@agent-native/core 0.157.0 → 0.157.1

package/corpus/templates/clips/chrome-extension/src/offscreen.ts

@@ -22,7 +22,6 @@ import {
   isLikelyPhoneMicLabel,
   type AudioInputFallback,
 } from "@shared/media-device-selection";
-import { createMicAudioCleanup } from "@shared/mic-audio-cleanup";
 import { scheduleReadyChime } from "@shared/recording-audio";
 import {
   SCREEN_CAPTURE_FRAME_RATE,
@@ -435,14 +434,9 @@ async function getCameraStream(deviceId: string): Promise<MediaStream> {
   }
 }
 
-// Phones/Continuity-style mics already apply their own echo cancellation,
-// noise suppression, and gain control before the audio ever reaches Chrome.
-// Stacking Chrome's versions of the same processing on top double-processes
-// the signal and audibly degrades it. We can't detect "is this device doing
-// its own processing" directly, so we use the device label as a heuristic and
-// ask the browser to SKIP its own noise suppression / AGC for those devices
-// (`{ ideal: false }` — a best-effort request, not a hard requirement). Echo
-// cancellation stays on for everything.
+// Let the browser apply its built-in voice processing. Device labels are not a
+// reliable signal for whether a hardware path already processed its audio, so
+// do not branch capture quality on a label heuristic.
 async function chooseFallbackMicDevice(
   requestedLabel: string,
   avoidDeviceIds: string[] = [],
@@ -459,13 +453,12 @@ async function chooseFallbackMicDevice(
 
 function voiceFocusedAudioConstraints(
   deviceId: string,
-  deviceLabel = "",
+  _deviceLabel = "",
 ): MediaTrackConstraints {
-  const skipRedundantProcessing = isLikelyPhoneMicLabel(deviceLabel);
   const audio: MediaTrackConstraints = {
     echoCancellation: true,
-    noiseSuppression: skipRedundantProcessing ? { ideal: false } : true,
-    autoGainControl: skipRedundantProcessing ? { ideal: false } : true,
+    noiseSuppression: true,
+    autoGainControl: true,
     channelCount: 1,
   };
   if (deviceId) audio.deviceId = { exact: deviceId };
@@ -811,20 +804,11 @@ async function createMixedAudio(
   const audioContext = new AudioContext();
   await audioContext.resume().catch(() => undefined);
   const destination = audioContext.createMediaStreamDestination();
-  const micCleanup: Array<{ stop: () => void }> = [];
   for (const input of audioInputs) {
-    let track = input.track;
-    if (input.isMicrophone) {
-      const cleanup = createMicAudioCleanup(new MediaStream([track]), {
-        audioContext,
-      });
-      micCleanup.push(cleanup);
-      track = cleanup.stream.getAudioTracks()[0] ?? track;
-    }
     // One source per track (not per stream) so each input is isolated in the
     // mix graph and can be detached independently.
     const source = audioContext.createMediaStreamSource(
-      new MediaStream([track]),
+      new MediaStream([input.track]),
     );
     source.connect(destination);
     // A mixed input track can end mid-recording — most commonly the shared
@@ -832,7 +816,7 @@ async function createMixedAudio(
     // just that dead source so the mixed output keeps carrying the surviving
     // inputs (the microphone). Without this the whole mixed destination can go
     // silent, dropping both the tab audio and the mic for the rest of the clip.
-    track.addEventListener("ended", () => {
+    input.track.addEventListener("ended", () => {
       try {
         source.disconnect();
       } catch {
@@ -847,7 +831,6 @@ async function createMixedAudio(
     cleanup() {
       if (cleanedUp) return;
       cleanedUp = true;
-      for (const cleanup of micCleanup) cleanup.stop();
     },
   };
 }

package/corpus/templates/clips/desktop/src/app.tsx

@@ -977,6 +977,7 @@ export function App() {
   );
   const localRecordingMode: LocalRecordingMode =
     featureConfig?.localRecordingMode ?? "off";
+  const voiceCleanupEnabled = featureConfig?.voiceCleanupEnabled !== false;
 
   const [pendingUploads, setPendingUploads] = useState<PendingDesktopUpload[]>(
     [],
@@ -3160,6 +3161,7 @@ export function App() {
         cameraOn,
         micOn,
         systemAudioOn,
+        voiceCleanupEnabled,
         localRecordingMode,
         preAcquiredCameraStream,
         preAcquiredDisplayStream: options?.resumeCapture?.displayStream ?? null,
@@ -5391,6 +5393,7 @@ function Setup({
   const featureConfig = useFeatureConfig();
   const updateStatus = useUpdateStatus();
   const voiceEnabled = featureConfig?.voiceEnabled !== false;
+  const voiceCleanupEnabled = featureConfig?.voiceCleanupEnabled !== false;
   const meetingsEnabled = featureConfig?.meetingsEnabled !== false;
   const launchAtLoginEnabled = featureConfig?.launchAtLoginEnabled !== false;
   const autoHidePopoverEnabled = featureConfig?.autoHidePopoverEnabled === true;
@@ -5540,6 +5543,15 @@ function Setup({
     );
   }
 
+  function setVoiceCleanupEnabled(enabled: boolean) {
+    if (!featureConfig) return;
+    invoke("set_feature_config", {
+      config: { ...featureConfig, voiceCleanupEnabled: enabled },
+    }).catch((err) =>
+      console.error("[settings] set_feature_config failed", err),
+    );
+  }
+
   function setMeetingsEnabled(enabled: boolean) {
     if (!featureConfig) return;
     invoke("set_feature_config", {
@@ -7258,6 +7270,18 @@ function Setup({
           title="Capture"
           description="Choose how recordings are saved and how the tray opens them."
         >
+          <DesktopSettingsRow
+            label="Voice cleanup"
+            description="Reduce steady background noise in microphone recordings."
+            control={
+              <Switch
+                on={voiceCleanupEnabled}
+                onChange={setVoiceCleanupEnabled}
+                disabled={captureControlsLocked}
+                label="Use voice cleanup"
+              />
+            }
+          />
           <DesktopSettingsRow
             label="Rewind"
             description={rewindStatusPresentation.title}

package/corpus/templates/clips/desktop/src/lib/media-capture-constraints.ts

@@ -13,12 +13,22 @@ export const VOICE_FOCUSED_AUDIO_CONSTRAINTS: MediaTrackConstraints = {
   channelCount: { ideal: 1 },
 };
 
+const RAW_AUDIO_CONSTRAINTS: MediaTrackConstraints = {
+  echoCancellation: false,
+  noiseSuppression: false,
+  autoGainControl: false,
+  channelCount: { ideal: 1 },
+};
+
 export function voiceFocusedAudioConstraints(
   deviceId?: string | null,
+  voiceCleanupEnabled = true,
 ): MediaTrackConstraints {
   const id = deviceId?.trim();
   return {
-    ...VOICE_FOCUSED_AUDIO_CONSTRAINTS,
+    ...(voiceCleanupEnabled
+      ? VOICE_FOCUSED_AUDIO_CONSTRAINTS
+      : RAW_AUDIO_CONSTRAINTS),
     ...(id ? { deviceId: { exact: id } } : {}),
   };
 }
@@ -73,9 +83,10 @@ function stopStream(stream: MediaStream): void {
 
 async function getVoiceFocusedAudioStream(
   deviceId?: string | null,
+  voiceCleanupEnabled = true,
 ): Promise<MediaStream> {
   return navigator.mediaDevices.getUserMedia({
-    audio: voiceFocusedAudioConstraints(deviceId),
+    audio: voiceFocusedAudioConstraints(deviceId, voiceCleanupEnabled),
     video: false,
   });
 }
@@ -101,6 +112,7 @@ async function fallbackAudioInput(
 async function tryFallbackAudioInput(
   savedLabel: string | null | undefined,
   avoidDeviceIds: Array<string | null | undefined>,
+  voiceCleanupEnabled = true,
 ): Promise<MediaStream | null> {
   const fallback = await fallbackAudioInput(savedLabel, avoidDeviceIds);
   if (!fallback) return null;
@@ -109,7 +121,10 @@ async function tryFallbackAudioInput(
       `[clips-recorder] selected mic unavailable; retrying ${fallback.reason} microphone`,
       { deviceId: fallback.deviceId, label: fallback.label },
     );
-    return await getVoiceFocusedAudioStream(fallback.deviceId);
+    return await getVoiceFocusedAudioStream(
+      fallback.deviceId,
+      voiceCleanupEnabled,
+    );
   } catch (fallbackErr) {
     if (!isMediaConstraintFailure(fallbackErr)) throw fallbackErr;
     console.warn(
@@ -122,9 +137,10 @@ async function tryFallbackAudioInput(
 
 async function getDefaultAudioStreamWithBasicFallback(
   reason: unknown,
+  voiceCleanupEnabled = true,
 ): Promise<MediaStream> {
   try {
-    return await getVoiceFocusedAudioStream();
+    return await getVoiceFocusedAudioStream(undefined, voiceCleanupEnabled);
   } catch (fallbackErr) {
     if (!isMediaConstraintFailure(fallbackErr)) throw fallbackErr;
     console.warn(
@@ -132,7 +148,7 @@ async function getDefaultAudioStreamWithBasicFallback(
       fallbackErr,
     );
     return navigator.mediaDevices.getUserMedia({
-      audio: true,
+      audio: voiceCleanupEnabled ? true : RAW_AUDIO_CONSTRAINTS,
       video: false,
     });
   } finally {
@@ -147,13 +163,16 @@ async function getDefaultAudioStreamWithBasicFallback(
 
 async function replacePhoneDefaultMicIfPossible(
   stream: MediaStream,
+  voiceCleanupEnabled = true,
 ): Promise<MediaStream> {
   const track = stream.getAudioTracks()[0];
   if (!track || !isLikelyPhoneMicLabel(track.label)) return stream;
   const settings = track.getSettings?.();
-  const replacement = await tryFallbackAudioInput(null, [
-    settings?.deviceId ?? null,
-  ]);
+  const replacement = await tryFallbackAudioInput(
+    null,
+    [settings?.deviceId ?? null],
+    voiceCleanupEnabled,
+  );
   if (!replacement) return stream;
   console.warn(
     "[clips-recorder] default mic resolved to a phone-like input; using explicit fallback mic",
@@ -198,12 +217,15 @@ export async function getCameraStreamWithFallback(
 export async function getAudioStreamWithFallback(
   deviceId?: string | null,
   savedLabel?: string | null,
+  voiceCleanupEnabled = true,
 ): Promise<MediaStream> {
   const id = deviceId?.trim();
 
   try {
-    const stream = await getVoiceFocusedAudioStream(id);
-    return id ? stream : replacePhoneDefaultMicIfPossible(stream);
+    const stream = await getVoiceFocusedAudioStream(id, voiceCleanupEnabled);
+    return id
+      ? stream
+      : replacePhoneDefaultMicIfPossible(stream, voiceCleanupEnabled);
   } catch (err) {
     let deviceGone = false;
     if (id && err instanceof DOMException && err.name === "NotAllowedError") {
@@ -227,11 +249,18 @@ export async function getAudioStreamWithFallback(
     }
     if (!isMediaConstraintFailure(err) && !deviceGone) throw err;
     if (id) {
-      const fallback = await tryFallbackAudioInput(savedLabel, [id]);
+      const fallback = await tryFallbackAudioInput(
+        savedLabel,
+        [id],
+        voiceCleanupEnabled,
+      );
       if (fallback) return fallback;
     }
 
-    const defaultStream = await getDefaultAudioStreamWithBasicFallback(err);
-    return replacePhoneDefaultMicIfPossible(defaultStream);
+    const defaultStream = await getDefaultAudioStreamWithBasicFallback(
+      err,
+      voiceCleanupEnabled,
+    );
+    return replacePhoneDefaultMicIfPossible(defaultStream, voiceCleanupEnabled);
   }
 }

package/corpus/templates/clips/desktop/src/lib/recorder.ts

@@ -55,10 +55,6 @@ import {
   waitForAcceptedRecordingAfterFinalizeError,
   waitForReadyRecordingAfterFinalizeError,
 } from "../../../shared/finalize-recovery";
-import {
-  createMicAudioCleanup,
-  type MicAudioCleanupHandle,
-} from "../../../shared/mic-audio-cleanup";
 import type { LocalRecordingMode } from "../shared/config";
 import { createAudioCue, type AudioCue } from "./audio-cue";
 import { createCameraCompositeStream } from "./camera-composite";
@@ -239,6 +235,8 @@ export interface StartParams {
   cameraOn: boolean;
   /** Record + transcribe system/desktop audio. Default true. */
   systemAudioOn?: boolean;
+  /** Apply the browser/native voice cleanup path to microphone audio. */
+  voiceCleanupEnabled?: boolean;
   /** Cancels or bounds the pre-record capture setup. */
   signal?: AbortSignal;
   localRecordingMode?: LocalRecordingMode;
@@ -548,9 +546,9 @@ interface RecordingAudio {
 }
 
 /**
- * Build the audio track(s) for the recording. Microphone tracks pass through
- * the cleanup graph whether or not system/display audio is also present, then
- * both sources are mixed into one track when needed.
+ * Build the audio track(s) for the recording. Browser capture requests its
+ * built-in voice processing; this graph only combines raw source tracks when
+ * MediaRecorder needs one mixed track.
  */
 function buildRecordingAudio(
   micTracks: MediaStreamTrack[],
@@ -571,22 +569,13 @@ function buildRecordingAudio(
   }
   const ctx: AudioContext = new AudioCtx();
   const destination = ctx.createMediaStreamDestination();
-  const micCleanup: MicAudioCleanupHandle[] = [];
-  const cleanedMicTracks = micTracks.map((track) => {
-    const cleanup = createMicAudioCleanup(new MediaStream([track]), {
-      audioContext: ctx,
-    });
-    micCleanup.push(cleanup);
-    return cleanup.stream.getAudioTracks()[0] ?? track;
-  });
-  for (const tracks of [cleanedMicTracks, systemTracks]) {
+  for (const tracks of [micTracks, systemTracks]) {
     if (!tracks.length) continue;
     ctx.createMediaStreamSource(new MediaStream(tracks)).connect(destination);
   }
   return {
     tracks: destination.stream.getAudioTracks(),
     cleanup() {
-      for (const cleanup of micCleanup) cleanup.stop();
       ctx.close().catch(() => {});
     },
   };
@@ -4307,7 +4296,11 @@ async function startRecordingInner(
   const audioStreamPromise: Promise<MediaStream> | null = resumedAudioStream
     ? Promise.resolve(resumedAudioStream)
     : wantsAudio
-      ? getAudioStreamWithFallback(params.micId, params.micLabel)
+      ? getAudioStreamWithFallback(
+          params.micId,
+          params.micLabel,
+          params.voiceCleanupEnabled,
+        )
       : null;
 
   // getDisplayMedia can remain pending in a long-lived WebKit tray webview.

package/corpus/templates/clips/desktop/src/overlays/onboarding.tsx

@@ -127,6 +127,7 @@ export function Onboarding() {
           clipsEnabled: clips,
           meetingsEnabled: meetings,
           voiceEnabled: voice,
+          voiceCleanupEnabled: true,
           launchAtLoginEnabled: true,
           autoHidePopoverEnabled: false,
           meetingTranscriptionMode: "ask",

package/corpus/templates/clips/desktop/src/shared/config.ts

@@ -120,6 +120,7 @@ export interface FeatureConfig {
   clipsEnabled: boolean;
   meetingsEnabled: boolean;
   voiceEnabled: boolean;
+  voiceCleanupEnabled: boolean;
   launchAtLoginEnabled: boolean;
   autoHidePopoverEnabled: boolean;
   meetingTranscriptionMode: "manual" | "ask" | "auto";

package/corpus/templates/clips/desktop/src/styles.css

@@ -6074,9 +6074,14 @@ body[data-clips-route="recording-pill"] #root {
   /* No fill, and the stroke is painted OUTWARD with an outset box-shadow so it
      lands just outside the captured rect — never inside the recorded video. */
   box-shadow:
-    0 0 0 2px rgba(239, 68, 68, 0.95),
-    0 0 0 4px rgba(239, 68, 68, 0.32),
-    0 0 26px 6px rgba(239, 68, 68, 0.26);
+    /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+    0 0 0 1px rgba(255, 255, 255, 0.96),
+    /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+    0 0 0 3px rgba(220, 38, 38, 0.98),
+    /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+    0 0 0 6px rgba(17, 24, 39, 0.78),
+    /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+    0 0 30px 8px rgba(220, 38, 38, 0.28);
   animation: region-record-border-pulse 1.8s ease-in-out infinite;
 }
 
@@ -6086,15 +6091,25 @@ body[data-clips-route="recording-pill"] #root {
   0%,
   100% {
     box-shadow:
-      0 0 0 2px rgba(239, 68, 68, 0.9),
-      0 0 0 4px rgba(239, 68, 68, 0.28),
-      0 0 22px 5px rgba(239, 68, 68, 0.2);
+      /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+      0 0 0 1px rgba(255, 255, 255, 0.98),
+      /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+        0 0 0 3px rgba(220, 38, 38, 1),
+      /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+        0 0 0 6px rgba(17, 24, 39, 0.76),
+      /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+        0 0 24px 7px rgba(220, 38, 38, 0.22);
   }
   50% {
     box-shadow:
-      0 0 0 2px rgba(239, 68, 68, 1),
-      0 0 0 4px rgba(239, 68, 68, 0.44),
-      0 0 30px 8px rgba(239, 68, 68, 0.34);
+      /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+      0 0 0 1px rgba(255, 255, 255, 1),
+      /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+        0 0 0 3px rgba(220, 38, 38, 1),
+      /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+        0 0 0 6px rgba(17, 24, 39, 0.88),
+      /* guard:allow-raw-color — recording frame uses fixed capture-safe contrast. */
+        0 0 34px 10px rgba(220, 38, 38, 0.36);
   }
 }
 

package/corpus/templates/clips/desktop/src-tauri/Cargo.toml

@@ -143,6 +143,9 @@ screencapturekit = { version = "2.0.0", features = ["macos_15_0"] }
 # AudioBufferList type so we can hand SCK's CMSampleBuffer audio bytes to
 # AVAudioPCMBuffer for the speech recognizer.
 objc2-core-audio-types = "0.3"
+# Pure-Rust RNNoise-derived microphone denoising for the realtime
+# ScreenCaptureKit mixer. Keep the CLI/dasp features out of the desktop bundle.
+nnnoiseless = { version = "0.5.2", default-features = false }
 # `metal` builds whisper.cpp's ggml-metal backend and links it in (the sys
 # crate's build script embeds the compiled metallib, so no app-bundle/runtime
 # asset changes are needed). Without it, whisper.cpp decode runs CPU-only.

package/corpus/templates/clips/desktop/src-tauri/src/config.rs

@@ -111,6 +111,8 @@ pub struct FeatureConfig {
     pub clips_enabled: bool,
     pub meetings_enabled: bool,
     pub voice_enabled: bool,
+    #[serde(default = "default_voice_cleanup_enabled")]
+    pub voice_cleanup_enabled: bool,
     #[serde(default = "default_launch_at_login_enabled")]
     pub launch_at_login_enabled: bool,
     #[serde(default)]
@@ -160,6 +162,10 @@ fn default_launch_at_login_enabled() -> bool {
     true
 }
 
+fn default_voice_cleanup_enabled() -> bool {
+    true
+}
+
 fn default_meeting_transcription_mode() -> MeetingTranscriptionMode {
     MeetingTranscriptionMode::Ask
 }
@@ -223,6 +229,7 @@ impl Default for FeatureConfig {
             clips_enabled: true,
             meetings_enabled: true,
             voice_enabled: true,
+            voice_cleanup_enabled: default_voice_cleanup_enabled(),
             launch_at_login_enabled: true,
             auto_hide_popover_enabled: false,
             meeting_transcription_mode: default_meeting_transcription_mode(),
@@ -436,6 +443,7 @@ mod tests {
         .unwrap();
 
         assert_eq!(config.whisper_model_id, "base");
+        assert!(config.voice_cleanup_enabled);
     }
 
     #[test]