diff --git a/apps/x/apps/renderer/src/App.tsx b/apps/x/apps/renderer/src/App.tsx
index f9019ede..03e8e998 100644
--- a/apps/x/apps/renderer/src/App.tsx
+++ b/apps/x/apps/renderer/src/App.tsx
@@ -6193,6 +6193,7 @@ function App() {
                             isRecording={isActive && isRecording}
                             recordingText={isActive ? voice.interimText : undefined}
                             recordingState={isActive ? (voice.state === 'connecting' ? 'connecting' : 'listening') : undefined}
+                            audioLevelsRef={voice.audioLevelsRef}
                             onStartRecording={isActive ? handleStartRecording : undefined}
                             onSubmitRecording={isActive ? handleSubmitRecording : undefined}
                             onCancelRecording={isActive ? handleCancelRecording : undefined}
@@ -6301,6 +6302,7 @@ function App() {
                 isRecording={isRecording}
                 recordingText={voice.interimText}
                 recordingState={voice.state === 'connecting' ? 'connecting' : 'listening'}
+                audioLevelsRef={voice.audioLevelsRef}
                 onStartRecording={handleStartRecording}
                 onSubmitRecording={handleSubmitRecording}
                 onCancelRecording={handleCancelRecording}
diff --git a/apps/x/apps/renderer/src/components/chat-input-with-mentions.tsx b/apps/x/apps/renderer/src/components/chat-input-with-mentions.tsx
index 00bcaa8e..412b4bbf 100644
--- a/apps/x/apps/renderer/src/components/chat-input-with-mentions.tsx
+++ b/apps/x/apps/renderer/src/components/chat-input-with-mentions.tsx
@@ -224,6 +224,8 @@ interface ChatInputInnerProps {
   isRecording?: boolean
   recordingText?: string
   recordingState?: 'connecting' | 'listening'
+  /** Live mic amplitude history (RMS per frame) driving the recording waveform. */
+  audioLevelsRef?: React.MutableRefObject<number[]>
   onStartRecording?: () => void
   onSubmitRecording?: () => void
   onCancelRecording?: () => void
@@ -260,7 +262,7 @@ function ChatInputInner({
   onDraftChange,
   isRecording,
   recordingText,
-  recordingState,
+  audioLevelsRef,
   onStartRecording,
   onSubmitRecording,
   onCancelRecording,
@@ -795,11 +797,10 @@ function ChatInputInner({
           >
             <X className="h-4 w-4" />
           </button>
-          <div className="flex flex-1 items-center gap-2 overflow-hidden">
-            <VoiceWaveform />
-            <span className="min-w-0 flex-1 truncate text-sm text-muted-foreground">
-              {recordingState === 'connecting' ? 'Connecting...' : recordingText || 'Listening...'}
-            </span>
+          {/* Audio-reactive waveform only — the transcribed words are intentionally
+              not shown while recording; they're still captured and submitted. */}
+          <div className="flex flex-1 items-center overflow-hidden">
+            <VoiceWaveform audioLevelsRef={audioLevelsRef} />
           </div>
           <Button
             size="icon"
@@ -1339,22 +1340,89 @@ function ChatInputInner({
 }
 
 /** Animated waveform bars for the recording indicator */
-function VoiceWaveform() {
+// Live recording waveform. Each bar is one captured audio frame; bars accumulate
+// from the left and grow rightward until they fill the width, then scroll (oldest
+// drops off the left). Bar height tracks that frame's mic amplitude, so the
+// waveform visibly reacts to how loud the user is speaking.
+const WAVE_BAR_WIDTH = 3 // px
+const WAVE_BAR_GAP = 2 // px
+const WAVE_BAR_PITCH = WAVE_BAR_WIDTH + WAVE_BAR_GAP
+const WAVE_BAR_MIN = 1.5 // px — floor so silence still shows a faint line
+const WAVE_BAR_MAX = 18 // px — fits inside the h-5 (20px) row
+const WAVE_CURVE = 0.8 // <1 lifts quiet speech slightly; near-linear keeps loud peaks tall
+
+function waveBarHeight(level: number): number {
+  // `level` is already auto-gained to ~0..1 in the hook, so map it close to linearly
+  // (a gentle curve) — louder voice ⇒ visibly taller bar, quiet ⇒ short.
+  const amp = Math.min(1, Math.max(0, level)) ** WAVE_CURVE
+  return WAVE_BAR_MIN + amp * (WAVE_BAR_MAX - WAVE_BAR_MIN)
+}
+
+function VoiceWaveform({ audioLevelsRef }: { audioLevelsRef?: React.MutableRefObject<number[]> }) {
+  const containerRef = useRef<HTMLDivElement>(null)
+  const [bars, setBars] = useState<number[]>([])
+  // How many bars fit in the current width; recomputed on resize.
+  const maxBarsRef = useRef(48)
+
+  useEffect(() => {
+    const el = containerRef.current
+    if (!el) return
+    const measure = () => {
+      maxBarsRef.current = Math.max(1, Math.floor(el.clientWidth / WAVE_BAR_PITCH))
+    }
+    measure()
+    const ro = new ResizeObserver(measure)
+    ro.observe(el)
+    return () => ro.disconnect()
+  }, [])
+
+  useEffect(() => {
+    if (!audioLevelsRef) return
+    let raf = 0
+    let lastSig = ''
+    const tick = () => {
+      const levels = audioLevelsRef.current
+      const maxBars = maxBarsRef.current
+      const next = levels.length > maxBars ? levels.slice(levels.length - maxBars) : levels
+      // Only re-render when the visible window actually changed. Length covers
+      // the growth phase; the trailing value covers the scrolling phase once full.
+      const sig = `${next.length}:${next.length ? next[next.length - 1] : 0}`
+      if (sig !== lastSig) {
+        lastSig = sig
+        setBars(next.slice())
+      }
+      raf = requestAnimationFrame(tick)
+    }
+    raf = requestAnimationFrame(tick)
+    return () => cancelAnimationFrame(raf)
+  }, [audioLevelsRef])
+
   return (
-    <div className="flex items-center gap-[3px] h-5">
-      {[0, 1, 2, 3, 4].map((i) => (
+    <div
+      ref={containerRef}
+      className="flex h-5 w-full items-center overflow-hidden"
+      style={{ gap: `${WAVE_BAR_GAP}px` }}
+    >
+      {/* Each newly-appended bar mounts with `voice-bar-in` (grows + fades in) so it
+          doesn't pop. Once the strip is full and values scroll through the bars, the
+          height transition makes them flow smoothly instead of stepping. */}
+      {bars.map((level, i) => (
         <span
           key={i}
-          className="w-[3px] rounded-full bg-primary"
+          className="shrink-0 rounded-full bg-primary"
           style={{
-            animation: `voice-wave 1.2s ease-in-out ${i * 0.15}s infinite`,
+            width: `${WAVE_BAR_WIDTH}px`,
+            height: `${waveBarHeight(level)}px`,
+            transformOrigin: 'center',
+            transition: 'height 90ms linear',
+            animation: 'voice-bar-in 130ms ease-out',
           }}
         />
       ))}
       <style>{`
-        @keyframes voice-wave {
-          0%, 100% { height: 4px; }
-          50% { height: 16px; }
+        @keyframes voice-bar-in {
+          from { transform: scaleY(0.15); opacity: 0; }
+          to { transform: scaleY(1); opacity: 1; }
         }
       `}</style>
     </div>
@@ -1378,6 +1446,7 @@ export interface ChatInputWithMentionsProps {
   isRecording?: boolean
   recordingText?: string
   recordingState?: 'connecting' | 'listening'
+  audioLevelsRef?: React.MutableRefObject<number[]>
   onStartRecording?: () => void
   onSubmitRecording?: () => void
   onCancelRecording?: () => void
@@ -1411,6 +1480,7 @@ export function ChatInputWithMentions({
   isRecording,
   recordingText,
   recordingState,
+  audioLevelsRef,
   onStartRecording,
   onSubmitRecording,
   onCancelRecording,
@@ -1441,6 +1511,7 @@ export function ChatInputWithMentions({
         isRecording={isRecording}
         recordingText={recordingText}
         recordingState={recordingState}
+        audioLevelsRef={audioLevelsRef}
         onStartRecording={onStartRecording}
         onSubmitRecording={onSubmitRecording}
         onCancelRecording={onCancelRecording}
diff --git a/apps/x/apps/renderer/src/components/chat-sidebar.tsx b/apps/x/apps/renderer/src/components/chat-sidebar.tsx
index 0f89570f..f0f29439 100644
--- a/apps/x/apps/renderer/src/components/chat-sidebar.tsx
+++ b/apps/x/apps/renderer/src/components/chat-sidebar.tsx
@@ -178,6 +178,7 @@ interface ChatSidebarProps {
   isRecording?: boolean
   recordingText?: string
   recordingState?: 'connecting' | 'listening'
+  audioLevelsRef?: React.MutableRefObject<number[]>
   onStartRecording?: () => void
   onSubmitRecording?: () => void
   onCancelRecording?: () => void
@@ -240,6 +241,7 @@ export function ChatSidebar({
   isRecording,
   recordingText,
   recordingState,
+  audioLevelsRef,
   onStartRecording,
   onSubmitRecording,
   onCancelRecording,
@@ -811,6 +813,7 @@ export function ChatSidebar({
                           isRecording={isActive && isRecording}
                           recordingText={isActive ? recordingText : undefined}
                           recordingState={isActive ? recordingState : undefined}
+                          audioLevelsRef={audioLevelsRef}
                           onStartRecording={isActive ? onStartRecording : undefined}
                           onSubmitRecording={isActive ? onSubmitRecording : undefined}
                           onCancelRecording={isActive ? onCancelRecording : undefined}
diff --git a/apps/x/apps/renderer/src/hooks/useVoiceMode.ts b/apps/x/apps/renderer/src/hooks/useVoiceMode.ts
index fd13c758..8263e45d 100644
--- a/apps/x/apps/renderer/src/hooks/useVoiceMode.ts
+++ b/apps/x/apps/renderer/src/hooks/useVoiceMode.ts
@@ -20,6 +20,17 @@ const DEEPGRAM_PARAMS = new URLSearchParams({
 });
 const DEEPGRAM_LISTEN_URL = `wss://api.deepgram.com/v1/listen?${DEEPGRAM_PARAMS.toString()}`;
 
+// Cap on retained per-frame amplitude samples (~64ms/frame ⇒ ~5 min of history).
+// The waveform only ever displays the most recent window, so older samples are dropped.
+const MAX_AUDIO_LEVELS = 4800;
+
+// Auto-gain for the waveform: each frame's amplitude is stored normalized against a
+// running peak (instant attack, slow release) so bar heights track the *relative*
+// loudness of the voice accurately regardless of mic/OS input gain. MIN_PEAK is a
+// floor so near-silence doesn't get amplified up into tall bars.
+const PEAK_DECAY = 0.97;
+const MIN_PEAK = 0.02;
+
 // Cache auth details so we don't need IPC round-trips on every mic click
 let cachedAuth: { type: 'rowboat'; url: string; token: string } | { type: 'local'; apiKey: string } | null = null;
 
@@ -35,6 +46,12 @@ export function useVoiceMode() {
     const interimRef = useRef('');
     // Buffer audio chunks captured before the WebSocket is ready
     const audioBufferRef = useRef<ArrayBuffer[]>([]);
+    // Rolling history of per-frame mic amplitude (auto-gained to 0..1), oldest first.
+    // Drives the live waveform — the UI reads this via requestAnimationFrame so
+    // amplitude updates never re-render the rest of the tree.
+    const audioLevelsRef = useRef<number[]>([]);
+    // Running peak amplitude for the waveform auto-gain (see PEAK_DECAY/MIN_PEAK).
+    const audioPeakRef = useRef(0);
 
     // Refresh cached auth details (called on warmup, not on mic click)
     const refreshAuth = useCallback(async () => {
@@ -132,6 +149,8 @@ export function useVoiceMode() {
             wsRef.current = null;
         }
         audioBufferRef.current = [];
+        audioLevelsRef.current = [];
+        audioPeakRef.current = 0;
         setInterimText('');
         transcriptBufferRef.current = '';
         interimRef.current = '';
@@ -145,6 +164,8 @@ export function useVoiceMode() {
         interimRef.current = '';
         setInterimText('');
         audioBufferRef.current = [];
+        audioLevelsRef.current = [];
+        audioPeakRef.current = 0;
 
         // Show listening immediately — don't wait for WebSocket
         setState('listening');
@@ -188,15 +209,32 @@ export function useVoiceMode() {
         const audioCtx = new AudioContext({ sampleRate: 16000 });
         audioCtxRef.current = audioCtx;
         const source = audioCtx.createMediaStreamSource(stream);
-        const processor = audioCtx.createScriptProcessor(2048, 1, 1);
+        // 1024-sample frames (~64ms at 16kHz) — smaller than the usual 2048 so the
+        // waveform gets ~16 amplitude updates/sec, making bars appear faster and
+        // flow more smoothly. Still a comfortable chunk size for Deepgram streaming.
+        const processor = audioCtx.createScriptProcessor(1024, 1, 1);
         processorRef.current = processor;
 
         processor.onaudioprocess = (e) => {
             const float32 = e.inputBuffer.getChannelData(0);
             const int16 = new Int16Array(float32.length);
+            let sumSquares = 0;
             for (let i = 0; i < float32.length; i++) {
                 const s = Math.max(-1, Math.min(1, float32[i]));
                 int16[i] = s < 0 ? s * 0x8000 : s * 0x7fff;
+                sumSquares += s * s;
+            }
+            // Record this frame's loudness for the live waveform, auto-gained against
+            // a running peak so bar heights accurately reflect the voice's dynamics.
+            // Instant attack (a louder frame raises the peak immediately), slow
+            // release (PEAK_DECAY), floored at MIN_PEAK so silence stays flat.
+            const rms = Math.sqrt(sumSquares / float32.length);
+            const peak = Math.max(rms, audioPeakRef.current * PEAK_DECAY, MIN_PEAK);
+            audioPeakRef.current = peak;
+            const levels = audioLevelsRef.current;
+            levels.push(rms / peak);
+            if (levels.length > MAX_AUDIO_LEVELS) {
+                levels.splice(0, levels.length - MAX_AUDIO_LEVELS);
             }
             const buffer = int16.buffer;
             if (wsRef.current?.readyState === WebSocket.OPEN) {
@@ -232,5 +270,5 @@ export function useVoiceMode() {
         refreshAuth().catch(() => {});
     }, [refreshAuth]);
 
-    return { state, interimText, start, submit, cancel, warmup };
+    return { state, interimText, audioLevelsRef, start, submit, cancel, warmup };
 }