diff --git a/src/lotus/lotusActions.test.ts b/src/lotus/lotusActions.test.ts
index d4b9401e..e1b8a9a9 100644
--- a/src/lotus/lotusActions.test.ts
+++ b/src/lotus/lotusActions.test.ts
@@ -50,5 +50,6 @@ describe("LotusWidgetActions", () => {
     expect(LOTUS_TO_WIDGET_ACTIONS).not.toContain(
       LotusWidgetActions.ControlsState,
     );
+    expect(LOTUS_TO_WIDGET_ACTIONS).not.toContain(LotusWidgetActions.MicLevel);
   });
 });
diff --git a/src/lotus/lotusActions.ts b/src/lotus/lotusActions.ts
index 1ffed38f..67e31bdc 100644
--- a/src/lotus/lotusActions.ts
+++ b/src/lotus/lotusActions.ts
@@ -80,6 +80,11 @@ export enum LotusWidgetActions {
    * them. Its arrival also tells the host this fork supports the actions above.
    */
   ControlsState = "io.lotus.controls_state",
+  /**
+   * fromWidget: local mic level `{ bars: 0 | 1 | 2 | 3 }` (cinny #146), sent
+   * only when it changes (≤ 10 Hz); 0 while muted or with no mic.
+   */
+  MicLevel = "io.lotus.mic_level",
 }
 
 /** toWidget Lotus actions that must be allow-listed in `initializeWidget`. */
diff --git a/src/lotus/lotusMicLevel.test.ts b/src/lotus/lotusMicLevel.test.ts
new file mode 100644
index 00000000..239ad608
--- /dev/null
+++ b/src/lotus/lotusMicLevel.test.ts
@@ -0,0 +1,37 @@
+/*
+Copyright 2026 Lotus Guild
+
+SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-Element-Commercial
+Please see LICENSE in the repository root for full details.
+*/
+
+import { describe, expect, it } from "vitest";
+
+import { MicLevelQuantizer } from "./lotusMicLevel";
+
+describe("MicLevelQuantizer", () => {
+  it("maps RMS to 0–3 bars", () => {
+    expect(new MicLevelQuantizer().push(0.001)).toBe(0);
+    expect(new MicLevelQuantizer().push(0.008)).toBe(1);
+    expect(new MicLevelQuantizer().push(0.02)).toBe(2);
+    expect(new MicLevelQuantizer().push(0.2)).toBe(3);
+  });
+
+  it("rises at once and falls one bar per sample", () => {
+    const q = new MicLevelQuantizer();
+    expect(q.push(0.2)).toBe(3);
+    expect(q.push(0)).toBe(2);
+    expect(q.push(0)).toBe(1);
+    expect(q.push(0.02)).toBe(2);
+    expect(q.push(0)).toBe(1);
+    expect(q.push(0)).toBe(0);
+    expect(q.push(0)).toBe(0);
+  });
+
+  it("reset drops straight to 0", () => {
+    const q = new MicLevelQuantizer();
+    q.push(0.2);
+    q.reset();
+    expect(q.push(0)).toBe(0);
+  });
+});
diff --git a/src/lotus/lotusMicLevel.ts b/src/lotus/lotusMicLevel.ts
new file mode 100644
index 00000000..6fc2f298
--- /dev/null
+++ b/src/lotus/lotusMicLevel.ts
@@ -0,0 +1,192 @@
+/*
+Copyright 2026 Lotus Guild
+
+SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-Element-Commercial
+Please see LICENSE in the repository root for full details.
+*/
+
+import {
+  type LocalTrackPublication,
+  type Room as LivekitRoom,
+  RoomEvent,
+  Track,
+} from "livekit-client";
+import { Observable, type Subscription, share, switchMap } from "rxjs";
+import { distinctUntilChanged, map } from "rxjs/operators";
+
+import { type CallViewModel } from "../state/CallViewModel/CallViewModel";
+import { widget } from "../widget";
+import { LotusWidgetActions, lotusFlag, lotusSendToHost } from "./lotusWidget";
+
+/**
+ * [lotus #146] One local mic sampler shared by the host's mic level meter
+ * (`io.lotus.mic_level`) and "talking while muted" (#37, lotusMutedSpeech).
+ * It taps a CLONE of the published mic track (the post-processor track when
+ * the in-source denoiser is active, so what's measured is what's sent) and
+ * reads RMS at ~10 Hz while a mic track is published, muted or not.
+ * Local only: nothing here reaches other participants.
+ */
+
+const SAMPLE_MS = 100;
+
+export interface LocalMicSample {
+  rms: number;
+  /** The mic is published but muted (the clone still hears it). */
+  muted: boolean;
+}
+
+const micPublication = (
+  room: LivekitRoom,
+): { track: MediaStreamTrack; muted: boolean } | null => {
+  const pub: LocalTrackPublication | undefined =
+    room.localParticipant.getTrackPublication(Track.Source.Microphone);
+  const track = pub?.track?.mediaStreamTrack;
+  return track && track.readyState === "live"
+    ? { track, muted: pub?.isMuted ?? false }
+    : null;
+};
+
+const samplers = new WeakMap<
+  CallViewModel,
+  Observable<LocalMicSample | null>
+>();
+
+/**
+ * RMS samples of the local mic, or `null` while no mic track is published.
+ * Shared per call view model, so the meter and the muted-speech detector use
+ * one AudioContext between them.
+ */
+export function observeLocalMicSample$(
+  vm: CallViewModel,
+): Observable<LocalMicSample | null> {
+  const cached = samplers.get(vm);
+  if (cached) return cached;
+  const sampler$ = vm.allConnections$.pipe(
+    switchMap(
+      (data) =>
+        new Observable<LocalMicSample | null>((subscriber) => {
+          const rooms = data.getConnections().map((c) => c.livekitRoom);
+          let ctx: AudioContext | null = null;
+          let clone: MediaStreamTrack | null = null;
+          let timer: ReturnType<typeof setInterval> | undefined;
+          let tapped: MediaStreamTrack | null = null;
+          let muted = false;
+
+          const stopTap = (): void => {
+            if (timer !== undefined) clearInterval(timer);
+            timer = undefined;
+            clone?.stop();
+            clone = null;
+            void ctx?.close().catch(() => undefined);
+            ctx = null;
+            tapped = null;
+            subscriber.next(null);
+          };
+
+          const startTap = (source: MediaStreamTrack): void => {
+            try {
+              clone = source.clone();
+              // Muting disables the published track; the clone must still hear.
+              clone.enabled = true;
+              ctx = new AudioContext();
+              const analyser = ctx.createAnalyser();
+              analyser.fftSize = 1024;
+              ctx
+                .createMediaStreamSource(new MediaStream([clone]))
+                .connect(analyser);
+              const buf = new Float32Array(analyser.fftSize);
+              tapped = source;
+              timer = setInterval(() => {
+                analyser.getFloatTimeDomainData(buf);
+                let sum = 0;
+                for (let i = 0; i < buf.length; i += 1) sum += buf[i] * buf[i];
+                subscriber.next({ rms: Math.sqrt(sum / buf.length), muted });
+              }, SAMPLE_MS);
+            } catch {
+              stopTap();
+            }
+          };
+
+          const reconcile = (): void => {
+            const pub =
+              rooms.map(micPublication).find((p) => p !== null) ?? null;
+            muted = pub?.muted ?? false;
+            const track = pub?.track ?? null;
+            if (track === tapped) return;
+            if (tapped) stopTap();
+            if (track) startTap(track);
+          };
+
+          const events = [
+            RoomEvent.TrackMuted,
+            RoomEvent.TrackUnmuted,
+            RoomEvent.LocalTrackPublished,
+            RoomEvent.LocalTrackUnpublished,
+            RoomEvent.Reconnected,
+            RoomEvent.Disconnected,
+          ] as const;
+          rooms.forEach((room) =>
+            events.forEach((ev) => room.on(ev, reconcile)),
+          );
+          subscriber.next(null);
+          reconcile();
+          return (): void => {
+            rooms.forEach((room) =>
+              events.forEach((ev) => room.off(ev, reconcile)),
+            );
+            if (tapped) stopTap();
+          };
+        }),
+    ),
+    share(),
+  );
+  samplers.set(vm, sampler$);
+  return sampler$;
+}
+
+/** RMS at which each bar lights: ≈ −46, −36 and −26 dBFS. */
+export const BAR_THRESHOLDS = [0.005, 0.015, 0.05] as const;
+
+/**
+ * Quantise RMS to 0–3 bars with a little hysteresis: rises at once, falls one
+ * bar per sample, so the meter doesn't flicker between words. Unit-tested.
+ */
+export class MicLevelQuantizer {
+  private bars = 0;
+
+  public reset(): void {
+    this.bars = 0;
+  }
+
+  public push(rms: number): number {
+    const target = BAR_THRESHOLDS.filter((t) => rms >= t).length;
+    this.bars = target >= this.bars ? target : this.bars - 1;
+    return this.bars;
+  }
+}
+
+/**
+ * Send the host `io.lotus.mic_level { bars }` (0–3) whenever the quantised
+ * level changes; 0 while muted or with no mic. At most one message per sample
+ * (10 Hz) and none during steady silence. Opt-in with the rest of the host
+ * state stream (`lotusCallState=1`). Returns a teardown function.
+ */
+export function startLotusMicLevel(vm: CallViewModel): () => void {
+  if (!lotusFlag("lotusCallState") || !widget) return (): void => undefined;
+  const quantizer = new MicLevelQuantizer();
+  const sub: Subscription = observeLocalMicSample$(vm)
+    .pipe(
+      map((sample) => {
+        if (!sample || sample.muted) {
+          quantizer.reset();
+          return 0;
+        }
+        return quantizer.push(sample.rms);
+      }),
+      distinctUntilChanged(),
+    )
+    .subscribe((bars) => {
+      lotusSendToHost(LotusWidgetActions.MicLevel, { bars });
+    });
+  return (): void => sub.unsubscribe();
+}
diff --git a/src/lotus/lotusMutedSpeech.ts b/src/lotus/lotusMutedSpeech.ts
index 14285886..938ba648 100644
--- a/src/lotus/lotusMutedSpeech.ts
+++ b/src/lotus/lotusMutedSpeech.ts
@@ -6,14 +6,15 @@ Please see LICENSE in the repository root for full details.
 */
 
 import {
-  type LocalTrackPublication,
-  type Room as LivekitRoom,
-  RoomEvent,
-  Track,
-} from "livekit-client";
-import { Observable, distinctUntilChanged, switchMap } from "rxjs";
+  type Observable,
+  distinctUntilChanged,
+  map,
+  scan,
+  startWith,
+} from "rxjs";
 
 import { type CallViewModel } from "../state/CallViewModel/CallViewModel";
+import { type LocalMicSample, observeLocalMicSample$ } from "./lotusMicLevel";
 
 /**
  * [lotus #37] "Talking while muted" detection for the LOCAL participant.
@@ -23,13 +24,12 @@ import { type CallViewModel } from "../state/CallViewModel/CallViewModel";
  * tells the user they are talking into a muted mic. This taps a CLONE of the
  * published track (the post-processor track when the in-source denoiser is
  * active, so keyboard noise doesn't count), samples RMS at ~10 Hz and emits a
- * debounced boolean. Zero cost when unmuted (tap torn down), local-only —
+ * debounced boolean. Local-only —
  * the flag rides `io.lotus.call_state` to the host and never reaches other
  * participants.
  */
 
 export const MUTED_SPEECH_RMS = 0.015; // ≈ −36 dBFS; normal speech into a headset is 0.05–0.3
-const SAMPLE_MS = 100;
 const ON_SAMPLES = 3; // 300 ms of voice before we say "talking"
 const OFF_SAMPLES = 8; // 800 ms of quiet before we drop it
 
@@ -61,94 +61,24 @@ export class MutedSpeechGate {
   }
 }
 
-const mutedMicTrack = (room: LivekitRoom): MediaStreamTrack | null => {
-  const pub: LocalTrackPublication | undefined =
-    room.localParticipant.getTrackPublication(Track.Source.Microphone);
-  const track = pub?.track?.mediaStreamTrack;
-  return pub?.isMuted && track && track.readyState === "live" ? track : null;
-};
-
 /**
  * Emits while the local mic is muted and voice is detected on it. Emits
  * `false` whenever the mic is unmuted, unpublished or the connection changes.
+ * [lotus #146] Fed by the shared local mic sampler (lotusMicLevel.ts), which
+ * also drives the host's mic level meter while unmuted.
  */
 export function observeSpeakingWhileMuted$(
   vm: CallViewModel,
 ): Observable<boolean> {
-  return vm.allConnections$.pipe(
-    switchMap(
-      (data) =>
-        new Observable<boolean>((subscriber) => {
-          const rooms = data.getConnections().map((c) => c.livekitRoom);
-          let ctx: AudioContext | null = null;
-          let clone: MediaStreamTrack | null = null;
-          let timer: ReturnType<typeof setInterval> | undefined;
-          let tapped: MediaStreamTrack | null = null;
-
-          const stopTap = (): void => {
-            if (timer !== undefined) clearInterval(timer);
-            timer = undefined;
-            clone?.stop();
-            clone = null;
-            void ctx?.close().catch(() => undefined);
-            ctx = null;
-            tapped = null;
-            subscriber.next(false);
-          };
-
-          const startTap = (source: MediaStreamTrack): void => {
-            try {
-              clone = source.clone();
-              clone.enabled = true; // the source is disabled by the mute — the clone must not be
-              ctx = new AudioContext();
-              const analyser = ctx.createAnalyser();
-              analyser.fftSize = 1024;
-              ctx
-                .createMediaStreamSource(new MediaStream([clone]))
-                .connect(analyser);
-              const buf = new Float32Array(analyser.fftSize);
-              const gate = new MutedSpeechGate();
-              tapped = source;
-              timer = setInterval(() => {
-                analyser.getFloatTimeDomainData(buf);
-                let sum = 0;
-                for (let i = 0; i < buf.length; i += 1) sum += buf[i] * buf[i];
-                subscriber.next(gate.push(Math.sqrt(sum / buf.length)));
-              }, SAMPLE_MS);
-            } catch {
-              stopTap();
-            }
-          };
-
-          const reconcile = (): void => {
-            const track =
-              rooms.map(mutedMicTrack).find((t) => t !== null) ?? null;
-            if (track === tapped) return;
-            if (tapped) stopTap();
-            if (track) startTap(track);
-          };
-
-          const events = [
-            RoomEvent.TrackMuted,
-            RoomEvent.TrackUnmuted,
-            RoomEvent.LocalTrackPublished,
-            RoomEvent.LocalTrackUnpublished,
-            RoomEvent.Reconnected,
-            RoomEvent.Disconnected,
-          ] as const;
-          rooms.forEach((room) =>
-            events.forEach((ev) => room.on(ev, reconcile)),
-          );
-          subscriber.next(false);
-          reconcile();
-          return () => {
-            rooms.forEach((room) =>
-              events.forEach((ev) => room.off(ev, reconcile)),
-            );
-            if (tapped) stopTap();
-          };
-        }),
-    ),
+  return observeLocalMicSample$(vm).pipe(
+    scan((gate: MutedSpeechGate | null, sample: LocalMicSample | null) => {
+      if (!sample?.muted) return null;
+      const g = gate ?? new MutedSpeechGate();
+      g.push(sample.rms);
+      return g;
+    }, null),
+    map((gate) => gate?.value ?? false),
+    startWith(false),
     distinctUntilChanged(),
   );
 }
diff --git a/src/room/InCallView.tsx b/src/room/InCallView.tsx
index c669bc97..8b4199a0 100644
--- a/src/room/InCallView.tsx
+++ b/src/room/InCallView.tsx
@@ -32,6 +32,7 @@ import { widget } from "../widget";
 import { startLotusCallState } from "../lotus/lotusCallState";
 import { startLotusFocus } from "../lotus/lotusFocus";
 import { startLotusControls } from "../lotus/lotusControls";
+import { startLotusMicLevel } from "../lotus/lotusMicLevel";
 import { startLotusAudioInject } from "../lotus/lotusAudioInject";
 import { startLotusQuality } from "../lotus/lotusQuality";
 import { startLotusDecorations } from "../lotus/lotusDecorations";
@@ -304,6 +305,8 @@ export const InCallView: FC<InCallViewProps> = ({
   // [cinny #43] layout / settings / reactions over the widget API, plus a
   // screensharing + layout report, replacing the host's DOM access.
   useEffect(() => startLotusControls(vm), [vm]);
+  // [cinny #146] Local mic level for the host's mute-button meter.
+  useEffect(() => startLotusMicLevel(vm), [vm]);
   // [lotus] Handle the host's io.lotus.inject_audio action to mix a soundboard
   // clip into the call as a separate track (#3). No-op unless the host sends it.
   useEffect(() => startLotusAudioInject(vm), [vm]);