @@
     scheduleSave();
   }
 
+  // Slash commands: typing "/todo" (etc.) at the start of a line and pressing
+  // Space/Enter swaps it for the matching Markdown prefix. Reuses the same
+  // line-prefix model as the toolbar buttons — no rich inline menu.
+  // ponytail: line-prefix slash only; add a picker popover if users ask.
+  const SLASH_COMMANDS: Record<string, string> = {
+    h1: "# ",
+    h2: "## ",
+    h3: "### ",
+    todo: "- [ ] ",
+    bullet: "- ",
+    num: "1. ",
+    quote: "> ",
+    divider: "---\n",
+  };
+  function handleNotesKeydown(e: KeyboardEvent) {
+    if (e.key !== "Enter" && e.key !== " ") return;
+    const el = editorEl;
+    if (!el) return;
+    const { selectionStart: s, value } = el;
+    const lineStart = value.lastIndexOf("\n", s - 1) + 1;
+    const match = /^\/(\w+)$/.exec(value.slice(lineStart, s));
+    if (!match) return;
+    const prefix = SLASH_COMMANDS[match[1].toLowerCase()];
+    if (prefix === undefined) return;
+    e.preventDefault();
+    const head = value.slice(0, lineStart) + prefix;
+    notesText = head + value.slice(s);
+    queueMicrotask(() => {
+      el.focus();
+      el.selectionStart = el.selectionEnd = head.length;
+    });
+    scheduleSave();
+  }
+
+  // AI-enhance (Granola-style): expand the user's rough notes into structured
+  // Markdown grounded in the transcript, via the configured LlmProvider (local
+  // by default, no new egress). Keeps a one-step Undo so we never silently lose
+  // what the user typed.
+  let enhancing = $state(false);
+  let enhanceError = $state<string | null>(null);
+  let notesBeforeEnhance = $state<string | null>(null);
+  async function enhanceNotes() {
+    const m = meetings.selected;
+    if (!m || enhancing) return;
+    enhancing = true;
+    enhanceError = null;
+    try {
+      const enhanced = await api.enhanceNotes(m.id, notesText);
+      notesBeforeEnhance = notesText;
+      notesText = enhanced;
+      scheduleSave();
+    } catch (e) {
+      enhanceError = errorMessage(e);
+    } finally {
+      enhancing = false;
+    }
+  }
+  function undoEnhance() {
+    if (notesBeforeEnhance === null) return;
+    notesText = notesBeforeEnhance;
+    notesBeforeEnhance = null;
+    scheduleSave();
+  }
+
   async function exportMd() {
     const m = meetings.selected;
     if (!m) return;
@@ -247,6 +332,15 @@
     reprocessing = true;
     try {
       await meetings.reprocess(m.id, reprocessModel, reprocessLanguage || undefined);
+      // Re-transcribe rebuilds notes.md server-side (merging saved manual notes),
+      // but the meeting stays selected (same id), so the buffer-sync $effect —
+      // which only fires on an id change — won't pick it up. Resync explicitly so
+      // the notes pane updates in place instead of only after a restart.
+      const updated = meetings.selected;
+      if (updated && updated.id === m.id) {
+        notesText = updated.notes_markdown;
+        loadedForId = updated.id;
+      }
     } finally {
       reprocessing = false;
     }
@@ -262,14 +356,23 @@
       onchange={onTitleChange}
       aria-label={t("transcript.title_aria")}
     />
+    {#if m.status === "transcribing" || imports.get(m.id)}
+      <div class="import-strip"><ImportTracker meetingId={m.id} /></div>
+    {:else if m.model_used}
+      <p class="engine-meta" title={t("transcript.transcribed_with")}>
+        {t("transcript.transcribed_with")}
+        <strong>{m.model_used}</strong>{#if m.backend_used}
+          · {m.backend_used}{/if}
+      </p>
+    {/if}
     <div
       class="split"
       style="grid-template-columns: {splitColumns()};"
       bind:clientWidth={splitWidth}
     >
-      <!-- svelte-ignore a11y_no_static_element_interactions -- wheel/touchmove
+      <!-- wheel/touchmove here only note "the user scrolled by hand" to pause
-           here only note "the user scrolled by hand" to pause playback
+           playback auto-scroll; the pane isn't an interactive control. -->
-           auto-scroll; the pane isn't an interactive control. -->
+      <!-- svelte-ignore a11y_no_static_element_interactions -->
       <div
         class="pane transcript"
         class:collapsed={layout.transcriptCollapsed}
@@ -423,6 +526,43 @@
             >
               <ListChecks size={14} aria-hidden="true" />
             </button>
+            <button
+              onclick={() => insertLinePrefix("### ")}
+              title={t("notes.h3")}
+              aria-label={t("notes.h3")}
+            >
+              <Heading3 size={14} aria-hidden="true" />
+            </button>
+            <button
+              onclick={() => insertLinePrefix("1. ")}
+              title={t("notes.numbered")}
+              aria-label={t("notes.numbered")}
+            >
+              <ListOrdered size={14} aria-hidden="true" />
+            </button>
+            <button
+              onclick={() => insertLinePrefix("> ")}
+              title={t("notes.quote")}
+              aria-label={t("notes.quote")}
+            >
+              <Quote size={14} aria-hidden="true" />
+            </button>
+            <button
+              onclick={() => insertLinePrefix("---\n")}
+              title={t("notes.divider")}
+              aria-label={t("notes.divider")}
+            >
+              <Minus size={14} aria-hidden="true" />
+            </button>
+            <button
+              class="enhance"
+              onclick={enhanceNotes}
+              disabled={enhancing}
+              title={t("notes.enhance_title")}
+            >
+              <Sparkles size={14} aria-hidden="true" class={enhancing ? "spin" : ""} />
+              {enhancing ? t("notes.enhancing") : t("notes.enhance")}
+            </button>
             <button
               class="toggle"
               onclick={() => (notesPreview = !notesPreview)}
@@ -459,6 +599,17 @@
               Obsidian
             </button>
           </div>
+          {#if enhanceError}
+            <p class="enhance-bar error" role="alert">{enhanceError}</p>
+          {:else if notesBeforeEnhance !== null}
+            <div class="enhance-bar">
+              <span>{t("notes.enhanced_note")}</span>
+              <button class="link" onclick={undoEnhance}>
+                <Undo2 size={13} aria-hidden="true" />
+                {t("notes.undo_enhance")}
+              </button>
+            </div>
+          {/if}
           <div class="editor-preview">
             {#if notesPreview}
               <!-- eslint-disable-next-line svelte/no-at-html-tags -- sanitized via renderMarkdown() -->
@@ -468,6 +619,7 @@
                 bind:this={editorEl}
                 bind:value={notesText}
                 oninput={scheduleSave}
+                onkeydown={handleNotesKeydown}
                 placeholder={t("notes.placeholder")}
               ></textarea>
             {/if}
@@ -524,7 +676,7 @@
                   onclick={() => (selectedSegmentMs = open ? null : s.start_ms)}
                 >
                   <span class="ts">{fmtTs(s.start_ms)}</span>
-                  <strong>{speakerName(s.speaker)}:</strong>
+                  <strong>{speakerName(s.speaker, recording.speakers)}:</strong>
                   {s.text}
                   {#if hasNote}
                     <span class="note-badge" title={t("transcript.has_note")}>📝</span>
@@ -610,6 +762,21 @@
     background: var(--border);
     outline: none;
   }
+  .import-strip {
+    flex: none;
+    padding: 0 1rem 0.5rem;
+  }
+  .engine-meta {
+    flex: none;
+    margin: 0;
+    padding: 0 1rem 0.4rem;
+    font-size: 0.75rem;
+    color: var(--muted);
+  }
+  .engine-meta strong {
+    font-weight: 600;
+    color: var(--fg);
+  }
   .pad {
     padding: 1rem;
     max-width: 760px;
@@ -842,6 +1009,40 @@
   .toolbar .spacer {
     flex: 1;
   }
+  .toolbar .enhance {
+    color: var(--accent);
+    border-color: color-mix(in srgb, var(--accent) 40%, var(--border));
+    font-weight: 600;
+  }
+  .toolbar .enhance:hover:not(:disabled) {
+    background: color-mix(in srgb, var(--accent) 12%, var(--bg));
+  }
+  .toolbar .enhance:disabled {
+    opacity: 0.6;
+    cursor: default;
+  }
+  .enhance-bar {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    margin-bottom: 0.5rem;
+    font-size: 0.8rem;
+    color: var(--muted);
+  }
+  .enhance-bar.error {
+    color: var(--danger);
+  }
+  .enhance-bar .link {
+    display: inline-flex;
+    align-items: center;
+    gap: 0.25rem;
+    background: none;
+    border: none;
+    color: var(--accent);
+    cursor: pointer;
+    font-size: 0.8rem;
+    padding: 0;
+  }
 
   .editor-preview {
     height: calc(100% - 2.5rem);