diff --git a/README.md b/README.md
index a691cc8..d62568f 100644
--- a/README.md
+++ b/README.md
@@ -1,8 +1,122 @@
-# Heard
+
+
+
-A macOS menu bar app that automatically detects Microsoft Teams meetings, records dual-track audio (app + mic), and produces on-device transcripts with speaker diarization. Also includes a real-time dictation mode that types transcribed speech into any focused text field.
+Heard
-No cloud, no LLM, no external APIs — everything runs on-device on Apple Silicon.
+
+ Stop taking meeting notes. Get them automatically.
+ Heard is a quiet macOS menu bar app that auto-detects your Microsoft Teams meetings, records them, and turns them into clean, speaker-labeled Markdown transcripts — fully on-device.
+
+
+
+ Install ·
+ Why Heard ·
+ A look inside ·
+ Features ·
+ Requirements
+
+
+
+
+
+
+---
+
+## Why Heard?
+
+**No cloud. No subscription. No sending your meetings to anyone's servers.**
+Every byte of audio, every transcript, every speaker embedding lives on your Mac. Heard never makes a network call to a transcription service — because there is no transcription service. It all runs locally on Apple Silicon.
+
+- **Zero‑click recording** — As soon as a Teams meeting starts, Heard starts. When the meeting ends, you have a transcript waiting in your Documents folder.
+- **Speaker-labeled transcripts** — Diarization separates each voice; once you name a speaker, Heard remembers them across every future meeting.
+- **Dual‑track audio** — Heard records the system audio (the people on the call) and your microphone as separate streams, then merges them after transcription. The result is dramatically cleaner than recording a single mixed track.
+- **Custom vocabulary** — Add product names, acronyms, or jargon and they'll come through correctly in both meeting transcripts and live dictation.
+- **Real‑time dictation, anywhere** — Press a hotkey and Heard types what you say into whatever app is focused — Notes, Slack, your IDE, anywhere.
+- **In‑meeting notes** — A second hotkey opens a quick composer. Your typed note is timestamped to the moment you started writing it and merged into the transcript chronologically.
+- **Built for macOS** — A native SwiftUI menu bar app with a calm "Paper" palette. No Electron. No tray icon misery.
+
+### How it stacks up
+
+| | Heard | Cloud meeting recorders |
+|------------------------------|-----------------------------|---------------------------------|
+| Audio leaves your Mac | Never | Always |
+| Works offline | Yes | No |
+| Subscription | Free, open source | $$$ / month |
+| Needs a meeting bot to join | No — runs on your machine | Yes |
+| Speaker memory across calls | Yes, on-device | Sometimes |
+| Transcribes in real time as you dictate | Yes | Different product |
+| Custom vocabulary | Yes | Sometimes |
+
+---
+
+## A look inside
+
+### Quietly lives in your menu bar
+Heard sits in the top bar and changes its icon to tell you what it's doing — pulsing while recording, a waveform while transcribing, a calm tape icon while it waits for your next call.
+
+
+
+
+
+
+
+### Outputs a clean, speaker-labeled Markdown transcript
+Timestamps, merged speaker blocks, and your own typed notes interleaved at the moment you wrote them.
+
+
+
+
+
+### Dictates into any app, in real time
+Hold `⌃⇧D` (or rebind it) and start talking. Words appear as soon as the sliding-window ASR confirms them, with filler words and "uhs" stripped automatically.
+
+
+
+
+
+### A native settings panel built for macOS
+Custom vocabulary, model management, speaker library, and permission status all live in a single 880 × 600 window.
+
+
+
+
+
+
+
+
+
+
+
+### Names new speakers with one playable clip each
+The first time Heard hears someone new, it surfaces a ~10-second sample of their clearest speech and asks you who they are. After that, it remembers them.
+
+
+
+
+
+---
+
+## Install
+
+**Homebrew (recommended):**
+
+```bash
+brew tap execsumo/heard
+brew install --cask heard
+```
+
+**Direct download:** Grab the latest `Heard-x.x.x.dmg` from [GitHub Releases](https://github.com/execsumo/Heard/releases), open it, and drag Heard to `/Applications`.
+
+After launching, grant Microphone (always required) and Accessibility (required for dictation, recommended for richer meeting metadata). Heard will guide you to the right System Settings pane from inside the app.
+
+## Requirements
+
+- macOS 15.0+
+- Apple Silicon (FluidAudio CoreML/ANE models are ARM-only)
+- [FluidAudio](https://github.com/FluidInference/FluidAudio) 0.14.3+ (resolved automatically via SPM)
+
+---
## Features
@@ -45,23 +159,6 @@ No cloud, no LLM, no external APIs — everything runs on-device on Apple Silico
- **Settings window** — 880×600, five tabs: **General** (launch at login, auto-watch, developer mode, custom vocabulary, output folder, permissions), **Dictation** (enable, push-to-talk, hotkey recorder with validation feedback, model keep-alive slider, live status, show dictation HUD checkbox), **Models** (download cards with progress, pipeline keep-alive, "Unload All Models"), **Speakers** (your name, inline rename, merge, delete, search, sort, with prompts to retroactively update all past transcripts), **About**.
- **Name Speakers window** — 560×520 standalone scene with audio-clip playback per candidate, roster-suggestion hints, save/skip, and a 120 s countdown.
-## Installation
-
-**Homebrew (recommended):**
-
-```bash
-brew tap execsumo/heard
-brew install --cask heard
-```
-
-**Direct download:** Grab the latest `Heard-x.x.x.dmg` from [GitHub Releases](https://github.com/execsumo/Heard/releases), open it, and drag Heard to `/Applications`.
-
-## Requirements
-
-- macOS 15.0+
-- Apple Silicon (FluidAudio CoreML/ANE models are ARM-only)
-- [FluidAudio](https://github.com/FluidInference/FluidAudio) 0.14.3+ (resolved automatically via SPM)
-
## Build & Run
```bash
@@ -236,3 +333,4 @@ After a notarized DMG is published to GitHub Releases, update `Casks/heard.rb` i
- [`handoff.md`](./handoff.md) — Current implementation status and known issues
- [`ROADMAP.md`](./ROADMAP.md) — Planned improvements and stretch ideas
- [`CLAUDE.md`](./CLAUDE.md) — Working rules for AI assistants
+- [`docs/screenshots/`](./docs/screenshots) — Where the README screenshots live, and how to capture new ones
diff --git a/docs/screenshots/README.md b/docs/screenshots/README.md
new file mode 100644
index 0000000..70c6b52
--- /dev/null
+++ b/docs/screenshots/README.md
@@ -0,0 +1,27 @@
+# Screenshots
+
+This folder holds the screenshots referenced from the top-level `README.md`. Replace any missing PNG with a real capture; keep the filenames exactly as listed so the README links don't break.
+
+All shots should be taken on a Retina display, 2x resolution, with the app's "Paper" light palette. Crop tightly to the surface — no surrounding desktop chrome unless noted.
+
+| File | What to capture | Suggested size |
+|---|---|---|
+| `hero.png` | Menu bar dropdown in `Recording` state, with the elapsed timer visible. Optional: composite over a soft Paper-palette backdrop. | ~1200 × 800 |
+| `menubar-dropdown.png` | Menu bar dropdown in `Watching` state, showing 2–3 Recent Meetings. | 268 × natural |
+| `recording.png` | Menu bar dropdown in `Recording` state with the dark `#2E3338` header strip and elapsed time. | 268 × natural |
+| `transcript.png` | A finished Markdown transcript open in a Markdown viewer (e.g. Obsidian, Marked 2, or VS Code preview). Speaker labels, timestamps, and at least one `[mm:ss] _**Note from …:**_` line should be visible. | ~1000 × 700 |
+| `dictation-hud.png` | The floating "Dictating" pill at the bottom of the screen, with a focused text field above it showing live transcribed text. | ~900 × 500 |
+| `settings-general.png` | Settings → General tab. Show the Behavior, Permissions, and Custom vocabulary cards. | 880 × 600 |
+| `settings-models.png` | Settings → Models tab with the dark hero card and a couple of model rows. | 880 × 600 |
+| `settings-speakers.png` | Settings → Speakers tab with a populated list, search visible. | 880 × 600 |
+| `name-speakers.png` | Speaker naming window with two candidate cards (one playing, one idle) and the auto-dismiss countdown. | 560 × 520 |
+
+## How to capture cleanly
+
+1. Launch the bundled app: `./scripts/bundle.sh && open build/Heard.app`.
+2. Enable Developer Mode (Settings → General) so the **Simulate Meeting** buttons appear.
+3. Resize the Settings window to its default 880 × 600 and snap window screenshots with `⌘⇧4 + space` then click the window.
+4. For the menu bar dropdown, use `⌘⇧5` → "Capture Selected Portion" and drag a rectangle around the panel.
+5. Save PNGs into this folder with the exact filenames above.
+
+Hi-DPI PNGs render crisply on GitHub; avoid JPEG.