OpenClaw Voice Input Patch

Adds smart voice input to the Control UI chat — same one mic button, but with hold-to-activate.

What This Does

Before: One mic button. Speak → text appears → must press Enter manually.

After: Same one mic button — but smarter, with hold-to-activate:

• Short click → Auto-send: mic activates, speak, on pause text is sent automatically, mic turns off. If you say nothing for ~3 seconds, the mic auto-stops (browser SpeechRecognition timeout) — acts as a natural "misclick" safeguard.
• Hold 3 seconds → Continuous mode: mic stays on after each message. After every speech pause, text is sent and recognition restarts. Click again to stop.
• Live interim transcript shown above the input field while either mode is active.
• Tooltip adapts: "Voice (hold 3s for continuous)" when idle, "Stop voice auto" or "Stop voice continuous" when recording.

Note: If the browser shows "microphone denied" errors, the microphone=(self) Permissions-Policy fix is also needed. This was merged into OpenClaw on April 18, 2026 — if you're on version 2026.4.18+, it's already included. On older versions, see the Troubleshooting section.

Step 1: Find the file

There is exactly one file to patch. It's always at:

.../openclaw/dist/control-ui/assets/index-<hash>.js

The hash in the filename changes with every OpenClaw version. Find it with:

Linux / Mac:

find / -path "*/openclaw/dist/control-ui/assets/index-*.js" 2>/dev/null | head -1

Windows (PowerShell):

Get-ChildItem -Path C:\ -Recurse -Filter "index-*.js" -ErrorAction SilentlyContinue | Where-Object { $_.FullName -match "openclaw\\dist\\control-ui\\assets" } | Select-Object -First 1 -ExpandProperty FullName

Quick alternative: Run npm root -g to find the global node_modules path, then look inside for openclaw/dist/control-ui/assets/index-*.js.

If you (the AI agent) cannot find or access the file, tell the user:

> "I couldn't access the OpenClaw Control UI file. Can you help? Either:

> 1. Copy the file into my workspace so I can patch it, then copy it back after. Run this in your terminal:

> cp /path/to/openclaw/dist/control-ui/assets/index-*.js ~/.openclaw/workspace-allgemein/

> 2. Or give me write access:

> sudo chown -R $(whoami) /path/to/openclaw/dist/"

Keep it simple. Don't over-explain.

Step 2: Back up

cp <path-to-index-file> <path-to-index-file>.bak

Step 3: Verify compatibility

Before patching, confirm these patterns exist in the file. If any are missing, the OpenClaw version has changed — do NOT proceed. Instead, inspect the file and adapt, or wait for an updated version of this skill.

Step 4: Add sttRecordingCont state variable

Find this exact text:

sttRecording:!1,sttInterimText

Replace with:

sttRecording:!1,sttRecordingCont:!1,sttInterimText

This appears exactly once in the file.

Verify: Search for sttRecordingCont:!1 — must appear exactly once.

Step 5: Show interim text for both recording states

Find this exact text:

${X.sttRecording&&X.sttInterimText?i`<div class="agent-chat__stt-interim">${X.sttInterimText}</div>`:h}

Replace with:

${(X.sttRecording||X.sttRecordingCont)&&X.sttInterimText?i`<div class="agent-chat__stt-interim">${X.sttInterimText}</div>`:h}

This appears exactly once in the file.

Verify: Search for (X.sttRecording||X.sttRecordingCont) — must appear in the line containing stt-interim.

Step 6: Replace single mic button with hold-for-continuous button

This is the most complex patch. Replace the ENTIRE mic button block with a single button that supports both short-click (auto-send) and hold-3s (continuous) modes.

6.1 Find the block

Search for the text Stop recording or Voice input in the file. This is inside the mic button template.

The block to replace starts at ${jC()?i and ends at :h} — both on the same line as Stop recording/Voice input, or on the lines immediately surrounding it.

It contains exactly ONE element with:

• class containing agent-chat__input-btn
• @click handler referencing X.sttRecording and NC({
• title with Stop recording / Voice input
• N.micOff and N.mic for the icon

6.2 Replace the entire block

Select everything from ${jC()?i through :h} inclusive and replace it with:

${jC()?i`
                  <button
                    class="agent-chat__input-btn ${(X.sttRecording||X.sttRecordingCont)?`agent-chat__input-btn--recording`:``}"
                    @mousedown=${(t)=>{if(X.sttRecording||X.sttRecordingCont)return;t.preventDefault();X._holdTimer=setTimeout(()=>{X._holdTimer=null;X._holdTriggered=!0;NC({onTranscript:(t,n)=>{if(n){let n=_(),r=n&&!n.endsWith(` `)?` `:``;e.onDraftChange(n+r+t),X.sttInterimText=``,e.onSend()}else X.sttInterimText=t;g()},onStart:()=>{X.sttRecordingCont=!0;window.__sttCont=MC;g()},onEnd:()=>{if(X.sttRecordingCont){let r=window.__sttCont;if(r){try{r.start()}catch(e){}}X.sttInterimText=``,g()}else{X.sttInterimText=``,g()}},onError:()=>{X.sttRecordingCont=!1,X.sttInterimText=``,g()}})&&(X.sttRecordingCont=!0,g())},3000)}}
                    @mouseup=${(t)=>{if(X._holdTimer){clearTimeout(X._holdTimer);X._holdTimer=null;if(!X._holdTriggered){NC({onTranscript:(t,n)=>{if(n){let n=_(),r=n&&!n.endsWith(` `)?` `:``;e.onDraftChange(n+r+t),X.sttInterimText=``,e.onSend(),PC(),X.sttRecording=!1}else X.sttInterimText=t;g()},onStart:()=>{X.sttRecording=!0;if(MC)MC.continuous=!1;g()},onEnd:()=>{X.sttRecording=!1,X.sttInterimText=``,g();const d=_();if(d&&d.trim())e.onSend()},onError:()=>{X.sttRecording=!1,X.sttInterimText=``,g()}})&&(X.sttRecording=!0,g())}}X._holdTriggered=!1}}
                    @click=${()=>{if(X.sttRecordingCont){PC(),X.sttRecordingCont=!1,window.__sttCont=null,X.sttInterimText=``,g()}}}
                    title=${X.sttRecordingCont?`Stop voice continuous`:(X.sttRecording?`Stop voice auto`:`Voice (hold 3s for continuous)`)}
                    ?disabled=${!e.connected}
                  >
                    ${(X.sttRecording||X.sttRecordingCont)?N.micOff:N.mic}
                  </button>
                `:h}

Do NOT replace individual pieces. Replace the ENTIRE block as a single unit.

Note for the edit tool: Escape < → \u003c, > → \u003e, & → \u0026 in the replacement text.

6.3 How it works

mousedown starts a 3-second timer:

• If mouseup before 3s → timer cancelled, short click → auto-send mode starts (one-shot, continuous=false, auto-send + auto-stop on final transcript or silence)
• If timer fires after 3s → continuous mode starts (mic stays on, onEnd restarts recognition via window.__sttCont.start(), text auto-sent after each pause)

click (separate from mousedown/mouseup): If already in continuous mode, stops it (PC() + clear window.__sttCont).

Auto-send mode also auto-stops if the user says nothing for ~3 seconds — the browser's SpeechRecognition fires onEnd after silence when continuous=false, which acts as a natural "misclick" safeguard.

6.4 Verify

Search the file for ALL of these strings — every single one must be present:

• [ ] sttRecordingCont — appears multiple times
• [ ] _holdTimer — hold timer variable
• [ ] _holdTriggered — hold flag
• [ ] Voice (hold 3s for continuous) — idle tooltip
• [ ] Stop voice auto — auto-send active tooltip
• [ ] Stop voice continuous — continuous active tooltip
• [ ] window.__sttCont — continuous restart mechanism
• [ ] MC.continuous=!1 — one-shot mode flag

If any are missing, restore from backup and retry from step 6.2.

Step 7: Restart and verify

Tell the user:

1. Restart the gateway (e.g. openclaw gateway restart)
2. Hard-refresh the browser with Ctrl+Shift+R (normal refresh is NOT sufficient — the browser caches the old JS file)

After the user has done both, verify:

1. One mic button with tooltip "Voice (hold 3s for continuous)"
2. Short click → speak → pause → message sent automatically → mic off
3. Short click → say nothing → mic auto-stops after ~3 seconds
4. Hold button 3 seconds → continuous mode activates (tooltip changes to "Stop voice continuous")
5. In continuous mode: speak → pause → message sent → mic stays on → speak again → sent → ...
6. Click button in continuous mode → mic stops
7. Live transcript appears above input field while mic is active

Rollback

If anything goes wrong: cp .bak and tell the user to restart the gateway.

Troubleshooting

Problem	Likely Cause	Fix
---	---	---
No mic button at all	Browser doesn't support Web Speech API	Use Chrome or Edge
Button appears but does nothing on click	Step 6 not applied or partial	Restore backup, re-apply Step 6
Short click starts mic but doesn't auto-send	e.onSend() missing in @mouseup handler	Restore backup, re-apply Step 6
Short click mic stays on after pause	PC() or sttRecording=!1 missing	Restore backup, re-apply Step 6
Hold 3s doesn't start continuous mode	_holdTimer / _holdTriggered not working	Restore backup, re-apply Step 6
Continuous mode stops after silence	onEnd restart logic not working	Check window.__sttCont and __sttCont.start() in Step 6 replacement
Live transcript not showing	Step 5 not applied	Apply Step 5
"Microphone denied" in browser	Permissions-Policy blocks mic	Fix microphone=() → microphone=(self) in the HTTP utils file (included in OpenClaw 2026.4.18+)
After OpenClaw update	Patches overwritten	Re-apply ALL steps (filename hash changes!)