Add local speech with optional cloud voice

NOTICE

@@ -67,3 +67,22 @@ The browser component is also published as
 https://github.com/TencentCloud/octop-browser
 Copyright (c) 2026 Octop
 Licensed under the MIT License.
+
+Jonathan Ai voice is adapted from Octop's voice manager: an active
+speech-to-text provider, an active text-to-speech provider, and a single
+place that dispatches a recording or a spoken reply. Jonathan Ai does not
+vendor Octop's browser, Edge, Tencent, or MiMo presets. The default
+speech-to-text engine is local faster-whisper and the default voice is
+local Piper. OpenAI and xAI are optional and run only after they are
+selected. Microphones and speakers, including Bluetooth headsets, are
+chosen from the Windows WASAPI devices or the Linux PulseAudio and
+PipeWire devices. Push-to-talk is the default. The wake word is off until
+the user turns it on, and that check stays on this machine. Model files
+are fetched from Hugging Face or the official Piper and Whisper sources
+and checked with SHA-256. The optional install extra is voice
+(pip install "clawd-codex[voice]").
+
+Octop voice
+https://github.com/TencentCloud/Octop
+Copyright (c) 2026 Octop
+Licensed under the MIT License.

pyproject.toml

@@ -43,6 +43,11 @@ dev = [
 browser = [
     "websocket-client>=1.7.0",
 ]
+voice = [
+    "faster-whisper>=1.0.0",
+    "sounddevice>=0.4.6",
+    "piper-tts>=1.2.0",
+]
 
 [project.scripts]
 clawd = "src.cli:main"

src/cli.py

@@ -34,6 +34,7 @@ def main():
   clawd memory delete <id> Delete one local memory
   clawd cron list          List local scheduled jobs
   clawd cron run <id>      Run one scheduled job now
+  clawd voice settings     Show local speech settings
   clawd                    Start interactive REPL
 """
     )
@@ -106,6 +107,9 @@ def main():
     cron_logs = cron_sub.add_parser('logs', help='Show scheduler logs')
     cron_logs.add_argument('job_id', nargs='?')
 
+    voice_parser = subparsers.add_parser('voice', help='Local speech, push-to-talk, and devices')
+    voice_parser.add_argument('words', nargs='*', help='settings, devices, stt, tts, ptt, wake, say')
+
     args = parser.parse_args()
 
     # Handle --version
@@ -183,6 +187,8 @@ def main():
         return handle_memory(args)
     elif args.command == 'cron':
         return handle_cron(args)
+    elif args.command == 'voice':
+        return handle_voice(args)
 
     # Default: start REPL
     return start_repl(stream=args.stream)
@@ -293,6 +299,15 @@ def handle_cron(args) -> int:
     return 0
 
 
+def handle_voice(args) -> int:
+    """Show or change speech settings stored on this machine."""
+    from src.voice.service import format_voice_command
+
+    words = getattr(args, "words", None) or []
+    print(format_voice_command(" ".join(words)))
+    return 0
+
+
 def start_repl(stream: bool = False):
     """Start interactive REPL."""
     from src.config import get_default_provider

src/command_system/__init__.py

@@ -15,6 +15,7 @@
     CRON_COMMAND,
     INIT_COMMAND,
     MEMORY_COMMAND,
+    VOICE_COMMAND,
     SKILLS_COMMAND,
     execute_command_async,
     execute_command_sync,
@@ -94,6 +95,7 @@
     "COMPACT_COMMAND",
     "CRON_COMMAND",
     "MEMORY_COMMAND",
+    "VOICE_COMMAND",
     "INIT_COMMAND",
     "get_builtin_commands",
     "register_builtin_commands",

src/command_system/builtins.py

@@ -636,6 +636,22 @@ def cron_command_call(args: str, context: CommandContext) -> LocalCommandResult:
 
     return LocalCommandResult(type="text", value=format_cron_command(args))
 
+
+VOICE_COMMAND = LocalCommand(
+    name="voice",
+    description="Listen and speak on this machine, or choose a cloud voice",
+    argument_hint="settings | devices | stt local|openai|xai | tts local|openai|xai | ptt on|off | wake on|off | say <text>",
+    supports_non_interactive=True,
+)
+
+
+def voice_command_call(args: str, context: CommandContext) -> LocalCommandResult:
+    """Show or change local speech settings. Audio stays here unless a cloud provider is selected."""
+    del context
+    from ..voice.service import format_voice_command
+
+    return LocalCommandResult(type="text", value=format_voice_command(args))
+
 INIT_COMMAND = PromptCommand(
     name="init",
     description="Initialize new CLAUDE.md file(s) and optional skills/hooks with codebase documentation",
@@ -684,6 +700,8 @@ def execute_command_sync(cmd_name: str, args: str, context: CommandContext) -> t
             result = memory_command_call(args, context)
         elif cmd is CRON_COMMAND:
             result = cron_command_call(args, context)
+        elif cmd is VOICE_COMMAND:
+            result = voice_command_call(args, context)
         else:
             return False, None, f"Command not implemented for sync execution: {cmd_name}"
 
@@ -702,6 +720,7 @@ def execute_command_sync(cmd_name: str, args: str, context: CommandContext) -> t
 COMPACT_COMMAND.set_call(compact_command_call)
 MEMORY_COMMAND.set_call(memory_command_call)
 CRON_COMMAND.set_call(cron_command_call)
+VOICE_COMMAND.set_call(voice_command_call)
 
 
 def get_builtin_commands() -> list[Command]:
@@ -716,6 +735,7 @@ def get_builtin_commands() -> list[Command]:
         COMPACT_COMMAND,
         MEMORY_COMMAND,
         CRON_COMMAND,
+        VOICE_COMMAND,
         INIT_COMMAND,
     ]
 

src/desktop/runtime.py

@@ -1173,6 +1173,7 @@ def _format_palette(self) -> str:
 - `/memory` — list, recall, or delete local memories (`/memory delete <id>`)
 - `/cron` — list, run, enable, disable, or delete local schedules (`/cron run <id>`)
 - Browser tool — local Chrome over CDP. Logins stay in the profile. Submits, purchases, and logins need approval.
+- `/voice` — local speech and speakers (`/voice settings`, `/voice say hello`)
 
 Type a message to run the existing agent loop. Destructive and network tools
 prompt for approval before they run.

src/desktop/server.py

@@ -231,6 +231,18 @@ def handle_api(self, method: str, path: str, query: dict[str, list[str]], body:
             return 200, runtime.update_status(refresh=True)
         if path == "/api/update/apply" and method == "POST":
             return 200, runtime.apply_update()
+        if path == "/api/voice" and method == "GET":
+            from src.voice.service import voice_status
+
+            return 200, voice_status()
+        if path == "/api/voice" and method == "POST":
+            from src.voice.service import apply_voice_settings
+
+            return 200, apply_voice_settings(body)
+        if path == "/api/voice/ptt" and method == "POST":
+            from src.voice.service import push_to_talk_action
+
+            return 200, push_to_talk_action(str(body.get("action") or ""))
         return 404, {"error": f"unknown route: {method} {path}"}
 
     def serve_forever(self) -> None:

src/desktop/web/app.js

@@ -679,6 +679,7 @@ function applyRoute() {
   const showSetup = route === "setup" || (route === "chat" && Boolean(state.status?.needs_setup));
   $("setup-modal")?.classList.toggle("hidden", !showSetup);
   if (route === "library") refreshLibrary().catch(reportError);
+  if (route === "settings") loadVoiceSettings().catch(reportError);
 }
 
 async function refreshLibrary() {
@@ -775,7 +776,90 @@ function renderScheduleList(items) {
   }
 }
 
+function fillVoiceDevices(select, devices, kind, selected) {
+  select.innerHTML = "";
+  const fallback = document.createElement("option");
+  fallback.value = "";
+  fallback.textContent = kind === "input" ? "Default microphone" : "Default speaker";
+  select.appendChild(fallback);
+  for (const device of devices.filter((item) => item.kind === kind)) {
+    const option = document.createElement("option");
+    option.value = device.id;
+    const marks = [device.host_api];
+    if (device.bluetooth) marks.push("Bluetooth");
+    if (device.default) marks.push("default");
+    option.textContent = `${device.name} (${marks.join(", ")})`;
+    select.appendChild(option);
+  }
+  select.value = selected || "";
+}
+
+async function loadVoiceSettings() {
+  try {
+    const data = await api("/api/voice");
+    const settings = data.settings || {};
+    $("voice-stt").value = settings.stt_provider || "local";
+    $("voice-tts").value = settings.tts_provider || "local";
+    $("voice-ptt-enabled").checked = settings.push_to_talk !== false;
+    $("voice-wake").checked = Boolean(settings.wake_word_enabled);
+    $("voice-wake-phrase").value = settings.wake_word || "jonathan";
+    fillVoiceDevices($("voice-input"), data.devices || [], "input", settings.input_device);
+    fillVoiceDevices($("voice-output"), data.devices || [], "output", settings.output_device);
+    $("voice-note").textContent = settings.note || data.device_error || "";
+  } catch (err) {
+    $("voice-note").textContent = err.message;
+  }
+}
+
+async function saveVoiceSettings() {
+  const data = await api("/api/voice", {
+    method: "POST",
+    body: JSON.stringify({
+      stt_provider: $("voice-stt").value,
+      tts_provider: $("voice-tts").value,
+      input_device: $("voice-input").value,
+      output_device: $("voice-output").value,
+      push_to_talk: $("voice-ptt-enabled").checked,
+      wake_word_enabled: $("voice-wake").checked,
+      wake_word: $("voice-wake-phrase").value,
+    }),
+  });
+  $("voice-note").textContent = data.settings?.note || "";
+}
+
+function bindVoiceTalk() {
+  const button = $("voice-ptt");
+  if (!button) return;
+  const start = async (event) => {
+    event.preventDefault();
+    button.classList.add("holding");
+    try {
+      await api("/api/voice/ptt", { method: "POST", body: JSON.stringify({ action: "start" }) });
+    } catch (err) {
+      button.classList.remove("holding");
+      addBubble("system", err.message);
+    }
+  };
+  const stop = async () => {
+    if (!button.classList.contains("holding")) return;
+    button.classList.remove("holding");
+    try {
+      const data = await api("/api/voice/ptt", { method: "POST", body: JSON.stringify({ action: "stop" }) });
+      if (data.text) {
+        const box = $("prompt");
+        box.value = `${box.value ? `${box.value} ` : ""}${data.text}`;
+      }
+    } catch (err) {
+      addBubble("system", err.message);
+    }
+  };
+  button.addEventListener("pointerdown", start);
+  button.addEventListener("pointerup", stop);
+  button.addEventListener("pointerleave", stop);
+}
+
 function bindUi() {
+  bindVoiceTalk();
   $("send").addEventListener("click", () => sendMessage("/api/chat"));
   $("workers")?.addEventListener("click", () => sendMessage("/api/multi-agent"));
   $("nav-toggle")?.addEventListener("click", () => {
@@ -919,6 +1003,7 @@ function bindUi() {
         });
       }
       $("settings-key").value = "";
+      await saveVoiceSettings();
       go("chat");
       await refreshStatus();
     } catch (err) {

src/desktop/web/index.html

@@ -84,6 +84,7 @@ <h1 id="chat-title">New chat</h1>
             <button id="attach-files" class="ghost" title="Attach files">Attach</button>
             <button id="attach-clipboard" class="ghost" title="Attach current clipboard text">Clipboard</button>
             <button id="attach-screenshot" class="ghost hidden" title="Capture the current screen">Screenshot</button>
+            <button id="voice-ptt" class="ghost" type="button" title="Hold to talk through the selected microphone">Hold to talk</button>
             <input id="file-input" type="file" multiple hidden />
           </div>
           <button id="workers" class="ghost" title="Plan this into isolated internal workers">Workers</button>
@@ -165,6 +166,34 @@ <h2>Provider &amp; model</h2>
         </div>
         <div id="settings-local-results" class="scan-list"></div>
       </div>
+      <h3>Voice</h3>
+      <p class="muted">Speech stays on this machine unless you choose OpenAI or xAI. Install the extra with <code>pip install "clawd-codex[voice]"</code>. Push-to-talk uses the selected microphone and speaker, including Bluetooth on Windows WASAPI and Linux PulseAudio or PipeWire.</p>
+      <label>Speech to text
+        <select id="voice-stt">
+          <option value="local">Local Whisper</option>
+          <option value="openai">OpenAI</option>
+          <option value="xai">xAI</option>
+        </select>
+      </label>
+      <label>Text to speech
+        <select id="voice-tts">
+          <option value="local">Local Piper</option>
+          <option value="openai">OpenAI</option>
+          <option value="xai">xAI</option>
+        </select>
+      </label>
+      <label>Microphone
+        <select id="voice-input"></select>
+      </label>
+      <label>Speaker
+        <select id="voice-output"></select>
+      </label>
+      <label class="check"><input id="voice-ptt-enabled" type="checkbox" checked /> Push to talk</label>
+      <label class="check"><input id="voice-wake" type="checkbox" /> Wake word (optional, stays on this machine)</label>
+      <label>Wake phrase
+        <input id="voice-wake-phrase" type="text" value="jonathan" />
+      </label>
+      <p id="voice-note" class="muted"></p>
       <p class="muted">Source and updates: <code>https://github.com/GoDeskio/Clawd-Code</code>. Default local tree: <code>~/Jonathan/Jonathan-Ai</code>.</p>
       <div class="row end">
         <button id="settings-close" class="ghost">Close</button>

src/desktop/web/styles.css

@@ -224,6 +224,9 @@ button:disabled { opacity: 0.55; cursor: default; }
   padding: 8px 10px;
 }
 .composer-row { justify-content: space-between; margin-top: 6px; }
+#voice-ptt.holding { outline: 2px solid var(--accent); }
+label.check { display: flex; align-items: center; }
+.dialog input[type="checkbox"] { width: auto; min-height: 0; margin: 0 8px 0 0; }
 .end { justify-content: flex-end; }
 .chips { display: flex; flex-wrap: wrap; gap: 4px; margin-bottom: 6px; }
 .chip {