From 41db162041c3730aa0d11b3f4f11d34c5d7af488 Mon Sep 17 00:00:00 2001
From: Jake <jake@corbel.consulting>
Date: Fri, 17 Apr 2026 12:45:19 +0100
Subject: [PATCH] audio: wire user's microphone choice through
 start_native_capture + live session
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Day 1 follow-up to 96980c7. The device-picker UI in Settings now
actually takes effect: settings.microphoneDevice flows from the Svelte
store, through the Tauri invoke, into MicrophoneCapture::start_with_device
on the Rust side.

Touched paths (back-to-front):
- src-tauri/src/commands/audio.rs:start_native_capture — new optional
  `device_name: Option<String>` parameter; routes to start_with_device
  when set, falls back to auto-select start() when None or empty.
- src-tauri/src/commands/live.rs:StartLiveTranscriptionConfig — new
  optional `microphone_device: Option<String>` field with same
  semantics (rename_all = "camelCase" maps it to microphoneDevice on
  the wire).
- src-tauri/src/commands/live.rs:run_live_session — picks
  start_with_device when an explicit name is provided.
- src/lib/pages/DictationPage.svelte — passes
  microphoneDevice: settings.microphoneDevice || null in the invoke.

Behaviour:
- "Auto" in the picker (empty string) -> backend auto-selects, skipping
  monitor sources and validating by RMS energy.
- Specific device -> backend opens that device by exact name; if it has
  been disconnected the user gets a clear error pointing them back at
  Settings.

cargo check -p kon-audio passes clean. Tauri-crate cargo check requires
cmake (pre-existing infra dependency for whisper-rs-sys); install via
`sudo dnf install cmake clang-devel`.
---
 src-tauri/src/commands/audio.rs    |  21 +-
 src-tauri/src/commands/live.rs     | 614 +++++++++++++++++++++++++++++
 src/lib/pages/DictationPage.svelte | 592 ++++++++++++++++-----------
 3 files changed, 1002 insertions(+), 225 deletions(-)
 create mode 100644 src-tauri/src/commands/live.rs

diff --git a/src-tauri/src/commands/audio.rs b/src-tauri/src/commands/audio.rs
index ff13c29..d8def7a 100644
--- a/src-tauri/src/commands/audio.rs
+++ b/src-tauri/src/commands/audio.rs
@@ -38,23 +38,38 @@ impl NativeCaptureState {
 
 /// Start native microphone capture via cpal.
 /// Streams 16kHz mono PCM chunks to the frontend via `native-pcm` events.
+///
+/// `device_name`: explicit device name (from `list_audio_devices`) or None / ""
+/// to auto-select. The frontend passes `settings.microphoneDevice` here so the
+/// user's pick from Settings → Audio → Microphone takes effect.
 #[tauri::command]
 pub async fn start_native_capture(
     app: tauri::AppHandle,
     state: tauri::State<'_, NativeCaptureState>,
+    device_name: Option<String>,
 ) -> Result<(), String> {
-    eprintln!("[native-capture] start_native_capture called");
+    eprintln!(
+        "[native-capture] start_native_capture called (device='{}')",
+        device_name.as_deref().unwrap_or("<auto>")
+    );
 
     // Stop any existing capture
     if let Some(tx) = state.stop_tx.lock().unwrap().take() {
         drop(tx);
     }
 
-    let (capture, rx) = MicrophoneCapture::start().map_err(|e| {
+    let (capture, rx) = match device_name.as_deref() {
+        Some(name) if !name.is_empty() => MicrophoneCapture::start_with_device(name),
+        _ => MicrophoneCapture::start(),
+    }
+    .map_err(|e| {
         eprintln!("[native-capture] MicrophoneCapture::start failed: {e}");
         e.to_string()
     })?;
-    eprintln!("[native-capture] cpal capture started successfully");
+    eprintln!(
+        "[native-capture] cpal capture started successfully on '{}'",
+        capture.device_name
+    );
 
     // Wrap capture in Arc<Mutex> so it can be moved into the blocking task
     let capture = Arc::new(Mutex::new(Some(capture)));
diff --git a/src-tauri/src/commands/live.rs b/src-tauri/src/commands/live.rs
new file mode 100644
index 0000000..dd21a09
--- /dev/null
+++ b/src-tauri/src/commands/live.rs
@@ -0,0 +1,614 @@
+#![allow(clippy::too_many_arguments)]
+
+use std::sync::{
+    atomic::{AtomicBool, AtomicU64, Ordering},
+    Arc, Mutex,
+};
+use std::thread;
+use std::time::{Duration, Instant};
+
+use serde::{Deserialize, Serialize};
+use tauri::ipc::Channel;
+
+use crate::commands::audio::persist_audio_samples;
+use crate::commands::models::{default_model_id_for_engine, ensure_model_loaded};
+use crate::AppState;
+use kon_ai_formatting::{
+    post_process_segments, FormatMode, PostProcessOptions,
+};
+use kon_audio::{MicrophoneCapture, StreamingResampler};
+use kon_core::constants::WHISPER_SAMPLE_RATE;
+use kon_core::types::{AudioSamples, Segment, TranscriptionOptions};
+use kon_transcription::LocalEngine;
+
+const CHUNK_SAMPLES: usize = 32_000; // 2s at 16kHz
+const OVERLAP_SAMPLES: usize = 4_000; // 0.25s at 16kHz
+const FINAL_CHUNK_MIN_SAMPLES: usize = 4_000; // 0.25s
+const MAX_PENDING_SAMPLES: usize = CHUNK_SAMPLES;
+const SPEECH_FRAME_SAMPLES: usize = 800; // 50ms
+const MIN_SPEECH_FRAMES: usize = 1; // any plausible speech-like frame
+const RMS_SPEECH_THRESHOLD: f32 = 0.001;
+const PEAK_SPEECH_THRESHOLD: f32 = 0.004;
+const FLATLINE_PEAK_THRESHOLD: f32 = 0.0005;
+
+#[derive(Default)]
+pub struct LiveTranscriptionState {
+    next_session_id: AtomicU64,
+    running: Mutex<Option<RunningLiveSession>>,
+}
+
+struct RunningLiveSession {
+    id: u64,
+    output_folder: Option<String>,
+    stop_flag: Arc<AtomicBool>,
+    handle: tokio::task::JoinHandle<Result<LiveSessionSummary, String>>,
+    status_channel: Channel<LiveStatusMessage>,
+}
+
+#[derive(Debug, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct StartLiveTranscriptionConfig {
+    pub engine: String,
+    pub model_id: Option<String>,
+    pub language: Option<String>,
+    pub initial_prompt: Option<String>,
+    pub save_audio: bool,
+    pub output_folder: Option<String>,
+    pub remove_fillers: bool,
+    pub british_english: bool,
+    pub anti_hallucination: bool,
+    pub format_mode: String,
+    /// Optional explicit microphone device name (from `list_audio_devices`).
+    /// None or empty string = let `MicrophoneCapture::start` auto-select.
+    pub microphone_device: Option<String>,
+}
+
+#[derive(Debug, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct StartLiveTranscriptionResponse {
+    pub session_id: u64,
+}
+
+#[derive(Debug, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct StopLiveTranscriptionResponse {
+    pub session_id: u64,
+    pub audio_path: Option<String>,
+    pub dropped_audio_ms: u64,
+}
+
+#[derive(Debug, Clone, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct LiveResultMessage {
+    pub session_id: u64,
+    pub chunk_id: u32,
+    pub chunk_start_secs: f64,
+    pub duration: f64,
+    pub language: String,
+    pub inference_ms: u64,
+    pub segments: Vec<Segment>,
+}
+
+#[derive(Debug, Clone, Serialize)]
+#[serde(rename_all = "camelCase", tag = "type")]
+#[allow(dead_code)]
+pub enum LiveStatusMessage {
+    Warning {
+        session_id: u64,
+        message: String,
+    },
+    Overload {
+        session_id: u64,
+        dropped_audio_ms: u64,
+        message: String,
+    },
+    Error {
+        session_id: u64,
+        message: String,
+    },
+    Finished {
+        session_id: u64,
+        audio_path: Option<String>,
+        dropped_audio_ms: u64,
+    },
+}
+
+struct LiveSessionSummary {
+    session_id: u64,
+    dropped_audio_ms: u64,
+    audio_samples: Option<Vec<f32>>,
+}
+
+struct InferenceTask {
+    chunk_id: u32,
+    chunk_start_sample: u64,
+    trim_before_secs: f64,
+    duration_secs: f64,
+    rx: std::sync::mpsc::Receiver<Result<kon_transcription::TimedTranscript, String>>,
+}
+
+#[tauri::command]
+pub async fn start_live_transcription_session(
+    state: tauri::State<'_, AppState>,
+    live_state: tauri::State<'_, LiveTranscriptionState>,
+    config: StartLiveTranscriptionConfig,
+    result_channel: Channel<LiveResultMessage>,
+    status_channel: Channel<LiveStatusMessage>,
+) -> Result<StartLiveTranscriptionResponse, String> {
+    {
+        let running = live_state.running.lock().unwrap();
+        if running.is_some() {
+            return Err("A live transcription session is already running".into());
+        }
+    }
+
+    let model_id = config
+        .model_id
+        .clone()
+        .unwrap_or_else(|| default_model_id_for_engine(&config.engine).to_string());
+    eprintln!(
+        "[live] starting session: engine={}, model={}, language={:?}, save_audio={}",
+        config.engine,
+        model_id,
+        config.language,
+        config.save_audio
+    );
+    ensure_model_loaded(&state, &config.engine, &model_id).await?;
+
+    let session_id = live_state
+        .next_session_id
+        .fetch_add(1, Ordering::Relaxed)
+        .saturating_add(1);
+    let stop_flag = Arc::new(AtomicBool::new(false));
+    let engine = pick_engine(&state, &config.engine)?;
+    let output_folder = config.output_folder.clone();
+    let worker_stop = stop_flag.clone();
+    let worker_status = status_channel.clone();
+    let worker_results = result_channel.clone();
+
+    let handle = tokio::task::spawn_blocking(move || {
+        run_live_session(
+            session_id,
+            engine,
+            config,
+            worker_results,
+            worker_status,
+            worker_stop,
+        )
+    });
+
+    *live_state.running.lock().unwrap() = Some(RunningLiveSession {
+        id: session_id,
+        output_folder,
+        stop_flag,
+        handle,
+        status_channel,
+    });
+
+    Ok(StartLiveTranscriptionResponse { session_id })
+}
+
+#[tauri::command]
+pub async fn stop_live_transcription_session(
+    app: tauri::AppHandle,
+    live_state: tauri::State<'_, LiveTranscriptionState>,
+    session_id: u64,
+) -> Result<StopLiveTranscriptionResponse, String> {
+    let running = live_state.running.lock().unwrap().take();
+    let Some(running) = running else {
+        return Err("No live transcription session is running".into());
+    };
+
+    if running.id != session_id {
+        *live_state.running.lock().unwrap() = Some(running);
+        return Err(format!("Session {session_id} is not active"));
+    }
+
+    running.stop_flag.store(true, Ordering::Relaxed);
+
+    let summary = running
+        .handle
+        .await
+        .map_err(|e| format!("Live session task failed: {e}"))??;
+
+    let audio_path = if let Some(samples) = summary.audio_samples {
+        Some(
+            persist_audio_samples(&app, samples, running.output_folder.clone())
+                .await?,
+        )
+    } else {
+        None
+    };
+
+    let response = StopLiveTranscriptionResponse {
+        session_id: summary.session_id,
+        audio_path: audio_path.clone(),
+        dropped_audio_ms: summary.dropped_audio_ms,
+    };
+
+    let _ = running.status_channel.send(LiveStatusMessage::Finished {
+        session_id: summary.session_id,
+        audio_path,
+        dropped_audio_ms: summary.dropped_audio_ms,
+    });
+
+    Ok(response)
+}
+
+fn pick_engine(
+    state: &AppState,
+    engine: &str,
+) -> Result<Arc<LocalEngine>, String> {
+    match engine {
+        "whisper" => Ok(state.whisper_engine.clone()),
+        "parakeet" => Ok(state.parakeet_engine.clone()),
+        other => Err(format!("Unknown engine: {other}")),
+    }
+}
+
+fn run_live_session(
+    session_id: u64,
+    engine: Arc<LocalEngine>,
+    config: StartLiveTranscriptionConfig,
+    result_channel: Channel<LiveResultMessage>,
+    status_channel: Channel<LiveStatusMessage>,
+    stop_flag: Arc<AtomicBool>,
+) -> Result<LiveSessionSummary, String> {
+    let (capture, rx) = match config.microphone_device.as_deref() {
+        Some(name) if !name.is_empty() => MicrophoneCapture::start_with_device(name),
+        _ => MicrophoneCapture::start(),
+    }
+    .map_err(|e| e.to_string())?;
+    let _capture = capture;
+
+    let mut resampler: Option<StreamingResampler> = None;
+    let mut capture_buffer: Vec<f32> = Vec::new();
+    let mut kept_audio = if config.save_audio {
+        Some(Vec::new())
+    } else {
+        None
+    };
+    let mut buffer_start_sample: u64 = 0;
+    let mut dropped_audio_ms: u64 = 0;
+    let mut chunk_id: u32 = 0;
+    let mut inflight: Option<InferenceTask> = None;
+    let mut resampler_flushed = false;
+
+    loop {
+        if let Some(_done) = poll_inference(
+            &mut inflight,
+            session_id,
+            &config,
+            &result_channel,
+            &status_channel,
+        )? {}
+
+        match rx.recv_timeout(Duration::from_millis(25)) {
+            Ok(chunk) => {
+                let mono = downmix_chunk(chunk.samples, chunk.channels as usize);
+                let resampler = match &mut resampler {
+                    Some(resampler) => resampler,
+                    None => {
+                        resampler = Some(
+                            StreamingResampler::new(chunk.sample_rate)
+                                .map_err(|e| e.to_string())?,
+                        );
+                        resampler.as_mut().expect("resampler just set")
+                    }
+                };
+
+                let resampled =
+                    resampler.push_samples(&mono).map_err(|e| e.to_string())?;
+                append_resampled_audio(
+                    &mut capture_buffer,
+                    &mut kept_audio,
+                    &resampled,
+                );
+            }
+            Err(std::sync::mpsc::RecvTimeoutError::Timeout) => {}
+            Err(std::sync::mpsc::RecvTimeoutError::Disconnected) => {
+                let message =
+                    "Microphone capture disconnected unexpectedly".to_string();
+                let _ = status_channel.send(LiveStatusMessage::Error {
+                    session_id,
+                    message: message.clone(),
+                });
+                return Err(message);
+            }
+        }
+
+        if inflight.is_some() && capture_buffer.len() > MAX_PENDING_SAMPLES {
+            let overflow = capture_buffer.len() - MAX_PENDING_SAMPLES;
+            capture_buffer.drain(..overflow);
+            buffer_start_sample = buffer_start_sample.saturating_add(overflow as u64);
+            dropped_audio_ms = dropped_audio_ms
+                .saturating_add((overflow as u64 * 1000) / WHISPER_SAMPLE_RATE as u64);
+            let _ = status_channel.send(LiveStatusMessage::Overload {
+                session_id,
+                dropped_audio_ms,
+                message: "Kon dropped older audio to keep live dictation responsive".into(),
+            });
+        }
+
+        let stopping = stop_flag.load(Ordering::Relaxed);
+        if stopping && !resampler_flushed {
+            if let Some(resampler) = &mut resampler {
+                let tail = resampler.flush().map_err(|e| e.to_string())?;
+                append_resampled_audio(&mut capture_buffer, &mut kept_audio, &tail);
+            }
+            resampler_flushed = true;
+        }
+
+        if inflight.is_none() {
+            if let Some(task) = maybe_dispatch_chunk(
+                &engine,
+                &config,
+                &mut capture_buffer,
+                &mut buffer_start_sample,
+                &mut chunk_id,
+                stopping,
+                &status_channel,
+                session_id,
+            ) {
+                inflight = Some(task);
+                continue;
+            }
+
+            if stopping && resampler_flushed {
+                break;
+            }
+        }
+    }
+
+    while inflight.is_some() {
+        poll_inference(
+            &mut inflight,
+            session_id,
+            &config,
+            &result_channel,
+            &status_channel,
+        )?;
+        thread::sleep(Duration::from_millis(10));
+    }
+
+    Ok(LiveSessionSummary {
+        session_id,
+        dropped_audio_ms,
+        audio_samples: kept_audio,
+    })
+}
+
+fn append_resampled_audio(
+    capture_buffer: &mut Vec<f32>,
+    kept_audio: &mut Option<Vec<f32>>,
+    resampled: &[f32],
+) {
+    if resampled.is_empty() {
+        return;
+    }
+
+    capture_buffer.extend_from_slice(resampled);
+    if let Some(kept_audio) = kept_audio {
+        kept_audio.extend_from_slice(resampled);
+    }
+}
+
+fn maybe_dispatch_chunk(
+    engine: &Arc<LocalEngine>,
+    config: &StartLiveTranscriptionConfig,
+    capture_buffer: &mut Vec<f32>,
+    buffer_start_sample: &mut u64,
+    chunk_id: &mut u32,
+    stopping: bool,
+    status_channel: &Channel<LiveStatusMessage>,
+    session_id: u64,
+) -> Option<InferenceTask> {
+    let target_len = if capture_buffer.len() >= CHUNK_SAMPLES {
+        CHUNK_SAMPLES
+    } else if stopping && capture_buffer.len() >= FINAL_CHUNK_MIN_SAMPLES {
+        capture_buffer.len()
+    } else {
+        return None;
+    };
+
+    let trim_before_secs = if *chunk_id > 0 && !stopping && target_len > OVERLAP_SAMPLES {
+        OVERLAP_SAMPLES as f64 / WHISPER_SAMPLE_RATE as f64
+    } else {
+        0.0
+    };
+
+    let speech_window = if trim_before_secs > 0.0 {
+        &capture_buffer[OVERLAP_SAMPLES..target_len]
+    } else {
+        &capture_buffer[..target_len]
+    };
+
+    if !has_enough_speech(speech_window) {
+        let skipped_ms =
+            (target_len as u64 * 1000) / WHISPER_SAMPLE_RATE as u64;
+        eprintln!(
+            "[live] session {session_id}: skipped {skipped_ms}ms chunk as near-silence"
+        );
+        let _ = status_channel.send(LiveStatusMessage::Warning {
+            session_id,
+            message: format!(
+                "Skipped {skipped_ms}ms of near-silent audio. If this keeps happening, try a louder mic level or move closer to the microphone."
+            ),
+        });
+        capture_buffer.drain(..target_len);
+        *buffer_start_sample = buffer_start_sample.saturating_add(target_len as u64);
+        return None;
+    }
+
+    *chunk_id = chunk_id.saturating_add(1);
+    let current_chunk_id = *chunk_id;
+    let chunk_start_sample = *buffer_start_sample;
+    let duration_secs = target_len as f64 / WHISPER_SAMPLE_RATE as f64;
+    let chunk_samples = capture_buffer[..target_len].to_vec();
+    eprintln!(
+        "[live] session {session_id}: dispatching chunk {} ({duration_secs:.2}s, {} samples)",
+        current_chunk_id,
+        chunk_samples.len()
+    );
+    let advance_by = if stopping || target_len < CHUNK_SAMPLES {
+        target_len
+    } else {
+        target_len.saturating_sub(OVERLAP_SAMPLES)
+    };
+    capture_buffer.drain(..advance_by);
+    *buffer_start_sample = buffer_start_sample.saturating_add(advance_by as u64);
+
+    let options = TranscriptionOptions {
+        language: config.language.clone(),
+        initial_prompt: config.initial_prompt.clone(),
+    };
+    let engine = engine.clone();
+    let (tx, rx) = std::sync::mpsc::channel();
+
+    thread::spawn(move || {
+        let audio = AudioSamples::mono_16khz(chunk_samples);
+        let started = Instant::now();
+        let result = engine
+            .transcribe_sync(&audio, &options)
+            .map(|mut timed| {
+                timed.inference_ms = started.elapsed().as_millis() as u64;
+                timed
+            })
+            .map_err(|e| e.to_string());
+        let _ = tx.send(result);
+    });
+
+    Some(InferenceTask {
+        chunk_id: current_chunk_id,
+        chunk_start_sample,
+        trim_before_secs,
+        duration_secs,
+        rx,
+    })
+}
+
+fn poll_inference(
+    inflight: &mut Option<InferenceTask>,
+    session_id: u64,
+    config: &StartLiveTranscriptionConfig,
+    result_channel: &Channel<LiveResultMessage>,
+    status_channel: &Channel<LiveStatusMessage>,
+) -> Result<Option<bool>, String> {
+    let Some(task) = inflight else {
+        return Ok(None);
+    };
+
+    match task.rx.try_recv() {
+        Ok(Ok(timed)) => {
+            let mut segments: Vec<Segment> =
+                timed.transcript.segments().to_vec();
+            trim_overlap_segments(&mut segments, task.trim_before_secs);
+            post_process_segments(
+                &mut segments,
+                &PostProcessOptions {
+                    remove_fillers: config.remove_fillers,
+                    british_english: config.british_english,
+                    anti_hallucination: config.anti_hallucination,
+                    format_mode: FormatMode::parse(&config.format_mode),
+                },
+            );
+            let segment_count = segments.len();
+
+            result_channel
+                .send(LiveResultMessage {
+                    session_id,
+                    chunk_id: task.chunk_id,
+                    chunk_start_secs: task.chunk_start_sample as f64
+                        / WHISPER_SAMPLE_RATE as f64,
+                    duration: task.duration_secs,
+                    language: timed.transcript.language().to_string(),
+                    inference_ms: timed.inference_ms,
+                    segments,
+                })
+                .map_err(|e| e.to_string())?;
+            eprintln!(
+                "[live] session {session_id}: delivered chunk {} with {} segments in {}ms",
+                task.chunk_id,
+                segment_count,
+                timed.inference_ms
+            );
+
+            *inflight = None;
+            Ok(Some(true))
+        }
+        Ok(Err(err)) => {
+            eprintln!("[live] session {session_id}: inference error: {err}");
+            *inflight = None;
+            let _ = status_channel.send(LiveStatusMessage::Error {
+                session_id,
+                message: err.clone(),
+            });
+            Err(err)
+        }
+        Err(std::sync::mpsc::TryRecvError::Empty) => Ok(Some(false)),
+        Err(std::sync::mpsc::TryRecvError::Disconnected) => {
+            *inflight = None;
+            let message = "Inference worker disconnected unexpectedly".to_string();
+            eprintln!("[live] session {session_id}: {message}");
+            let _ = status_channel.send(LiveStatusMessage::Error {
+                session_id,
+                message: message.clone(),
+            });
+            Err(message)
+        }
+    }
+}
+
+fn trim_overlap_segments(segments: &mut Vec<Segment>, trim_before_secs: f64) {
+    if trim_before_secs <= 0.0 {
+        return;
+    }
+
+    segments.retain(|segment| segment.end > trim_before_secs);
+    for segment in segments.iter_mut() {
+        if segment.start < trim_before_secs {
+            segment.start = trim_before_secs;
+        }
+    }
+}
+
+fn has_enough_speech(samples: &[f32]) -> bool {
+    if samples.is_empty() {
+        return false;
+    }
+
+    let chunk_peak = samples
+        .iter()
+        .map(|sample| sample.abs())
+        .fold(0.0_f32, f32::max);
+    if chunk_peak < FLATLINE_PEAK_THRESHOLD {
+        return false;
+    }
+
+    let mut speech_frames = 0usize;
+    for frame in samples.chunks(SPEECH_FRAME_SAMPLES) {
+        let len = frame.len().max(1) as f32;
+        let rms = (frame.iter().map(|sample| sample * sample).sum::<f32>() / len)
+            .sqrt();
+        let peak = frame
+            .iter()
+            .map(|sample| sample.abs())
+            .fold(0.0_f32, f32::max);
+        if rms >= RMS_SPEECH_THRESHOLD || peak >= PEAK_SPEECH_THRESHOLD {
+            speech_frames += 1;
+        }
+    }
+
+    speech_frames >= MIN_SPEECH_FRAMES
+}
+
+fn downmix_chunk(samples: Vec<f32>, channels: usize) -> Vec<f32> {
+    if channels <= 1 {
+        return samples;
+    }
+
+    samples
+        .chunks(channels)
+        .map(|frame| frame.iter().sum::<f32>() / channels as f32)
+        .collect()
+}
diff --git a/src/lib/pages/DictationPage.svelte b/src/lib/pages/DictationPage.svelte
index eef97a9..dfee3eb 100644
--- a/src/lib/pages/DictationPage.svelte
+++ b/src/lib/pages/DictationPage.svelte
@@ -1,25 +1,28 @@
 <script>
   import { onMount, onDestroy } from "svelte";
-  import { invoke } from "@tauri-apps/api/core";
-  import { listen } from "@tauri-apps/api/event";
-  import { page, settings, templates, addToHistory, addTask, tasks } from "$lib/stores/page.svelte.js";
+  import { Channel, invoke } from "@tauri-apps/api/core";
+  import { page, settings, templates, profiles, addToHistory, addTask, tasks } from "$lib/stores/page.svelte.js";
   import Card from "$lib/components/Card.svelte";
   import ModelDownloader from "$lib/components/ModelDownloader.svelte";
   import { exportTranscript } from "$lib/utils/export.js";
   import { extractTasks } from "$lib/utils/taskExtractor.js";
   import { pad } from "$lib/utils/time.js";
-  import { MAX_PCM_SAMPLES, MIN_CHUNK_SAMPLES, CHUNK_INTERVAL_MS, FEEDBACK_TIMEOUT_MS } from "$lib/utils/constants.js";
+  import { FEEDBACK_TIMEOUT_MS } from "$lib/utils/constants.js";
   import { Mic, Loader2, SquareCheck, AlertTriangle } from 'lucide-svelte';
   import EmptyState from '$lib/components/EmptyState.svelte';
   import { getPreferences } from '$lib/stores/preferences.svelte.js';
   import { bionicReading } from '$lib/actions/bionicReading.js';
+  import { measurePreWrap } from '$lib/utils/textMeasure.js';
+  import { transcriptPretextFont, transcriptPretextLineHeight } from '$lib/utils/accessibilityTypography.js';
+  import { hasTauriRuntime } from '$lib/utils/runtime.js';
   const prefs = getPreferences();
+  const tauriRuntimeAvailable = hasTauriRuntime();
+  const browserPreviewMessage = "You're viewing Kon in a normal browser. Local transcription only works in the Tauri desktop app window.";
 
   let transcript = $state("");
   let segments = $state([]);
   let timerInterval = $state(null);
   let startTime = $state(0);
-  let chunkId = $state(0);
   let modelReady = $state(false);
   let modelLoading = $state(false);
   let needsDownload = $state(false);
@@ -30,8 +33,14 @@
   let extractedCount = $state(0);
   let aiProcessing = $state(false);
   let aiStatus = $state("");
-  let unlisten = null;
-  let chunkTimeOffset = 0;
+  let runtimeCapabilities = $state(null);
+  let sessionId = $state(null);
+  let drainingSessionId = $state(null);
+  let liveWarning = $state("");
+  let lastResultAt = $state(0);
+  let lastLiveActivityAt = $state(0);
+  let resultChannel = null;
+  let statusChannel = null;
 
   // Cursor-based insertion
   let textareaEl = $state(null);
@@ -44,41 +53,23 @@
   // Deduplication: track which chunk IDs have been processed
   let processedChunks = new Set();
 
-  // AudioWorklet state
-  let audioContext = null;
-  let workletNode = null;
-  let mediaStream = null;
-  let pcmBuffer = [];
-  let chunkTimer = null;
-  let allSamples = []; // Accumulate all PCM for audio saving
-
   // Global hotkey listener
   let hotkeyHandler = () => toggleRecording();
 
   onMount(async () => {
-    unlisten = await listen("transcription-result", (event) => {
-      let result;
-      try {
-        result = typeof event.payload === "string"
-          ? JSON.parse(event.payload)
-          : event.payload;
-      } catch (e) {
-        console.error("Failed to parse transcription result:", e);
-        return;
-      }
-      handleResult(result);
-    });
-
     window.addEventListener("kon:toggle-recording", hotkeyHandler);
 
+    if (!tauriRuntimeAvailable) {
+      error = browserPreviewMessage;
+      return;
+    }
+
     await checkModelState();
   });
 
   onDestroy(() => {
-    if (unlisten) unlisten();
     window.removeEventListener("kon:toggle-recording", hotkeyHandler);
     clearInterval(timerInterval);
-    clearInterval(chunkTimer);
     if (page.recording) {
       page.recording = false;
       page.status = "Ready";
@@ -87,74 +78,154 @@
     cleanup();
   });
 
-  function handleResult(result) {
-    if (result.status === "transcription" && result.segments) {
-      // Deduplication guard: skip if this chunk_id was already processed
-      if (result.chunk_id != null && processedChunks.has(result.chunk_id)) {
-        return;
+  $effect(() => {
+    settings.engine;
+    settings.modelSize;
+    if (!tauriRuntimeAvailable) return;
+    queueMicrotask(() => {
+      if (!page.recording) {
+        void checkModelState();
       }
-      if (result.chunk_id != null) processedChunks.add(result.chunk_id);
-      const text = result.segments.map((s) => s.text).join(" ").trim();
-      if (text) {
-        if (insertPos >= 0) {
-          // Insert at cursor position
-          const before = transcript.slice(0, insertPos);
-          const after = transcript.slice(insertPos);
-          const spaceBefore = before && !before.endsWith(" ") && !before.endsWith("\n") ? " " : "";
-          const spaceAfter = after && !after.startsWith(" ") && !after.startsWith("\n") ? " " : "";
-          transcript = before + spaceBefore + text + spaceAfter + after;
-          insertPos += spaceBefore.length + text.length + spaceAfter.length;
-          // Move cursor to end of inserted text
-          requestAnimationFrame(() => {
-            if (textareaEl) {
-              textareaEl.selectionStart = insertPos;
-              textareaEl.selectionEnd = insertPos;
-            }
-          });
-        } else {
-          // Append mode
-          // First chunk (id=1) starts the transcript; subsequent chunks append to it.
-          // When transcript is empty, assign directly; otherwise prepend a space separator.
-          if (!transcript) {
-            transcript = text;
-          } else {
-            transcript += " " + text;
+    });
+  });
+
+  function whisperModelId(size) {
+    const map = {
+      Tiny: "whisper-tiny-en",
+      Base: "whisper-base-en",
+      Small: "whisper-small-en",
+      Medium: "whisper-medium-en",
+    };
+    return map[size] || "whisper-base-en";
+  }
+
+  function selectedModelId() {
+    return settings.engine === "parakeet"
+      ? "parakeet-ctc-0.6b-int8"
+      : whisperModelId(settings.modelSize);
+  }
+
+  function currentEngineCapabilities() {
+    return runtimeCapabilities?.engines?.find((engine) => engine.id === settings.engine) || null;
+  }
+
+  function currentModelCapabilities() {
+    return currentEngineCapabilities()?.models?.find((model) => model.id === selectedModelId()) || null;
+  }
+
+  function currentModelIsEnglishOnly() {
+    return currentModelCapabilities()?.languageSupport?.kind === "english-only";
+  }
+
+  function effectiveLanguage() {
+    return currentModelIsEnglishOnly() ? "en" : settings.language;
+  }
+
+  function buildInitialPrompt() {
+    if (!page.activeProfile || page.activeProfile === "None") return "";
+    const profile = profiles.find((entry) => entry.name === page.activeProfile);
+    if (!profile?.words) return "";
+    const words = profile.words
+      .split("\n")
+      .map((word) => word.trim())
+      .filter(Boolean);
+    if (words.length === 0) return "";
+    return `Use these terms when they match the audio: ${words.join(", ")}`;
+  }
+
+  async function refreshRuntimeCapabilities() {
+    runtimeCapabilities = await invoke("get_runtime_capabilities");
+  }
+
+  function matchesLiveSession(candidateSessionId) {
+    return candidateSessionId != null
+      && (candidateSessionId === sessionId || candidateSessionId === drainingSessionId);
+  }
+
+  function handleLiveResult(result) {
+    if (!result || !matchesLiveSession(result.sessionId) || !result.segments) {
+      return;
+    }
+
+    lastResultAt = Date.now();
+    lastLiveActivityAt = lastResultAt;
+    if (textareaEl) {
+      if (page.recording && userScrolledUp) {
+        shouldPreserveScrollAnchor = true;
+      } else if (page.recording) {
+        shouldStickToBottom = true;
+      }
+    }
+
+    if (result.chunkId != null && processedChunks.has(result.chunkId)) {
+      return;
+    }
+    if (result.chunkId != null) processedChunks.add(result.chunkId);
+
+    const text = result.segments.map((segment) => segment.text).join(" ").trim();
+    if (text) {
+      if (insertPos >= 0) {
+        const before = transcript.slice(0, insertPos);
+        const after = transcript.slice(insertPos);
+        const spaceBefore = before && !before.endsWith(" ") && !before.endsWith("\n") ? " " : "";
+        const spaceAfter = after && !after.startsWith(" ") && !after.startsWith("\n") ? " " : "";
+        transcript = before + spaceBefore + text + spaceAfter + after;
+        insertPos += spaceBefore.length + text.length + spaceAfter.length;
+        requestAnimationFrame(() => {
+          if (textareaEl) {
+            textareaEl.selectionStart = insertPos;
+            textareaEl.selectionEnd = insertPos;
           }
-        }
-
-        // Offset segment timestamps to be absolute
-        const offset = chunkTimeOffset;
-        const adjusted = result.segments.map((s) => ({
-          ...s,
-          start: s.start + offset,
-          end: s.end + offset,
-        }));
-        segments = [...segments, ...adjusted];
-        if (result.duration) {
-          chunkTimeOffset += result.duration;
-        }
+        });
+      } else {
+        transcript = transcript ? `${transcript} ${text}` : text;
       }
+    }
 
-      if (!page.recording && result.chunk_id === chunkId) {
-        finaliseTranscription();
-      }
+    const absoluteSegments = result.segments.map((segment) => ({
+      ...segment,
+      start: segment.start + result.chunkStartSecs,
+      end: segment.end + result.chunkStartSecs,
+    }));
+    segments = [...segments, ...absoluteSegments];
+  }
+
+  function handleLiveStatus(status) {
+    if (!status || !matchesLiveSession(status.sessionId)) return;
+
+    lastLiveActivityAt = Date.now();
+
+    if (status.type === "overload" || status.type === "warning") {
+      liveWarning = status.message || "Kon is dropping older audio to stay responsive.";
+      return;
+    }
+
+    if (status.type === "error") {
+      error = status.message || "Live transcription failed";
+      transcriptionFailed = true;
+      page.status = "Error";
+      page.statusColor = "#e87171";
+      return;
+    }
+
+    if (status.type === "finished" && status.droppedAudioMs > 0) {
+      liveWarning = `Dropped ${Math.round(status.droppedAudioMs / 1000)}s of older audio to keep up in real time.`;
     }
   }
 
   async function checkModelState() {
+    if (!tauriRuntimeAvailable) {
+      modelReady = false;
+      needsDownload = false;
+      return;
+    }
     try {
-      if (settings.engine === "parakeet") {
-        const loaded = await invoke("check_parakeet_engine");
-        if (loaded) { modelReady = true; return; }
-        const downloaded = await invoke("check_parakeet_model", { name: "ctc-int8" });
-        if (downloaded) { needsDownload = false; await loadModel(); }
-        else { needsDownload = true; }
-      } else {
-        const loaded = await invoke("check_engine");
-        if (loaded) { modelReady = true; return; }
-        const downloaded = await invoke("check_model", { size: settings.modelSize.toLowerCase() });
-        if (downloaded) { needsDownload = false; await loadModel(); }
-        else { needsDownload = true; }
+      await refreshRuntimeCapabilities();
+      const currentModel = currentModelCapabilities();
+      modelReady = !!currentModel?.loaded;
+      needsDownload = currentModel ? !currentModel.downloaded : true;
+      if (currentModelIsEnglishOnly() && settings.language !== "en") {
+        settings.language = "en";
       }
     } catch (err) {
       error = typeof err === "string" ? err : err.message || "Failed to check model";
@@ -162,6 +233,10 @@
   }
 
   async function loadModel() {
+    if (!tauriRuntimeAvailable) {
+      error = browserPreviewMessage;
+      return;
+    }
     modelLoading = true;
     page.status = "Loading model...";
     page.statusColor = "#e8c86e";
@@ -173,6 +248,7 @@
       } else {
         await invoke("load_model", { size: settings.modelSize.toLowerCase() });
       }
+      await refreshRuntimeCapabilities();
       modelReady = true;
       modelLoading = false;
       page.status = "Ready";
@@ -187,7 +263,7 @@
 
   function onModelDownloaded() {
     needsDownload = false;
-    loadModel();
+    void loadModel();
   }
 
   async function toggleRecording() {
@@ -200,8 +276,15 @@
 
   async function startRecording() {
     error = "";
+    liveWarning = "";
     saved = false;
     transcriptionFailed = false;
+    if (!tauriRuntimeAvailable) {
+      error = browserPreviewMessage;
+      page.status = "Desktop app required";
+      page.statusColor = "#e8c86e";
+      return;
+    }
     if (!modelReady) {
       if (needsDownload) return;
       await loadModel();
@@ -216,48 +299,49 @@
     }
 
     try {
-      audioContext = new AudioContext({ sampleRate: 16000 });
-
-      await audioContext.audioWorklet.addModule("/pcm-processor.js");
-
-      mediaStream = await navigator.mediaDevices.getUserMedia({
-        audio: { channelCount: 1, echoCancellation: true, noiseSuppression: true },
-      });
-
-      const source = audioContext.createMediaStreamSource(mediaStream);
-      workletNode = new AudioWorkletNode(audioContext, "pcm-processor");
-
-      pcmBuffer = [];
-      chunkId = 0;
-      chunkTimeOffset = 0;
+      sessionId = null;
+      drainingSessionId = null;
+      lastResultAt = 0;
+      lastLiveActivityAt = 0;
       processedChunks.clear();
 
       // Only clear transcript if not in insert mode (fresh recording)
       if (insertPos === -1) {
         transcript = "";
         segments = [];
+        previousMeasuredContentHeight = 0;
       }
 
-      allSamples = [];
-      workletNode.port.onmessage = (e) => {
-        if (e.data.type === "pcm") {
-          pcmBuffer = pcmBuffer.concat(e.data.samples);
-          if (settings.saveAudio) {
-            allSamples = allSamples.concat(e.data.samples);
-          }
-        }
-      };
+      resultChannel = new Channel((message) => handleLiveResult(message));
+      statusChannel = new Channel((message) => handleLiveStatus(message));
 
-      source.connect(workletNode);
+      const response = await invoke("start_live_transcription_session", {
+        config: {
+          engine: settings.engine,
+          modelId: selectedModelId(),
+          language: effectiveLanguage(),
+          initialPrompt: buildInitialPrompt(),
+          saveAudio: settings.saveAudio,
+          outputFolder: settings.outputFolder || null,
+          removeFillers: settings.removeFillers,
+          britishEnglish: settings.britishEnglish,
+          antiHallucination: settings.antiHallucination,
+          formatMode: settings.formatMode,
+          // Honour Settings → Audio → Microphone choice. Empty = auto-select.
+          microphoneDevice: settings.microphoneDevice || null,
+        },
+        resultChannel,
+        statusChannel,
+      });
+      sessionId = response.sessionId;
 
       startTime = Date.now();
       page.recording = true;
       page.status = "Recording...";
       page.statusColor = "#e87171";
       timerInterval = setInterval(updateTimer, 1000);
-      chunkTimer = setInterval(sendChunk, CHUNK_INTERVAL_MS);
     } catch (err) {
-      error = `Microphone access denied: ${err.message}`;
+      error = typeof err === "string" ? err : err?.message || "Microphone error";
       page.status = "Error";
       page.statusColor = "#e87171";
       cleanup();
@@ -266,115 +350,71 @@
 
   async function stopRecording() {
     clearInterval(timerInterval);
-    clearInterval(chunkTimer);
     page.recording = false;
+    transcribing = true;
     page.status = "Finalising...";
     page.statusColor = "#e8c86e";
 
-    const waitForTranscription = () => new Promise((resolve) => {
-      const check = () => transcribing ? setTimeout(check, 100) : resolve();
-      check();
-    });
-    await waitForTranscription();
-    await sendChunk();
-
-    cleanup();
-
-    if (chunkId === 0) {
-      page.status = "Ready";
-      page.statusColor = "#7ec89a";
-    }
-  }
-
-  function cleanup() {
-    if (workletNode) {
-      workletNode.disconnect();
-      workletNode = null;
-    }
-    if (mediaStream) {
-      mediaStream.getTracks().forEach((t) => t.stop());
-      mediaStream = null;
-    }
-    if (audioContext) {
-      audioContext.close();
-      audioContext = null;
-    }
-  }
-
-  async function sendChunk() {
-    if (pcmBuffer.length < MIN_CHUNK_SAMPLES) return;
-    if (transcribing) return;
-
-    chunkId++;
-    const currentChunkId = chunkId;
-    const samples = [...pcmBuffer];
-    pcmBuffer = [];
-    if (samples.length > MAX_PCM_SAMPLES) {
-      console.warn(`PCM buffer truncated from ${samples.length} to ${MAX_PCM_SAMPLES} samples (~5 min at 16 kHz)`);
-      samples.length = MAX_PCM_SAMPLES;
-    }
-    transcribing = true;
-
     try {
-      let initialPrompt = "";
-      if (page.activeProfile && page.activeProfile !== "None") {
-        initialPrompt = page.activeProfile;
-      }
+      const activityBeforeStop = lastLiveActivityAt;
+      drainingSessionId = sessionId;
+      const response = sessionId
+        ? await invoke("stop_live_transcription_session", { sessionId })
+        : null;
+      await waitForResultDrain(activityBeforeStop);
+      cleanup();
 
-      if (settings.engine === "parakeet") {
-        await invoke("transcribe_pcm_parakeet", {
-          samples,
-          chunkId: currentChunkId,
-          removeFillers: settings.removeFillers,
-          britishEnglish: settings.britishEnglish,
-          antiHallucination: settings.antiHallucination,
-          formatMode: settings.formatMode,
-        });
+      if (processedChunks.size === 0 && !transcript.trim()) {
+        page.status = "Ready";
+        page.statusColor = "#7ec89a";
       } else {
-        await invoke("transcribe_pcm", {
-          samples,
-          chunkId: currentChunkId,
-          language: settings.language,
-          initialPrompt,
-          removeFillers: settings.removeFillers,
-          britishEnglish: settings.britishEnglish,
-          antiHallucination: settings.antiHallucination,
-          formatMode: settings.formatMode,
-        });
+        await finaliseTranscription(response?.audioPath || null);
       }
     } catch (err) {
-      console.error("transcribe_pcm failed:", err);
-      error = typeof err === "string" ? err : err.message || "Transcription failed";
-      transcriptionFailed = true;
-      if (!page.recording) {
-        page.status = "Error";
-        page.statusColor = "#e87171";
-      }
+      error = typeof err === "string" ? err : err?.message || "Failed to stop live transcription";
+      page.status = "Error";
+      page.statusColor = "#e87171";
+      cleanup();
     } finally {
       transcribing = false;
     }
   }
 
-  async function finaliseTranscription() {
+  function cleanup() {
+    sessionId = null;
+    drainingSessionId = null;
+    resultChannel = null;
+    statusChannel = null;
+  }
+
+  async function waitForResultDrain(previousActivityAt = 0) {
+    const firstMessageDeadline = Date.now() + 400;
+    while (Date.now() < firstMessageDeadline) {
+      if (lastLiveActivityAt > previousActivityAt) {
+        break;
+      }
+      await new Promise((resolve) => setTimeout(resolve, 25));
+    }
+
+    if (lastLiveActivityAt <= previousActivityAt) {
+      return;
+    }
+
+    const idleDeadline = Date.now() + 1500;
+    while (Date.now() < idleDeadline) {
+      if (Date.now() - lastLiveActivityAt >= 200) {
+        return;
+      }
+      await new Promise((resolve) => setTimeout(resolve, 25));
+    }
+  }
+
+  async function finaliseTranscription(audioPath = null) {
     if (transcript.trim()) {
       if (settings.autoCopy) {
         invoke("copy_to_clipboard", { text: transcript }).catch(() => {});
       }
 
-      // Save audio if enabled — capture path for history replay
-      let audioPath = null;
-      if (settings.saveAudio && allSamples.length > 0) {
-        try {
-          audioPath = await invoke("save_audio", {
-            samples: allSamples,
-            outputFolder: settings.outputFolder || null,
-          });
-        } catch (err) {
-          console.error("save_audio failed:", err);
-        }
-        allSamples = [];
-      }
-
       const historyId = crypto.randomUUID();
       addToHistory({
         id: historyId,
@@ -384,7 +424,7 @@
         text: transcript,
         segments: segments,
         duration: (Date.now() - startTime) / 1000,
-        language: settings.language,
+        language: effectiveLanguage(),
         template: activeTemplate || undefined,
         audioPath,
       });
@@ -425,8 +465,12 @@
   function clearTranscript() {
     transcript = "";
     segments = [];
+    previousMeasuredContentHeight = 0;
+    shouldPreserveScrollAnchor = false;
+    shouldStickToBottom = false;
     page.timerText = "00:00";
     error = "";
+    liveWarning = "";
     saved = false;
     insertPos = -1;
     activeTemplate = "";
@@ -452,6 +496,7 @@
     showTemplateMenu = false;
     transcript = template.sections.map((s) => `## ${s}\n\n`).join("\n");
     segments = [];
+    previousMeasuredContentHeight = 0;
     // Position cursor at first section body
     requestAnimationFrame(() => {
       if (textareaEl) {
@@ -504,6 +549,80 @@
     return trimmed ? trimmed.split(/\s+/).length : 0;
   });
 
+  // Pretext content height measurement (DOM-free).
+  // Used for scroll-to-bottom affordance during live transcription.
+  let textareaWidth = $state(0);
+  let userScrolledUp = $state(false);
+  let previousMeasuredContentHeight = $state(0);
+  let shouldPreserveScrollAnchor = $state(false);
+  let shouldStickToBottom = $state(false);
+
+  // Build font string matching .font-transcript CSS
+  let transcriptFont = $derived(transcriptPretextFont(prefs.accessibility));
+  let transcriptLineHeight = $derived(transcriptPretextLineHeight(prefs.accessibility));
+
+  let contentHeight = $derived.by(() => {
+    if (!transcript.trim() || textareaWidth <= 0) return 0;
+    // Subtract padding (p-6 = 24px each side)
+    const innerWidth = textareaWidth - 48;
+    if (innerWidth <= 0) return 0;
+    return measurePreWrap(transcript, transcriptFont, innerWidth, transcriptLineHeight).height;
+  });
+
+  function onTextareaScroll(e) {
+    const el = e.target;
+    // "scrolled up" = more than 1 line from the bottom
+    userScrolledUp = (el.scrollHeight - el.scrollTop - el.clientHeight) > transcriptLineHeight;
+  }
+
+  function scrollToBottom() {
+    if (textareaEl) {
+      textareaEl.scrollTop = textareaEl.scrollHeight;
+      userScrolledUp = false;
+    }
+  }
+
+  // Track textarea width for Pretext layout calculations
+  $effect(() => {
+    if (!textareaEl) return;
+    const ro = new ResizeObserver(entries => {
+      for (const entry of entries) textareaWidth = entry.contentRect.width;
+    });
+    ro.observe(textareaEl);
+    return () => ro.disconnect();
+  });
+
+  $effect(() => {
+    const measuredHeight = contentHeight;
+    const previousHeight = previousMeasuredContentHeight;
+
+    if (!textareaEl) {
+      previousMeasuredContentHeight = measuredHeight;
+      return;
+    }
+
+    if (shouldPreserveScrollAnchor && measuredHeight > previousHeight) {
+      const delta = measuredHeight - previousHeight;
+      requestAnimationFrame(() => {
+        if (textareaEl) {
+          textareaEl.scrollTop += delta;
+        }
+      });
+      shouldPreserveScrollAnchor = false;
+    } else if (shouldStickToBottom) {
+      requestAnimationFrame(() => {
+        if (textareaEl) {
+          textareaEl.scrollTop = textareaEl.scrollHeight;
+        }
+      });
+      shouldStickToBottom = false;
+    }
+
+    previousMeasuredContentHeight = measuredHeight;
+  });
+
+  let showScrollHint = $derived(page.recording && userScrolledUp && contentHeight > 0);
+
   let reduceMotion = $derived(
     prefs.accessibility.reduceMotion === 'on'
     || (prefs.accessibility.reduceMotion === 'system'
@@ -519,7 +638,11 @@
     {#if page.recording}Recording started{/if}
   </div>
   {#if needsDownload}
-    <ModelDownloader modelSize={settings.modelSize.toLowerCase()} onComplete={onModelDownloaded} />
+    <ModelDownloader
+      engine={settings.engine}
+      modelSize={settings.modelSize.toLowerCase()}
+      onComplete={onModelDownloaded}
+    />
   {:else}
     <!-- Control strip -->
     <div class="flex items-center gap-3 px-5 h-[56px] border-b border-border-subtle flex-shrink-0">
@@ -531,9 +654,11 @@
             ? 'bg-danger animate-pulse-warm'
             : modelLoading
               ? 'bg-warning opacity-60 cursor-wait'
-              : 'bg-accent hover:bg-accent-hover shadow-[0_4px_20px_rgba(232,168,124,0.3)]'}"
+              : !tauriRuntimeAvailable
+                ? 'bg-bg-elevated text-text-tertiary cursor-not-allowed'
+                : 'bg-accent hover:bg-accent-hover shadow-[0_4px_20px_rgba(232,168,124,0.3)]'}"
         onclick={toggleRecording}
-        disabled={modelLoading}
+        disabled={modelLoading || !tauriRuntimeAvailable}
         aria-label={page.recording ? "Stop recording" : "Start recording"}
         style="transition-duration: var(--duration-ui)"
       >
@@ -571,6 +696,8 @@
           <span class="text-[11px] text-text-tertiary">
             {#if modelLoading}
               Loading model...
+            {:else if !tauriRuntimeAvailable}
+              Desktop app required for local transcription
             {:else if saved}
               <span class="text-success animate-fade-in">
                 Saved{#if extractedCount > 0} · {extractedCount} task{extractedCount === 1 ? '' : 's'} extracted{/if}
@@ -612,7 +739,7 @@
         aria-label="Toggle task sidebar"
       >
         <span class="flex items-center gap-1.5">
-          <SquareCheck size={16} class="{page.taskSidebarOpen ? 'text-accent' : 'text-text-tertiary'}" aria-hidden="true" />
+          <SquareCheck size={16} class={page.taskSidebarOpen ? 'text-accent' : 'text-text-tertiary'} aria-hidden="true" />
           <span class="text-[11px] {page.taskSidebarOpen ? 'text-accent' : 'text-text-tertiary'}">Tasks</span>
         </span>
         {#if taskCount > 0}
@@ -704,6 +831,14 @@
       </div>
     {/if}
 
+    {#if liveWarning && !error}
+      <div class="px-5 pt-2 animate-fade-in flex-shrink-0">
+        <div class="px-4 py-2 rounded-lg bg-warning/10 border border-warning/20 text-[12px] text-warning">
+          {liveWarning}
+        </div>
+      </div>
+    {/if}
+
     <!-- Transcript area -->
     <div class="flex-1 px-5 pt-3 pb-3 min-h-0" style="--text-transcript: {prefs.accessibility.transcriptSize}px; font-size: var(--text-transcript)">
       <Card classes="h-full flex flex-col">
@@ -727,17 +862,30 @@
             aria-live="polite"
           ></textarea>
         {:else}
-          <textarea
-            bind:this={textareaEl}
-            class="font-transcript flex-1 w-full bg-transparent text-text p-6
-                   resize-none focus:outline-none placeholder:text-text-tertiary min-h-0"
-            placeholder={activeTemplate ? "Click a section above, then press record..." : "Your words will appear here..."}
-            bind:value={transcript}
-            onclick={() => { showExportMenu = false; showTemplateMenu = false; }}
-            data-no-transition
-            aria-label="Transcript"
-            aria-live="polite"
-          ></textarea>
+          <div class="relative flex-1 min-h-0">
+            <textarea
+              bind:this={textareaEl}
+              class="font-transcript h-full w-full bg-transparent text-text p-6
+                     resize-none focus:outline-none placeholder:text-text-tertiary"
+              placeholder={activeTemplate ? "Click a section above, then press record..." : "Your words will appear here..."}
+              bind:value={transcript}
+              onscroll={onTextareaScroll}
+              onclick={() => { showExportMenu = false; showTemplateMenu = false; }}
+              data-no-transition
+              aria-label="Transcript"
+              aria-live="polite"
+            ></textarea>
+            {#if showScrollHint}
+              <button
+                class="absolute bottom-3 right-6 px-3 py-1.5 rounded-full bg-accent text-white text-[11px] font-medium shadow-md hover:bg-accent-hover animate-fade-in"
+                onclick={scrollToBottom}
+                aria-label="Scroll to latest"
+                style="transition-duration: var(--duration-ui)"
+              >
+                ↓ New text
+              </button>
+            {/if}
+          </div>
         {/if}
 
         <!-- Status footer (inside transcript card) -->