@agent-native/core 0.86.1 → 0.88.0

package/corpus/templates/clips/changelog/2026-07-05-desktop-recording-shortcut-start-stop.md

@@ -0,0 +1,6 @@
+---
+type: added
+date: 2026-07-05
+---
+
+Desktop Clips can now use a custom global shortcut to start or stop recording.

package/corpus/templates/clips/changelog/2026-07-05-desktop-recordings-no-longer-fail-when-saved-capture-constra.md

@@ -0,0 +1,6 @@
+---
+type: fixed
+date: 2026-07-05
+---
+
+Desktop recordings no longer fail when saved capture constraints are stale after an app update.

package/corpus/templates/clips/changelog/2026-07-05-permanently-deleted-clips-now-clean-up-s3-media-objects.md

@@ -0,0 +1,6 @@
+---
+type: fixed
+date: 2026-07-05
+---
+
+Permanently deleted clips now clean up matching video and thumbnail objects from configured S3-compatible storage.

package/corpus/templates/clips/changelog/2026-07-05-timeline-zoom-controls-are-now-visible-in-the-clip-editor-an.md

@@ -0,0 +1,6 @@
+---
+type: improved
+date: 2026-07-05
+---
+
+Timeline zoom controls are now visible in the clip editor, support trackpad pinch gestures, and stay anchored while selecting ranges.

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

@@ -156,6 +156,7 @@ const VOICE_SHORTCUT_KEY = "clips:voice-shortcut";
 const VOICE_SHORTCUT_CONFIGURED_KEY = "clips:voice-shortcut-configured";
 const VOICE_CUSTOM_SHORTCUT_KEY = "clips:voice-custom-shortcut";
 const POPOVER_CUSTOM_SHORTCUT_KEY = "clips:popover-custom-shortcut";
+const RECORD_CUSTOM_SHORTCUT_KEY = "clips:record-custom-shortcut";
 const VOICE_MODE_KEY = "clips:voice-mode";
 const VOICE_PROVIDER_KEY = "clips:voice-provider";
 const VOICE_INSTRUCTIONS_KEY = "clips:voice-instructions";
@@ -596,6 +597,9 @@ export function App() {
   const [popoverCustomShortcut, setPopoverCustomShortcut] = useState<string>(
     () => loadStringAllowEmpty(POPOVER_CUSTOM_SHORTCUT_KEY, ""),
   );
+  const [recordCustomShortcut, setRecordCustomShortcut] = useState<string>(() =>
+    loadStringAllowEmpty(RECORD_CUSTOM_SHORTCUT_KEY, ""),
+  );
   const [voiceMode, setVoiceMode] = useState<VoiceMode>(() => {
     const saved = loadString(VOICE_MODE_KEY, "push-to-talk");
     return saved === "toggle" ? "toggle" : "push-to-talk";
@@ -656,6 +660,7 @@ export function App() {
   const isRecording = recorder !== null;
   // Whether the popover window is shown; driven by the visibility effect below.
   const [popoverVisible, setPopoverVisible] = useState(false);
+  const recordShortcutHandlerRef = useRef<() => void>(() => {});
   // Mirrors `bubbleActive` (assigned below once it is computed) so device
   // probes can synchronously tell whether the camera bubble owns the grant.
   const bubbleActiveRef = useRef(false);
@@ -779,6 +784,7 @@ export function App() {
     invoke("set_custom_shortcuts", {
       voice: voiceShortcut === "custom" ? voiceCustomShortcut : null,
       popover: popoverCustomShortcut.trim() ? popoverCustomShortcut : null,
+      record: recordCustomShortcut.trim() ? recordCustomShortcut : null,
     })
       .then(() => {
         if (!cancelled) setShortcutRegistrationError(null);
@@ -794,7 +800,12 @@ export function App() {
     return () => {
       cancelled = true;
     };
-  }, [popoverCustomShortcut, voiceCustomShortcut, voiceShortcut]);
+  }, [
+    popoverCustomShortcut,
+    recordCustomShortcut,
+    voiceCustomShortcut,
+    voiceShortcut,
+  ]);
 
   // ---- auth status --------------------------------------------------------
   // The Tauri WebView has its own cookie jar (separate from the user's
@@ -1571,6 +1582,10 @@ export function App() {
     () => saveString(POPOVER_CUSTOM_SHORTCUT_KEY, popoverCustomShortcut),
     [popoverCustomShortcut],
   );
+  useEffect(
+    () => saveString(RECORD_CUSTOM_SHORTCUT_KEY, recordCustomShortcut),
+    [recordCustomShortcut],
+  );
   useEffect(() => saveString(VOICE_MODE_KEY, voiceMode), [voiceMode]);
   useEffect(
     () => saveString(VOICE_PROVIDER_KEY, voiceProvider),
@@ -1952,6 +1967,66 @@ export function App() {
     setRecError(message);
   }
 
+  recordShortcutHandlerRef.current = () => {
+    if (recorder) {
+      emit("clips:recorder-stop").catch(() => {});
+      return;
+    }
+    if (recordingFlowGateRef.current || recordingFlowActive) {
+      emit("clips:countdown-cancel").catch(() => {});
+      return;
+    }
+
+    setShowSettings(false);
+    if (authStatus === "anon" && localRecordingMode === "off") {
+      setRecError("Sign in to Clips before using the recording shortcut.");
+      invoke("show_popover").catch(() => {});
+      return;
+    }
+
+    const canStartFromGlobalShortcut =
+      mode === "camera" || nativeFullscreenRecordingActive;
+    if (!canStartFromGlobalShortcut) {
+      setRecError(
+        "Open Clips and click Start recording to use the selected source.",
+      );
+      invoke("show_popover").catch(() => {});
+      return;
+    }
+
+    void handleStartRecording({ ignoreActiveRecorder: true });
+  };
+
+  useEffect(() => {
+    let cancelled = false;
+    let unlisten: (() => void) | null = null;
+    listen("clips:record-shortcut", () => {
+      recordShortcutHandlerRef.current();
+    })
+      .then((u) => {
+        if (cancelled) {
+          try {
+            u();
+          } catch {
+            // ignore
+          }
+          return;
+        }
+        unlisten = u;
+      })
+      .catch(() => {});
+    return () => {
+      cancelled = true;
+      if (unlisten) {
+        try {
+          unlisten();
+        } catch {
+          // ignore
+        }
+      }
+    };
+  }, []);
+
   function updateReadinessOpen(next: boolean) {
     setReadinessOpen(next);
     if (!next) saveBool(READINESS_REVIEWED_KEY, true);
@@ -2128,6 +2203,7 @@ export function App() {
           voiceShortcut={voiceShortcut}
           voiceCustomShortcut={voiceCustomShortcut}
           popoverCustomShortcut={popoverCustomShortcut}
+          recordCustomShortcut={recordCustomShortcut}
           voiceMode={voiceMode}
           voiceProvider={voiceProvider}
           voiceInstructions={voiceInstructions}
@@ -2135,6 +2211,7 @@ export function App() {
           onVoiceShortcutChange={updateVoiceShortcut}
           onVoiceCustomShortcutChange={setVoiceCustomShortcut}
           onPopoverCustomShortcutChange={setPopoverCustomShortcut}
+          onRecordCustomShortcutChange={setRecordCustomShortcut}
           onVoiceModeChange={setVoiceMode}
           onVoiceProviderChange={setVoiceProvider}
           onVoiceInstructionsChange={setVoiceInstructions}
@@ -3059,6 +3136,7 @@ function Setup({
   voiceShortcut,
   voiceCustomShortcut,
   popoverCustomShortcut,
+  recordCustomShortcut,
   voiceMode,
   voiceProvider,
   voiceInstructions,
@@ -3066,6 +3144,7 @@ function Setup({
   onVoiceShortcutChange,
   onVoiceCustomShortcutChange,
   onPopoverCustomShortcutChange,
+  onRecordCustomShortcutChange,
   onVoiceModeChange,
   onVoiceProviderChange,
   onVoiceInstructionsChange,
@@ -3079,6 +3158,7 @@ function Setup({
   voiceShortcut: VoiceShortcutPreference;
   voiceCustomShortcut: string;
   popoverCustomShortcut: string;
+  recordCustomShortcut: string;
   voiceMode: VoiceMode;
   voiceProvider: VoiceProvider;
   voiceInstructions: string;
@@ -3086,6 +3166,7 @@ function Setup({
   onVoiceShortcutChange: (value: VoiceShortcutPreference) => void;
   onVoiceCustomShortcutChange: (value: string) => void;
   onPopoverCustomShortcutChange: (value: string) => void;
+  onRecordCustomShortcutChange: (value: string) => void;
   onVoiceModeChange: (value: VoiceMode) => void;
   onVoiceProviderChange: (value: VoiceProvider) => void;
   onVoiceInstructionsChange: (value: string) => void;
@@ -3967,6 +4048,22 @@ function Setup({
         </div>
       </details>
 
+      <div className="setup-section">
+        <SettingLabel
+          label="Start/stop recording shortcut"
+          hint="Optional global shortcut for starting full-screen, region, or camera recordings and stopping the active recording."
+        />
+        <ShortcutRecorder
+          value={recordCustomShortcut}
+          placeholder="Record shortcut"
+          onChange={onRecordCustomShortcutChange}
+        />
+        <p className="setup-hint">
+          Window and browser-tab sources still open Clips first so the picker
+          can use a click.
+        </p>
+      </div>
+
       <div className="setup-section">
         <SettingLabel
           label="Open Clips shortcut"

package/corpus/templates/clips/desktop/src/components/MediaDeviceRow.tsx

@@ -104,6 +104,7 @@ export function MediaDeviceRow({
   const { open, setOpen, rowRef } = useRowMenu();
 
   const disabled = !on;
+  const canOpenMenu = !disabled || (kind === "mic" && !!onSystemAudioToggle);
   const defaultLabel = kind === "camera" ? "Default camera" : "Default mic";
   const accessLabel =
     kind === "camera" ? "Allow camera access" : "Allow microphone access";
@@ -119,9 +120,9 @@ export function MediaDeviceRow({
         type="button"
         className="row-button"
         onClick={() => {
-          if (!disabled) setOpen((v) => !v);
+          if (canOpenMenu) setOpen((v) => !v);
         }}
-        disabled={disabled}
+        disabled={!canOpenMenu}
         title={label}
       >
         <span className="row-label">{label}</span>

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

@@ -0,0 +1,121 @@
+export const VOICE_FOCUSED_AUDIO_CONSTRAINTS: MediaTrackConstraints = {
+  echoCancellation: { ideal: true },
+  noiseSuppression: { ideal: true },
+  autoGainControl: { ideal: true },
+  channelCount: { ideal: 1 },
+};
+
+export function voiceFocusedAudioConstraints(
+  deviceId?: string | null,
+): MediaTrackConstraints {
+  const id = deviceId?.trim();
+  return {
+    ...VOICE_FOCUSED_AUDIO_CONSTRAINTS,
+    ...(id ? { deviceId: { exact: id } } : {}),
+  };
+}
+
+export function buildDesktopDisplayMediaOptions({
+  audio,
+  frameRate,
+  maxWidth,
+  maxHeight,
+}: {
+  audio: boolean;
+  frameRate: number;
+  maxWidth: number;
+  maxHeight: number;
+}): DisplayMediaStreamOptions {
+  return {
+    video: {
+      frameRate: {
+        ideal: frameRate,
+        max: frameRate,
+      },
+      width: { ideal: maxWidth },
+      height: { ideal: maxHeight },
+    },
+    audio,
+  };
+}
+
+export function isMediaConstraintFailure(err: unknown): boolean {
+  const name =
+    err instanceof DOMException || err instanceof Error ? err.name : "";
+  const message = err instanceof Error ? err.message : String(err);
+  return (
+    name === "OverconstrainedError" ||
+    name === "ConstraintNotSatisfiedError" ||
+    name === "NotFoundError" ||
+    /invalid constraint|overconstrained|could not satisfy constraint|device not found|requested device not found/i.test(
+      message,
+    )
+  );
+}
+
+export async function getCameraStreamWithFallback(
+  deviceId?: string | null,
+): Promise<MediaStream> {
+  const id = deviceId?.trim();
+  const exactConstraints: MediaStreamConstraints = {
+    video: id ? { deviceId: { exact: id } } : true,
+    audio: false,
+  };
+
+  try {
+    return await navigator.mediaDevices.getUserMedia(exactConstraints);
+  } catch (err) {
+    if (!id || !isMediaConstraintFailure(err)) throw err;
+    console.warn(
+      "[clips-recorder] selected camera constraint failed; retrying default camera",
+      err,
+    );
+    return navigator.mediaDevices.getUserMedia({ video: true, audio: false });
+  }
+}
+
+export async function getAudioStreamWithFallback(
+  deviceId?: string | null,
+): Promise<MediaStream> {
+  const id = deviceId?.trim();
+
+  try {
+    return await navigator.mediaDevices.getUserMedia({
+      audio: voiceFocusedAudioConstraints(id),
+      video: false,
+    });
+  } catch (err) {
+    if (!isMediaConstraintFailure(err)) throw err;
+    if (id) {
+      console.warn(
+        "[clips-recorder] selected mic constraint failed; retrying default mic",
+        err,
+      );
+      try {
+        return await navigator.mediaDevices.getUserMedia({
+          audio: VOICE_FOCUSED_AUDIO_CONSTRAINTS,
+          video: false,
+        });
+      } catch (fallbackErr) {
+        if (!isMediaConstraintFailure(fallbackErr)) throw fallbackErr;
+        console.warn(
+          "[clips-recorder] voice-focused mic constraints failed; retrying basic audio",
+          fallbackErr,
+        );
+        return navigator.mediaDevices.getUserMedia({
+          audio: true,
+          video: false,
+        });
+      }
+    }
+
+    console.warn(
+      "[clips-recorder] voice-focused mic constraints failed; retrying basic audio",
+      err,
+    );
+    return navigator.mediaDevices.getUserMedia({
+      audio: true,
+      video: false,
+    });
+  }
+}

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

@@ -64,6 +64,11 @@ import {
   type LocalExportedFile,
   type LocalRecordingTarget,
 } from "./local-export";
+import {
+  buildDesktopDisplayMediaOptions,
+  getAudioStreamWithFallback,
+  getCameraStreamWithFallback,
+} from "./media-capture-constraints";
 import { buildCaptureTitle, type CaptureTitleResult } from "./recording-title";
 import {
   startTranscriptionCapture,
@@ -384,22 +389,6 @@ function createCloudMediaRecorder(
   throw lastError instanceof Error ? lastError : new Error(String(lastError));
 }
 
-const VOICE_FOCUSED_AUDIO_CONSTRAINTS: MediaTrackConstraints = {
-  echoCancellation: { ideal: true },
-  noiseSuppression: { ideal: true },
-  autoGainControl: { ideal: true },
-  channelCount: { ideal: 1 },
-};
-
-function voiceFocusedAudioConstraints(
-  deviceId?: string | null,
-): MediaTrackConstraints {
-  return {
-    ...VOICE_FOCUSED_AUDIO_CONSTRAINTS,
-    ...(deviceId ? { deviceId: { exact: deviceId } } : {}),
-  };
-}
-
 interface RecordingAudio {
   tracks: MediaStreamTrack[];
   cleanup: () => void;
@@ -444,13 +433,13 @@ function localRecordingTargetsForMode({
   localRecordingMode,
   displayStream,
   bubbleCameraStream,
-  audioStream,
+  recordingAudio,
   combined,
 }: {
   localRecordingMode: Exclude<LocalRecordingMode, "off">;
   displayStream: MediaStream | null;
   bubbleCameraStream: MediaStream | null;
-  audioStream: MediaStream | null;
+  recordingAudio: RecordingAudio;
   combined: MediaStream;
 }): LocalRecordingTarget[] {
   if (localRecordingMode === "composed") {
@@ -461,7 +450,7 @@ function localRecordingTargetsForMode({
   if (displayStream) {
     const desktopTracks = [
       ...displayStream.getVideoTracks(),
-      ...(audioStream?.getAudioTracks() ?? displayStream.getAudioTracks()),
+      ...recordingAudio.tracks,
     ];
     targets.push({
       role: "desktop",
@@ -1698,8 +1687,10 @@ async function startNativeFullscreenRecording(
   // clock and the toolbar-enable behind the real recording start.
   let startedAt = 0;
   let nativeTranscriptFailureSaved = false;
+  const wantsSystemAudio = params.systemAudioOn !== false;
+  const wantsRecordedAudio = wantsAudio || wantsSystemAudio;
   const saveTranscriptFailure = async (failureReason: string) => {
-    if (!wantsAudio || nativeTranscriptFailureSaved || !id) return;
+    if (!wantsRecordedAudio || nativeTranscriptFailureSaved || !id) return;
     nativeTranscriptFailureSaved = true;
     await saveRecordingTranscriptFailure(
       params.serverUrl,
@@ -1724,12 +1715,7 @@ async function startNativeFullscreenRecording(
     if (localOnly && localRecordingMode === "separate" && wantsCamera) {
       localCameraStream =
         params.preAcquiredCameraStream ??
-        (await navigator.mediaDevices.getUserMedia({
-          video: params.cameraId
-            ? { deviceId: { exact: params.cameraId } }
-            : true,
-          audio: false,
-        }));
+        (await getCameraStreamWithFallback(params.cameraId));
       localOwnsCameraStream =
         localCameraStream !== params.preAcquiredCameraStream;
       localCameraExport = await prepareLocalRecordingExport(
@@ -1757,16 +1743,12 @@ async function startNativeFullscreenRecording(
         ? "[clips-recorder] invoking show_countdown for native local recording"
         : "[clips-recorder] invoking show_countdown + createServerRecording",
     );
-    const wantsSystemAudio = wantsAudio && params.systemAudioOn !== false;
-    // Resolve the mic's REAL device name for the native recorder. WebKit's
-    // deviceId is a salted hash that never equals ScreenCaptureKit's CoreAudio
-    // device UID, so the Rust side can only pin the input by NAME. The stored
-    // label can be stale or empty (device list locked when picked, or a rotated
-    // deviceId salt after an app update) — that's what makes "only Default
-    // works": the hash matches nothing and there's no name to fall back to.
-    // A one-shot getUserMedia gives the exact current device name, the same
-    // string ScreenCaptureKit exposes, so name resolution succeeds. Done before
-    // warming so both phases pin the same input.
+    // Resolve the mic's REAL device name before native setup. WebKit's deviceId
+    // is a salted hash that never equals CoreAudio's device UID, so any native
+    // path that needs a mic can only pin the input by NAME. The stored label can
+    // be stale or empty (device list locked when picked, or a rotated deviceId
+    // salt after an app update), so a one-shot getUserMedia gives the exact
+    // current device name.
     let micDeviceLabel = params.micLabel || null;
     if (wantsAudio && params.micId) {
       try {
@@ -1797,11 +1779,10 @@ async function startNativeFullscreenRecording(
       micDeviceLabel,
       captureRegion,
     };
-    // Warm the mic DURING the countdown. ScreenCaptureKit delivers its first
-    // mic sample ~1s after capture starts; warming now lets `begin` attach the
-    // recording output to an already-live mic, so the clip no longer starts
-    // with a silent second. No-op when there's nothing to pre-warm (mic off /
-    // SCK unavailable) — `begin` then does a normal immediate start.
+    // Warm ScreenCaptureKit DURING the countdown without recording frames yet.
+    // This keeps the capture start off the critical path while letting `begin`
+    // attach the recording output at the exact start moment. No-op when SCK is
+    // unavailable — `begin` then does a normal immediate start.
     const warmMic = (recordingId: string) =>
       invoke("native_fullscreen_recording_warm", {
         recordingId,
@@ -1822,13 +1803,13 @@ async function startNativeFullscreenRecording(
       const recordingPromise = createServerRecording(
         params.serverUrl,
         wantsCamera,
-        wantsAudio,
+        wantsRecordedAudio,
         captureTitle,
       ).finally(() => {
         console.timeEnd("[clips-recorder] createServerRecording duration");
       });
       // The recording id usually lands well before the countdown ends — warm
-      // the mic as soon as it does so the warm-up overlaps the 3-2-1.
+      // native capture as soon as it does so setup overlaps the 3-2-1.
       const warmAndId = (async () => {
         const createRes = await recordingPromise;
         await warmMic(createRes.id);
@@ -2049,7 +2030,7 @@ async function startNativeFullscreenRecording(
             recordingId: id,
             authToken: params.authToken ?? "",
             cookie: params.cookie ?? "",
-            hasAudio: wantsAudio,
+            hasAudio: wantsRecordedAudio,
             hasCamera: wantsCamera,
           },
         );
@@ -2074,9 +2055,9 @@ async function startNativeFullscreenRecording(
               capturedTranscript,
               params.authToken,
             );
-          } else if (wantsAudio) {
+          } else if (wantsRecordedAudio) {
             await saveTranscriptFailure(
-              "No speech was captured during this recording. If you spoke, check Microphone input, Speech Recognition permission, and the selected mic, then retry transcription.",
+              "No speech was captured during this recording. If you spoke or played system audio, check System Audio, Microphone input, Speech Recognition permission, and the selected mic, then retry transcription.",
             );
           }
 
@@ -2265,8 +2246,7 @@ async function startNativeFullscreenRecording(
 
   if (!localOnly) {
     await showRegionGuidesForRecording(true);
-    const wantsSystemAudio = wantsAudio && params.systemAudioOn !== false;
-    transcriptionCapture = wantsAudio
+    transcriptionCapture = wantsRecordedAudio
       ? await startTranscriptionCapture(
           {
             deviceId: params.micId,
@@ -2288,12 +2268,12 @@ async function startNativeFullscreenRecording(
       );
       void transcriptionCapture.pause().catch(() => {});
     } else if (
-      wantsAudio &&
+      wantsRecordedAudio &&
       !transcriptionCapture &&
       shouldSaveLocalTranscriptionStartupFailure()
     ) {
       void saveTranscriptFailure(
-        "macOS Speech recognition could not start for this recording. Check Speech Recognition and Microphone permissions, then retry transcription.",
+        "macOS Speech recognition could not start for this recording. Check Speech Recognition, System Audio, and Microphone permissions, then retry transcription.",
       );
     }
   }
@@ -2409,7 +2389,8 @@ async function startRecordingInner(
   const wantsScreen = params.mode !== "camera";
   const wantsCamera = params.mode !== "screen" && params.cameraOn;
   const wantsAudio = params.micOn;
-  const wantsSystemAudio = wantsAudio && params.systemAudioOn !== false;
+  const wantsSystemAudio = wantsScreen && params.systemAudioOn !== false;
+  const wantsRecordedAudio = wantsAudio || wantsSystemAudio;
   const audioCue = createAudioCue();
   const captureSource = params.source ?? "window";
   const localRecordingMode = params.localRecordingMode ?? "off";
@@ -2421,6 +2402,7 @@ async function startRecordingInner(
     wantsScreen,
     wantsCamera,
     wantsAudio,
+    wantsSystemAudio,
   });
 
   if (wantsScreen && shouldUseNativeFullscreenRecording(captureSource)) {
@@ -2466,21 +2448,18 @@ async function startRecordingInner(
   const displayStreamPromise: Promise<MediaStream> | null = wantsScreen
     ? (() => {
         if (!devSyntheticCapture) {
-          const displaySurface =
-            captureSource === "window" ? "window" : "monitor";
-          return navigator.mediaDevices.getDisplayMedia({
-            video: {
-              frameRate: {
-                ideal: CLOUD_CAPTURE_FRAME_RATE,
-                max: CLOUD_CAPTURE_FRAME_RATE,
-              },
-              width: { ideal: CLOUD_CAPTURE_MAX_WIDTH },
-              height: { ideal: CLOUD_CAPTURE_MAX_HEIGHT },
-              displaySurface,
-            },
-            // System/desktop audio is gated by the system-audio toggle, not mic.
-            audio: wantsSystemAudio,
-          });
+          // Do not pass displaySurface as an input constraint. Modern runtimes
+          // can reject it with "Invalid constraint", and it cannot reliably
+          // pre-filter the OS picker anyway; the selected track reports its
+          // actual surface through getSettings() after capture starts.
+          return navigator.mediaDevices.getDisplayMedia(
+            buildDesktopDisplayMediaOptions({
+              audio: wantsSystemAudio,
+              frameRate: CLOUD_CAPTURE_FRAME_RATE,
+              maxWidth: CLOUD_CAPTURE_MAX_WIDTH,
+              maxHeight: CLOUD_CAPTURE_MAX_HEIGHT,
+            }),
+          );
         }
         console.warn(
           "[clips-recorder] using opt-in dev synthetic screen capture; remove localStorage clips:dev-synthetic-capture to use the native picker",
@@ -2506,18 +2485,10 @@ async function startRecordingInner(
   }
   const bubbleCameraStreamPromise: Promise<MediaStream> | null =
     wantsCamera && !reusedCameraStream
-      ? navigator.mediaDevices.getUserMedia({
-          video: params.cameraId
-            ? { deviceId: { exact: params.cameraId } }
-            : true,
-          audio: false,
-        })
+      ? getCameraStreamWithFallback(params.cameraId)
       : null;
   const audioStreamPromise: Promise<MediaStream> | null = wantsAudio
-    ? navigator.mediaDevices.getUserMedia({
-        audio: voiceFocusedAudioConstraints(params.micId),
-        video: false,
-      })
+    ? getAudioStreamWithFallback(params.micId)
     : null;
 
   // Use allSettled so a single rejection (e.g. user cancels the macOS screen
@@ -2700,7 +2671,7 @@ async function startRecordingInner(
       localRecordingMode,
       displayStream,
       bubbleCameraStream,
-      audioStream,
+      recordingAudio,
       combined,
     });
 
@@ -2896,7 +2867,7 @@ async function startRecordingInner(
   const recordingPromise = createServerRecording(
     params.serverUrl,
     wantsCamera,
-    wantsAudio,
+    recordingAudio.tracks.length > 0,
     captureTitle,
     { mimeType: mimeType || "video/webm", requestStreaming: true },
   ).finally(() => {
@@ -2922,7 +2893,7 @@ async function startRecordingInner(
   console.log("[clips-recorder] recording row created", { id, uploadMode });
   let nativeTranscriptFailureSaved = false;
   const saveTranscriptFailure = async (failureReason: string) => {
-    if (!wantsAudio || nativeTranscriptFailureSaved) return;
+    if (!wantsRecordedAudio || nativeTranscriptFailureSaved) return;
     nativeTranscriptFailureSaved = true;
     await saveRecordingTranscriptFailure(
       params.serverUrl,
@@ -3182,7 +3153,7 @@ async function startRecordingInner(
   // delaying capture, so the first ~1s the user expected to record was lost
   // (and the recording felt cut at the end). It's a separate capture from the
   // recorded audio tracks, so starting it slightly late is safe.
-  transcriptionCapture = wantsAudio
+  transcriptionCapture = wantsRecordedAudio
     ? await startTranscriptionCapture(
         {
           deviceId: params.micId,
@@ -3204,12 +3175,12 @@ async function startRecordingInner(
     );
     void transcriptionCapture.pause().catch(() => {});
   } else if (
-    wantsAudio &&
+    wantsRecordedAudio &&
     !transcriptionCapture &&
     shouldSaveLocalTranscriptionStartupFailure()
   ) {
     void saveTranscriptFailure(
-      "macOS Speech recognition could not start for this recording. Check Speech Recognition and Microphone permissions, then retry transcription.",
+      "macOS Speech recognition could not start for this recording. Check Speech Recognition, System Audio, and Microphone permissions, then retry transcription.",
     );
   }
 
@@ -3289,9 +3260,9 @@ async function startRecordingInner(
           capturedTranscript,
           params.authToken,
         );
-      } else if (wantsAudio) {
+      } else if (wantsRecordedAudio) {
         await saveTranscriptFailure(
-          "No speech was captured during this recording. If you spoke, check Microphone input, Speech Recognition permission, and the selected mic, then retry transcription.",
+          "No speech was captured during this recording. If you spoke or played system audio, check System Audio, Microphone input, Speech Recognition permission, and the selected mic, then retry transcription.",
         );
       }
       await thumbnailUploadPromise;