@hanzo/rooms 0.1.3 → 0.1.4

package/dist/Chat.js

@@ -35,14 +35,14 @@ import { freeCopy, grantConsent, hasConsent, isFree, } from "@hanzo/ai";
 import { ASK, Chat as ChatSurface, Code, Failure, Parts, words } from "@hanzo/ui/chat";
 import { compose, hold, images, partsOf } from "./lib/attach.js";
 import { Face, brief as roomBrief, join, roleOf, roomTurn, roster as rosterLine, rosterOf, speakers, speaksOf, voiceOf } from "./cast.js";
-import { SpeakButton, speakAgent } from "./speech.js";
+import { SpeakButton, cleanForSpeech } from "./speech.js";
 import { Crew, HOUSE, useCrew } from "./crew.js";
 import { called } from "./team.js";
 import { Anchor, Box, Button, Text, View, XStack, YStack } from "@hanzo/ui";
 import { Control } from "@hanzo/composer";
 import { useIam, useIamToken } from "@hanzo/iam/react";
 import { hasSession, org } from "./lib/session.js";
-import { clock, sku, useLimits } from "./lib/limits.js";
+import { chats, clock, sku, useLimits } from "./lib/limits.js";
 import { Meters } from "./meters.js";
 import { useHydrated } from "./lib/hydrated.js";
 import { base, served } from "./lib/ai.js";
@@ -431,7 +431,11 @@ function Thread({ placeholder = MESSAGE, initial, anonymous = false, heading, })
     // bare "not found", after it has already recorded the question. `served`
     // keeps the house name where it is served and otherwise falls to the first
     // model that is, so a room on any catalogue can be spoken to.
-    const { models } = useModels();
+    // A SPEECH MODEL NEVER ANSWERS A CHAT TURN. The catalog lists the
+    // transcriber and the voice beside the chat models; one that names its
+    // outputs and names no text (`chats`) is neither offered nor fallen back to.
+    const { models: catalog } = useModels();
+    const models = useMemo(() => catalog.filter(chats), [catalog]);
     // HANZO MODELS ALONE are offered here, to everyone: an Enso or Zen id
     // (`sku`). A preference kept from before for any other model is not one this
     // room offers, so the house router is asked for instead, and `served` keeps
@@ -615,42 +619,40 @@ function Thread({ placeholder = MESSAGE, initial, anonymous = false, heading, })
         setFileSaid(refused.length ? refused.join(" ") : null);
     }, []);
     useSeed(initial, setDraft);
-    // DICTATION, the same one the site's own composer offers. `speech()` defaults
-    // to the Hanzo gateway and `base()` is the one place a host is read from, so
+    // DICTATION through Hanzo's own ear: the mic records and each pause is
+    // posted to /v1/audio/transcriptions (`zen-scribe`). `speech()` defaults to
+    // the Hanzo gateway and `base()` is the one place a host is read from, so
     // pointing a local cloud at one points both. The token is the reader's own —
-    // the package never invents a credential — and where a browser has no
-    // recogniser `<Voice/>` wears the reason rather than failing silently.
+    // the package never invents a credential — and a refusal is worn by
+    // `<Voice/>`, never passed off as a transcript.
     const { token } = useIamToken();
-    const agentVoice = voiceOf(withWhom);
-    const ear = useMemo(() => speech({
-        ...base(),
-        ...(token ? { token } : {}),
-        voice: { name: agentVoice },
-    }), [token, agentVoice]);
+    const ear = useMemo(() => speech({ ...base(), ...(token ? { token } : {}) }), [token]);
     const voice = useVoice({
         speech: ear,
         onPartial: setDraft,
         onUtterance: setDraft,
     });
-    // When voice dictation is open, automatically speak incoming assistant replies
+    // WHILE THE MIC IS OPEN, A FINISHED REPLY IS READ ALOUD by the platform's
+    // voice, through the conversation's own mouth so speaking over it stops it.
+    // A room's reply is read part by part, each in its speaker's cast voice.
     const wasStreaming = useRef(false);
     useEffect(() => {
         if (wasStreaming.current && !chat.streaming && voice.open) {
             const msgs = chat.messages;
             const lastMsg = msgs[msgs.length - 1];
             if (lastMsg && lastMsg.role === "assistant" && typeof lastMsg.content === "string") {
-                // A room's reply is read part by part, each in its speaker's voice.
                 const parts = room.length > 1 ? speakers(lastMsg.content, room) : [{ who: withWhom, text: lastMsg.content }];
-                const read = (i) => {
-                    if (i >= parts.length)
-                        return;
-                    speakAgent(parts[i].text, { agent: parts[i].who ?? withWhom ?? undefined, onEnd: () => read(i + 1) });
-                };
-                read(0);
+                void (async () => {
+                    for (const part of parts) {
+                        const said = cleanForSpeech(part.text);
+                        if (said)
+                            await voice.say(said, voiceOf(part.who ?? withWhom ?? undefined));
+                    }
+                })();
             }
         }
         wasStreaming.current = chat.streaming;
-    }, [chat.streaming, voice.open, chat.messages, withWhom, room]);
+    }, [chat.streaming, voice.open, voice.say, chat.messages, withWhom, room]);
     // A SEND THAT WAITS ONE RENDER. `send` closes over the turns it was rendered
     // with, so a send that first tidies the thread is held here and made from the
     // next render, where the tidied turns are the ones it sends.

package/dist/Meet.js

@@ -1,27 +1,31 @@
 'use client';
 import { Fragment as _Fragment, jsx as _jsx, jsxs as _jsxs } from "react/jsx-runtime";
-import { useState, useEffect, useRef, useCallback } from 'react';
+import { useState, useEffect, useRef, useCallback, useMemo } from 'react';
 import { Video, VideoOff, Mic, MicOff, PhoneOff, Copy, Check, Plus, Calendar, ExternalLink, Bot, Volume2, Square, Send, UserPlus, Tv, LayoutGrid, ArrowRight, } from 'lucide-react';
 import { Text, View, XStack, YStack } from '@hanzo/gui';
 import { Button } from '@hanzo/ui';
+import { useIamToken } from '@hanzo/iam/react';
+import { REFUSED, speech, useVoice } from '@hanzo/voice';
 import { Face, voiceOf } from './cast.js';
 import { member } from './team.js';
-import { speakAgent, stopAgentSpeech, AudioWave, useAgentSpeech } from './speech.js';
+import { cleanForSpeech, speakAgent, stopAgentSpeech, AudioWave, useAgentSpeech } from './speech.js';
 import { GOOD } from './lib/mix.js';
+import { api } from './lib/api.js';
+import { ENSO, base } from './lib/ai.js';
 /** A core member as a meeting seats them: the team's name and part, the cast's voice. */
 const seated = (id) => {
     const one = member(id);
     return { id, name: one.name, role: one.role, voice: voiceOf(id) };
 };
 export const AVAILABLE_MEETING_AGENTS = [
-    { id: 'hanzo-coder', name: 'Hanzo Coder', role: 'Engineering Lead & Coder', voice: 'ash' },
-    { id: 'hanzo-researcher', name: 'Hanzo Researcher', role: 'Research & DeSci Scientist', voice: 'fable' },
+    { id: 'hanzo-coder', name: 'Hanzo Coder', role: 'Engineering Lead & Coder', voice: voiceOf('Hanzo Coder') },
+    { id: 'hanzo-researcher', name: 'Hanzo Researcher', role: 'Research & DeSci Scientist', voice: voiceOf('Hanzo Researcher') },
     seated('maya'),
     seated('des'),
     seated('vi'),
     seated('nora'),
     seated('einstein'),
-    { id: 'zach', name: 'Zach', role: 'Quantitative Analyst', voice: 'onyx' },
+    { id: 'zach', name: 'Zach', role: 'Quantitative Analyst', voice: voiceOf('zach') },
 ];
 const DEFAULT_MEETINGS = [
     {
@@ -64,65 +68,37 @@ const DEFAULT_MEETINGS = [
         url: 'https://meet.hanzo.ai/security-sync',
     },
 ];
-async function generateAgentMeetingResponse(prompt, agentName) {
-    const token = typeof window !== 'undefined' ? localStorage.getItem('hanzo_iam_access_token') : null;
-    if (token) {
-        try {
-            const res = await fetch('https://api.hanzo.ai/v1/chat/completions', {
-                method: 'POST',
-                headers: {
-                    'Content-Type': 'application/json',
-                    Authorization: `Bearer ${token}`,
+/**
+ * The co-pilot's answer to one thing said in the call, in a sentence or two a
+ * voice can read. Asked of the gateway as the reader (`token`, from
+ * `useIamToken`) through the one API host; a refusal is thrown with what the
+ * gateway said, never answered with a line nobody wrote.
+ */
+async function generateAgentMeetingResponse(prompt, agentName, token) {
+    if (!token)
+        throw new Error('Sign in to ask the co-pilot.');
+    const res = await fetch(`${api()}/v1/chat/completions`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json', Authorization: `Bearer ${token}` },
+        body: JSON.stringify({
+            model: ENSO,
+            messages: [
+                {
+                    role: 'system',
+                    content: `You are ${agentName}, participating in a live Hanzo Meet video call. Respond succinctly, conversationally, and informatively in 1 to 2 sentences suitable for voice synthesis.`,
                 },
-                body: JSON.stringify({
-                    model: 'gpt-4o-mini',
-                    messages: [
-                        {
-                            role: 'system',
-                            content: `You are ${agentName}, participating in a live Hanzo Meet video call. Respond succinctly, conversationally, and informatively in 1 to 2 sentences suitable for voice synthesis.`,
-                        },
-                        {
-                            role: 'user',
-                            content: prompt,
-                        },
-                    ],
-                    max_tokens: 150,
-                }),
-            });
-            if (res.ok) {
-                const data = await res.json();
-                const text = data.choices?.[0]?.message?.content?.trim();
-                if (text)
-                    return text;
-            }
-        }
-        catch { }
-    }
-    // Domain-specific intelligent persona responses
-    const lower = prompt.toLowerCase();
-    const agent = agentName.toLowerCase();
-    if (agent.includes('coder')) {
-        if (lower.includes('audio') || lower.includes('webrtc') || lower.includes('media')) {
-            return 'The WebRTC audio track is streaming over LiveKit SFU at 24kHz with PCM capture and real-time turn-taking.';
-        }
-        if (lower.includes('deploy') || lower.includes('build') || lower.includes('status')) {
-            return 'All services and container builds are verified green across Kubernetes pods with zero regressions.';
-        }
-        return `As Hanzo Coder, I am tracking that: our engineering pipeline is synchronized and ready for production deployment.`;
-    }
-    if (agent.includes('researcher')) {
-        return `From our DeSci research benchmarks, model convergence and validation loss show optimal scaling across the cluster.`;
-    }
-    if (agent.includes('maya')) {
-        return `I have recorded this decision and scheduled the follow-up milestone in our calendar.`;
-    }
-    if (agent.includes('des')) {
-        return `The UI hierarchy, contrast ratios, and layout flow are optimized for clarity across both desktop and mobile.`;
-    }
-    if (agent.includes('vi')) {
-        return `Cluster health, P99 response latencies, and auto-scaling ingress controllers are reporting 100% operational.`;
-    }
-    return `Understood. I have logged that item for our meeting sync and will monitor the execution.`;
+                { role: 'user', content: prompt },
+            ],
+            max_tokens: 150,
+        }),
+    });
+    if (!res.ok)
+        throw new Error(`The co-pilot is unavailable (${res.status}).`);
+    const data = await res.json();
+    const text = data.choices?.[0]?.message?.content?.trim();
+    if (!text)
+        throw new Error('The co-pilot answered with nothing.');
+    return text;
 }
 /** design's smallest rung (`--text-floor`, 10px), below gui's `$1`; gui's font size takes a rung or a number. */
 const FLOOR = 10;
@@ -143,16 +119,11 @@ export function Meet() {
     const [meetingView, setMeetingView] = useState('iframe');
     const [agentQuestion, setAgentQuestion] = useState('');
     // Agent Speech in Call
+    const { token } = useIamToken();
     const { speaking, speak, toggle } = useAgentSpeech(activeCoPilotAgent);
     const [audioTesting, setAudioTesting] = useState(false);
-    const [handsFree, setHandsFree] = useState(false);
-    const recognitionRef = useRef(null);
-    const [liveNotes, setLiveNotes] = useState([
-        '18:02: Meeting started with participants. AI Bot transcription active.',
-        '18:04: Discussed latency metrics on local S3 drive server (port 9005). Average response: 1.2ms.',
-        '18:08: Agreed on channel connector architecture: channels connect via bot instances exclusively.',
-        '18:12: Action Item assigned to Hanzo Coder: Integrate Cal.com scheduling engine across bot triggers.',
-    ]);
+    // The call's notes: what was said and answered here, and nothing before it.
+    const [liveNotes, setLiveNotes] = useState([]);
     const videoRef = useRef(null);
     const localStreamRef = useRef(null);
     useEffect(() => {
@@ -230,80 +201,37 @@ export function Meet() {
             return;
         }
         setAudioTesting(true);
-        speakAgent(`Welcome to Hanzo Meet. Speaker test complete. All agent audio and video systems are ready for this call.`, {
+        speakAgent(`Welcome to Hanzo Meet. This is ${activeCoPilotAgent}'s voice. If you can hear this, your speakers are working.`, {
             agent: activeCoPilotAgent,
+            token,
             onEnd: () => setAudioTesting(false),
             onError: () => setAudioTesting(false),
         });
     };
+    // HANDS-FREE is one spoken conversation with the co-pilot, on @hanzo/voice:
+    // the mic records, each pause goes to Hanzo's /v1/audio/transcriptions, the
+    // utterance is answered, and the answer is read in the co-pilot's cast voice
+    // by /v1/audio/speech. Speaking over the reply stops it (the machine's
+    // barge-in). Every browser that can record takes part, not only the ones with
+    // a working SpeechRecognition — Chromium on Linux and Firefox have none.
+    const ear = useMemo(() => speech({ ...base(), ...(token ? { token } : {}) }), [token]);
+    const handsFree = useVoice({
+        speech: ear,
+        onUtterance: (transcript) => void handleSpokenInput(transcript),
+    });
+    const handsFreeSay = handsFree.say;
     const handleSpokenInput = useCallback(async (transcript) => {
         const time = new Date().toLocaleTimeString().slice(0, 5);
-        const userNote = `${time}: You (voice): ${transcript}`;
-        setLiveNotes((prev) => [...prev, userNote]);
-        const reply = await generateAgentMeetingResponse(transcript, activeCoPilotAgent);
-        const agentNote = `${time}: ${activeCoPilotAgent} (AI): ${reply}`;
-        setLiveNotes((prev) => [...prev, agentNote]);
-        speak(reply, activeCoPilotAgent);
-    }, [activeCoPilotAgent, speak]);
-    // Real-time hands-free speech recognition & barge-in turn-taking
-    useEffect(() => {
-        if (!handsFree || !activeMeetingRoom) {
-            if (recognitionRef.current) {
-                try {
-                    recognitionRef.current.stop();
-                }
-                catch { }
-                recognitionRef.current = null;
-            }
-            return;
-        }
-        if (typeof window === 'undefined')
-            return;
-        const SpeechRec = window.SpeechRecognition || window.webkitSpeechRecognition;
-        if (!SpeechRec)
-            return;
-        const recognition = new SpeechRec();
-        recognition.continuous = true;
-        recognition.interimResults = true;
-        recognition.lang = 'en-US';
-        recognition.onresult = (event) => {
-            // Barge-in: if human starts speaking while agent is talking, halt speech immediately!
-            if (speaking) {
-                stopAgentSpeech();
-            }
-            for (let i = event.resultIndex; i < event.results.length; i++) {
-                if (event.results[i].isFinal) {
-                    const transcript = event.results[i][0]?.transcript?.trim();
-                    if (transcript) {
-                        handleSpokenInput(transcript);
-                    }
-                }
-            }
-        };
-        recognition.onerror = () => { };
-        recognition.onend = () => {
-            if (handsFree && activeMeetingRoom) {
-                try {
-                    recognition.start();
-                }
-                catch { }
-            }
-        };
+        setLiveNotes((prev) => [...prev, `${time}: You (voice): ${transcript}`]);
         try {
-            recognition.start();
-            recognitionRef.current = recognition;
+            const reply = await generateAgentMeetingResponse(transcript, activeCoPilotAgent, token);
+            setLiveNotes((prev) => [...prev, `${time}: ${activeCoPilotAgent} (AI): ${reply}`]);
+            void handsFreeSay(cleanForSpeech(reply), voiceOf(activeCoPilotAgent));
         }
-        catch { }
-        return () => {
-            if (recognitionRef.current) {
-                try {
-                    recognitionRef.current.stop();
-                }
-                catch { }
-                recognitionRef.current = null;
-            }
-        };
-    }, [handsFree, activeMeetingRoom, speaking, handleSpokenInput]);
+        catch (err) {
+            setLiveNotes((prev) => [...prev, `${time}: ${activeCoPilotAgent} did not answer: ${err.message}`]);
+        }
+    }, [activeCoPilotAgent, token, handsFreeSay]);
     const handleAskAgent = async (e) => {
         e.preventDefault();
         const q = agentQuestion.trim();
@@ -311,26 +239,37 @@ export function Meet() {
             return;
         setAgentQuestion('');
         const time = new Date().toLocaleTimeString().slice(0, 5);
-        const userNote = `${time}: You: ${q}`;
-        setLiveNotes((prev) => [...prev, userNote]);
-        const reply = await generateAgentMeetingResponse(q, activeCoPilotAgent);
-        const agentNote = `${time}: ${activeCoPilotAgent} (AI): ${reply}`;
-        setLiveNotes((prev) => [...prev, agentNote]);
-        speak(reply, activeCoPilotAgent);
+        setLiveNotes((prev) => [...prev, `${time}: You: ${q}`]);
+        try {
+            const reply = await generateAgentMeetingResponse(q, activeCoPilotAgent, token);
+            setLiveNotes((prev) => [...prev, `${time}: ${activeCoPilotAgent} (AI): ${reply}`]);
+            speak(reply, activeCoPilotAgent);
+        }
+        catch (err) {
+            setLiveNotes((prev) => [...prev, `${time}: ${activeCoPilotAgent} did not answer: ${err.message}`]);
+        }
     };
+    // The briefing is the call's own notes, read aloud; with none there is nothing to brief.
     const handleReadBriefing = () => {
-        const summary = `Executive briefing for ${activeMeetingRoom || 'call'}. Key items discussed: latency metrics on local S3 drive server average 1.2ms, channel connector architecture verified, Cal scheduling integration underway.`;
-        toggle(summary, activeCoPilotAgent);
+        if (!liveNotes.length)
+            return;
+        toggle(`Briefing for ${activeMeetingRoom || 'this call'}. ${liveNotes.join('. ')}`, activeCoPilotAgent);
     };
+    // What the hands-free control says: why it cannot run, or that Hanzo's speech
+    // refused and the browser stood in (or could not) — never silence.
+    const handsFreeNote = handsFree.reason ?? (handsFree.refusal ? REFUSED[handsFree.refusal.covered ? 'covered' : 'lost'] : null);
     return (_jsxs(YStack, { width: "100%", height: "100%", bg: "$background", overflowX: "auto", overflowY: "auto", children: [_jsxs(XStack, { minH: 56, px: 20, py: 8, gap: 10, flexWrap: "wrap", bg: "$panel", borderBottomWidth: 1, borderBottomColor: "$borderColor", items: "center", justify: "space-between", children: [_jsxs(XStack, { items: "center", gap: 12, shrink: 1, minW: "auto", children: [_jsx(XStack, { width: 32, height: 32, rounded: 8, bg: "$raised", items: "center", justify: "center", shrink: 1, minW: "auto", children: _jsx(Video, { size: 18, color: "var(--foreground)" }) }), _jsxs(YStack, { shrink: 1, minW: "auto", children: [_jsxs(XStack, { items: "center", gap: 8, children: [_jsx(Text, { fontSize: "$4", fontWeight: "700", color: "$ink", children: "Hanzo Meet" }), _jsx(Text, { fontSize: FLOOR, fontWeight: "700", bg: "$hover", color: "$soft", borderWidth: 1, borderColor: "$borderColor", px: 6, py: 1, rounded: 4, children: "LIVE \u00B7 meet.hanzo.ai" })] }), _jsx(Text, { fontSize: "$1", color: "$soft", children: "Native WebRTC video conferencing with AI meeting bot transcription & voice" })] })] }), activeMeetingRoom ? (_jsxs(XStack, { items: "center", gap: 8, shrink: 1, minW: "auto", flexWrap: "wrap", children: [_jsx(XStack, { bg: "$hover", rounded: 8, p: 2, gap: 2, borderWidth: 1, borderColor: "$borderColor", shrink: 1, minW: "auto", children: [
                                     ['iframe', 'SFU Stage', Tv],
                                     ['grid', 'Agent Grid', LayoutGrid],
                                 ].map(([view, label, Icon]) => (_jsxs(Button, { size: "sm", variant: meetingView === view ? 'primary' : 'ghost', ...(meetingView === view ? null : { color: '$soft' }), "aria-pressed": meetingView === view, onClick: () => setMeetingView(view), children: [_jsx(Icon, { size: 12 }), label] }, view))) }), botAssisting ? (_jsx(View, { display: "none", "$max-md": { display: 'flex' }, children: _jsxs(Button, { size: "sm", variant: copilot ? 'secondary' : 'default', onClick: () => setCopilot((v) => !v), "aria-label": "Co-pilot", "aria-expanded": copilot, "aria-controls": "meet-copilot", children: [_jsx(Bot, { size: 13 }), "Co-pilot"] }) })) : null, _jsxs(Button, { size: "sm", onClick: () => handleCopyLink(activeMeetingRoom), children: [copied ? _jsx(Check, { size: 13 }) : _jsx(Copy, { size: 13 }), copied ? 'Link Copied' : 'Copy Room Link'] }), _jsxs(Button, { size: "sm", variant: "destructive", onClick: () => {
                                     stopAgentSpeech();
+                                    // Leaving the call hangs up the hands-free conversation with it.
+                                    if (handsFree.open)
+                                        handsFree.toggle();
                                     setActiveMeetingRoom(null);
                                 }, children: [_jsx(PhoneOff, { size: 14 }), "Leave Call"] })] })) : (_jsx(XStack, { items: "center", gap: 8, shrink: 1, minW: "auto", children: _jsx(Button, { asChild: true, variant: "linkMuted", size: "sm", px: 0, children: _jsxs("a", { href: "https://meet.hanzo.ai", target: "_blank", rel: "noopener noreferrer", children: ["meet.hanzo.ai", _jsx(ExternalLink, { size: 12 })] }) }) }))] }), activeMeetingRoom ? (_jsxs(XStack, { flex: 1, overflow: "hidden", position: "relative", children: [_jsx(XStack, { flex: 1, minW: "auto", position: "relative", bg: "$sunken", children: meetingView === 'iframe' ? (_jsx(View, { render: _jsx("iframe", { src: `https://meet.hanzo.ai/${activeMeetingRoom}#config.prejoinPageEnabled=false&config.startWithAudioMuted=${!micOn}&config.startWithVideoMuted=${!cameraOn}`, allow: "camera; microphone; display-capture; autoplay; clipboard-write; speaker-selection; screen-wake-lock", title: `Hanzo Meet Room: ${activeMeetingRoom}` }), width: "100%", height: "100%", borderWidth: 0, bg: "$sunken" }, "stage")) : (_jsxs(View, { flex: 1, display: "grid", gridTemplateColumns: "repeat(auto-fit, minmax(280px, 1fr))", gap: 12, p: 16, overflowY: "auto", bg: "$background", children: [_jsx(YStack, { bg: "$panel", rounded: 12, borderWidth: 1, borderColor: "$borderColor", position: "relative", overflow: "hidden", items: "center", justify: "center", minH: 220, children: _jsxs(YStack, { items: "center", children: [_jsx(YStack, { width: 72, height: 72, rounded: 9999, bg: "$raised", items: "center", justify: "center", mb: 8, children: _jsx(Text, { fontSize: "$8", fontWeight: "700", color: "$ink", children: "You" }) }), _jsx(Text, { fontSize: "$2", fontWeight: "600", color: "$ink", children: "Alex (Host)" }), _jsx(Text, { fontSize: "$1", color: micOn ? GOOD : '$soft', children: micOn ? 'Mic Active' : 'Muted' })] }) }), invitedAgents.map((agentName) => {
                                     const voice = voiceOf(agentName);
-                                    const isThisSpeaking = speaking && activeCoPilotAgent.toLowerCase() === agentName.toLowerCase();
+                                    const isThisSpeaking = (speaking || handsFree.state === 'speaking') && activeCoPilotAgent.toLowerCase() === agentName.toLowerCase();
                                     return (_jsxs(YStack, { bg: "$panel", rounded: 12, borderWidth: isThisSpeaking ? 2 : 1, borderColor: isThisSpeaking ? 'var(--state-success)' : '$borderColor', position: "relative", overflow: "hidden", items: "center", justify: "center", p: 20, minH: 220, transition: "200ms", children: [_jsxs(View, { position: "relative", mb: 10, children: [_jsx(Face, { name: agentName, size: 72 }), isThisSpeaking ? (_jsx(View, { position: "absolute", b: -4, r: -4, bg: "var(--popover)", borderWidth: 1, borderColor: "var(--state-success)", rounded: 9999, p: 4, children: _jsx(Volume2, { size: 12, color: GOOD }) })) : null] }), _jsx(Text, { fontSize: "$3", fontWeight: "700", color: "$ink", mb: 2, children: agentName }), _jsxs(Text, { fontSize: "$1", color: "$soft", mb: 8, children: ["Voice: ", voice] }), _jsx(XStack, { items: "center", gap: 6, bg: "$hover", px: 10, py: 4, rounded: 12, children: isThisSpeaking ? (_jsxs(_Fragment, { children: [_jsx(AudioWave, { active: true, color: "var(--state-success)" }), _jsx(Text, { fontSize: "$1", fontWeight: "600", color: GOOD, children: "Speaking in Call" })] })) : (_jsxs(_Fragment, { children: [_jsx(Bot, { size: 11, color: "var(--muted-foreground)" }), _jsx(Text, { fontSize: "$1", fontWeight: "600", color: "$quiet", children: "AI Agent \u00B7 In Call" })] })) })] }, agentName));
                                 })] }, "grid")) }), botAssisting ? (_jsxs(YStack, { id: "meet-copilot", width: 360, minW: 360, bg: "$panel", borderLeftWidth: 1, borderLeftColor: "$borderColor", p: 16, gap: 12, "$max-md": {
                             display: copilot ? 'flex' : 'none',
@@ -343,9 +282,13 @@ export function Meet() {
                             minW: 0,
                             z: 'var(--z-sheet)',
                             borderLeftWidth: 0,
-                        }, children: [_jsxs(XStack, { items: "center", justify: "space-between", children: [_jsxs(XStack, { items: "center", gap: 8, children: [_jsx(Face, { name: activeCoPilotAgent, size: 28 }), _jsxs(YStack, { children: [_jsx(Text, { fontSize: "$2", fontWeight: "700", color: "$ink", children: activeCoPilotAgent }), _jsxs(Text, { fontSize: FLOOR, color: "$faint", children: ["AI Co-pilot \u00B7 Voice: ", voiceOf(activeCoPilotAgent)] })] })] }), _jsx(Text, { fontSize: FLOOR, fontWeight: "700", color: GOOD, bg: "$hover", px: 6, py: 2, rounded: 4, children: "Live in call" })] }), _jsx(XStack, { gap: 4, flexWrap: "wrap", children: invitedAgents.map((agent) => (_jsx(Button, { size: "sm", variant: activeCoPilotAgent === agent ? 'secondary' : 'ghost', color: activeCoPilotAgent === agent ? '$ink' : '$soft', "aria-pressed": activeCoPilotAgent === agent, onClick: () => setActiveCoPilotAgent(agent), children: agent }, agent))) }), _jsxs(XStack, { gap: 8, children: [_jsx(Button, { size: "sm", flex: 1, variant: speaking ? 'secondary' : 'default', onClick: handleReadBriefing, children: speaking ? (_jsxs(_Fragment, { children: [_jsx(AudioWave, { active: true, color: "var(--state-success)" }), "Stop Briefing", _jsx(Square, { size: 11, fill: "currentColor" })] })) : (_jsxs(_Fragment, { children: [_jsx(Volume2, { size: 14 }), "Briefing"] })) }), _jsxs(Button, { size: "sm", variant: handsFree ? 'secondary' : 'default', "data-testid": "hands-free-voice-toggle", "aria-pressed": handsFree, title: handsFree ? 'Hands-Free Voice Active (Speaks and interrupts on voice)' : 'Enable Hands-Free Voice Mode', onClick: () => setHandsFree(!handsFree), children: [_jsx(Mic, { size: 14, color: handsFree ? 'var(--state-success)' : 'currentColor' }), handsFree ? 'Hands-Free On' : 'Hands-Free'] })] }), _jsx(YStack, { flex: 1, overflowY: "auto", bg: "$background", rounded: 8, borderWidth: 1, borderColor: "$borderColor", p: 12, gap: 8, children: liveNotes.map((note, i) => (_jsx(Text, { fontSize: "$1", color: "$quiet", lineHeight: 16.5, pb: 8, borderBottomWidth: 1, borderBottomColor: "$borderColor", children: note }, i))) }), _jsxs(XStack, { render: "form", onSubmit: handleAskAgent, gap: 6, children: [_jsx(Text, { render: _jsx("input", { type: "text", value: agentQuestion, onChange: (e) => setAgentQuestion(e.target.value), placeholder: `Ask ${activeCoPilotAgent} in meeting...` }), flex: 1, bg: "$background", borderWidth: 1, borderColor: "var(--border-control)", rounded: 6, px: 10, py: 8, color: "$ink", fontSize: "$1", outlineStyle: "none" }), _jsx(Button, { type: "submit", variant: "primary", size: "icon-sm", disabled: !agentQuestion.trim(), "aria-label": "Ask", children: _jsx(Send, { size: 12 }) })] }), _jsxs(XStack, { gap: 8, children: [_jsx(Button, { size: "sm", flex: 1, onClick: () => {
-                                            const sample = `${new Date().toLocaleTimeString().slice(0, 5)}: Participant discussed deployment timeline for next release.`;
-                                            setLiveNotes((prev) => [...prev, sample]);
+                        }, children: [_jsxs(XStack, { items: "center", justify: "space-between", children: [_jsxs(XStack, { items: "center", gap: 8, children: [_jsx(Face, { name: activeCoPilotAgent, size: 28 }), _jsxs(YStack, { children: [_jsx(Text, { fontSize: "$2", fontWeight: "700", color: "$ink", children: activeCoPilotAgent }), _jsxs(Text, { fontSize: FLOOR, color: "$faint", children: ["AI Co-pilot \u00B7 Voice: ", voiceOf(activeCoPilotAgent)] })] })] }), _jsx(Text, { fontSize: FLOOR, fontWeight: "700", color: GOOD, bg: "$hover", px: 6, py: 2, rounded: 4, children: "Live in call" })] }), _jsx(XStack, { gap: 4, flexWrap: "wrap", children: invitedAgents.map((agent) => (_jsx(Button, { size: "sm", variant: activeCoPilotAgent === agent ? 'secondary' : 'ghost', color: activeCoPilotAgent === agent ? '$ink' : '$soft', "aria-pressed": activeCoPilotAgent === agent, onClick: () => setActiveCoPilotAgent(agent), children: agent }, agent))) }), _jsxs(XStack, { gap: 8, children: [_jsx(Button, { size: "sm", flex: 1, variant: speaking ? 'secondary' : 'default', disabled: !speaking && !liveNotes.length, onClick: handleReadBriefing, children: speaking ? (_jsxs(_Fragment, { children: [_jsx(AudioWave, { active: true, color: "var(--state-success)" }), "Stop Briefing", _jsx(Square, { size: 11, fill: "currentColor" })] })) : (_jsxs(_Fragment, { children: [_jsx(Volume2, { size: 14 }), "Briefing"] })) }), _jsxs(Button, { size: "sm", variant: handsFree.open ? 'secondary' : 'default', "data-testid": "hands-free-voice-toggle", "data-state": handsFree.state, "data-refusal": handsFree.refusal ? handsFree.refusal.service : undefined, "aria-pressed": handsFree.open, disabled: !!handsFree.blocked, title: handsFreeNote ??
+                                            (handsFree.open ? 'Hands-Free Voice Active (Speaks and interrupts on voice)' : 'Enable Hands-Free Voice Mode'), onClick: handsFree.toggle, children: [_jsx(Mic, { size: 14, color: handsFree.open ? 'var(--state-success)' : 'currentColor' }), handsFree.open ? 'Hands-Free On' : 'Hands-Free'] })] }), handsFreeNote ? (_jsx(Text, { fontSize: "$1", color: "$soft", role: "status", children: handsFreeNote })) : null, _jsx(YStack, { flex: 1, overflowY: "auto", bg: "$background", rounded: 8, borderWidth: 1, borderColor: "$borderColor", p: 12, gap: 8, children: liveNotes.map((note, i) => (_jsx(Text, { fontSize: "$1", color: "$quiet", lineHeight: 16.5, pb: 8, borderBottomWidth: 1, borderBottomColor: "$borderColor", children: note }, i))) }), _jsxs(XStack, { render: "form", onSubmit: handleAskAgent, gap: 6, children: [_jsx(Text, { render: _jsx("input", { type: "text", value: agentQuestion, onChange: (e) => setAgentQuestion(e.target.value), placeholder: `Ask ${activeCoPilotAgent} in meeting...` }), flex: 1, bg: "$background", borderWidth: 1, borderColor: "var(--border-control)", rounded: 6, px: 10, py: 8, color: "$ink", fontSize: "$1", outlineStyle: "none" }), _jsx(Button, { type: "submit", variant: "primary", size: "icon-sm", disabled: !agentQuestion.trim(), "aria-label": "Ask", children: _jsx(Send, { size: 12 }) })] }), _jsxs(XStack, { gap: 8, children: [_jsx(Button, { size: "sm", flex: 1, disabled: !agentQuestion.trim(), title: "Add what is typed above to the notes, without asking the co-pilot", onClick: () => {
+                                            const note = agentQuestion.trim();
+                                            if (!note)
+                                                return;
+                                            setAgentQuestion('');
+                                            setLiveNotes((prev) => [...prev, `${new Date().toLocaleTimeString().slice(0, 5)}: You (note): ${note}`]);
                                         }, children: "+ Add Live Note" }), _jsx(Button, { variant: "primary", size: "sm", onClick: () => {
                                             const text = liveNotes.join('\n');
                                             navigator.clipboard.writeText(text);

package/dist/cast.d.ts

@@ -24,18 +24,17 @@ interface Role {
      */
     ground?: string;
     /**
-     * The voice the platform reads this character's replies in, by
-     * /v1/audio/speech name.
+     * The voice the platform reads this character's replies in: a speech-service
+     * voice id, sent to /v1/audio/speech as it is. The first letter is the accent
+     * (a American, b British), the second the register (f female, m male).
      *
-     * ONE VOICE PER CHARACTER, and never two who would be in the same room. The
-     * service serves eleven and there are fourteen of them, so three must double
-     * up — the question is which three, and the answer is people who work apart:
-     * the five the demo puts in one call are all distinct. `echo` was reading for
-     * three at once, which in a call is one person doing three parts.
+     * ONE VOICE PER CHARACTER, and never two who would be in the same room. A
+     * house part shares the voice of the character whose face it wears, so Hanzo
+     * Coder sounds like Dev; no two characters share one.
      *
      * CAST, NOT COPIED. These are the speech service's own voices, chosen for fit
-     * — a warm animated reader for the physicist, something low and dry for the
-     * founder. None of them is anybody's actual voice and none is trying to be:
+     * — a playful reader for the physicist, something low and measured for the
+     * quant. None of them is anybody's actual voice and none is trying to be:
      * synthesising a real person saying words they never said is a different act
      * from casting a part, and this is the second one.
      */
@@ -88,18 +87,22 @@ export declare function speakers(text: string, names: string[]): {
     who: string | null;
     text: string;
 }[];
+/**
+ * Every American and British voice the speech service reads in, which is the
+ * pool an agent nobody cast is given a voice from. `am_santa` is left out: it
+ * is a character voice, not a reader.
+ */
 export declare const VOICES: string[];
+/** A voice for a name nobody cast: the same name always gets the same one. */
 export declare function fallbackVoice(name: string | undefined): string;
 /** The voice to read a character's replies in. Always returns a valid voice. */
 export declare const voiceOf: (name: string | undefined) => string;
+/** Who a voice sounds like, read off its id — what a browser voice standing in is matched on. */
 export interface VoiceProfile {
     voice: string;
-    pitch: number;
-    rate: number;
-    description: string;
-    gender: 'male' | 'female' | 'neutral';
+    gender: 'male' | 'female';
+    accent: 'american' | 'british';
 }
-export declare const VOICE_PROFILES: Record<string, VoiceProfile>;
 export declare function voiceProfileOf(name: string | undefined): VoiceProfile;
 export declare function fallbackMemoji(name: string | undefined): string;
 /**

package/dist/cast.js

@@ -24,30 +24,30 @@ export const CAST = {
     //
     // Dev wears its own memoji (black ground baked in, so it fills the disc and
     // needs no rim colour); hanzo coder is that same face under the house name.
-    dev: { portrait: '/agents/dev.png', voice: 'ash' },
-    des: { portrait: '/agents/des.png', ground: '#74569F', voice: 'echo' },
-    vi: { portrait: '/agents/vi.png', ground: '#41649F', voice: 'sage' },
-    antje: { portrait: '/agents/antje.jpg', voice: 'nova' },
-    zach: { portrait: '/agents/zach.jpg', voice: 'onyx' },
-    feynman: { portrait: '/agents/feynman.png', ground: '#9C7C4F', voice: 'fable' },
-    einstein: { portrait: '/agents/einstein.png', ground: '#7D6450', voice: 'verse' },
-    nora: { portrait: '/agents/nora.png', voice: 'shimmer' },
-    creative: { portrait: '/agents/creative.png', voice: 'ballad' },
-    maya: { portrait: '/agents/maya.png', voice: 'coral' },
-    leo: { portrait: '/agents/leo.png', voice: 'alloy' },
+    dev: { portrait: '/agents/dev.png', voice: 'am_michael' },
+    des: { portrait: '/agents/des.png', ground: '#74569F', voice: 'af_bella' },
+    vi: { portrait: '/agents/vi.png', ground: '#41649F', voice: 'am_fenrir' },
+    antje: { portrait: '/agents/antje.jpg', voice: 'af_nova' },
+    zach: { portrait: '/agents/zach.jpg', voice: 'am_onyx' },
+    feynman: { portrait: '/agents/feynman.png', ground: '#9C7C4F', voice: 'am_puck' },
+    einstein: { portrait: '/agents/einstein.png', ground: '#7D6450', voice: 'bm_george' },
+    nora: { portrait: '/agents/nora.png', voice: 'af_heart' },
+    creative: { portrait: '/agents/creative.png', voice: 'bm_fable' },
+    maya: { portrait: '/agents/maya.png', voice: 'bf_emma' },
+    leo: { portrait: '/agents/leo.png', voice: 'am_liam' },
     // THE HOUSE AGENTS, UNDER THE NAMES A READER SEES. Every entry above is keyed
     // by the character; these are keyed by the part.
-    'hanzo coder': { portrait: '/agents/dev.png', voice: 'ash' },
-    'hanzo researcher': { portrait: '/agents/feynman.png', voice: 'fable' },
-    'hanzo designer': { portrait: '/agents/des.png', voice: 'echo' },
-    'hanzo devops': { voice: 'sage' },
-    'hanzo quant': { voice: 'onyx' },
-    'hanzo copywriter': { portrait: '/agents/creative.png', voice: 'ballad' },
-    'hanzo legal & compliance': { portrait: '/agents/einstein.png', voice: 'verse' },
-    'hanzo executive assistant': { portrait: '/agents/maya.png', voice: 'coral' },
-    'hanzo architect': { voice: 'verse' },
-    'hanzo vision': { voice: 'ballad' },
-    'hanzo support': { portrait: '/agents/nora.png', voice: 'shimmer' },
+    'hanzo coder': { portrait: '/agents/dev.png', voice: 'am_michael' },
+    'hanzo researcher': { portrait: '/agents/feynman.png', voice: 'am_puck' },
+    'hanzo designer': { portrait: '/agents/des.png', voice: 'af_bella' },
+    'hanzo devops': { voice: 'am_fenrir' },
+    'hanzo quant': { voice: 'am_onyx' },
+    'hanzo copywriter': { portrait: '/agents/creative.png', voice: 'bm_fable' },
+    'hanzo legal & compliance': { portrait: '/agents/einstein.png', voice: 'bm_george' },
+    'hanzo executive assistant': { portrait: '/agents/maya.png', voice: 'bf_emma' },
+    'hanzo architect': { voice: 'bm_george' },
+    'hanzo vision': { voice: 'bm_fable' },
+    'hanzo support': { portrait: '/agents/nora.png', voice: 'af_heart' },
 };
 /** The role for an agent name, or undefined for one nobody cast. */
 export const roleOf = (name) => CAST[name?.toLowerCase() ?? ''];
@@ -132,22 +132,21 @@ export function speakers(text, names) {
     }
     return out.map((one) => ({ ...one, text: one.text.trim() })).filter((one) => one.text || one.who);
 }
+/**
+ * Every American and British voice the speech service reads in, which is the
+ * pool an agent nobody cast is given a voice from. `am_santa` is left out: it
+ * is a character voice, not a reader.
+ */
 export const VOICES = [
-    'ash',
-    'echo',
-    'sage',
-    'nova',
-    'onyx',
-    'fable',
-    'shimmer',
-    'alloy',
-    'verse',
-    'ballad',
-    'coral',
+    'af_alloy', 'af_aoede', 'af_bella', 'af_heart', 'af_jessica', 'af_kore', 'af_nicole', 'af_nova', 'af_river', 'af_sarah', 'af_sky',
+    'am_adam', 'am_echo', 'am_eric', 'am_fenrir', 'am_liam', 'am_michael', 'am_onyx', 'am_puck',
+    'bf_alice', 'bf_emma', 'bf_isabella', 'bf_lily',
+    'bm_daniel', 'bm_fable', 'bm_george', 'bm_lewis',
 ];
+/** A voice for a name nobody cast: the same name always gets the same one. */
 export function fallbackVoice(name) {
     if (!name)
-        return VOICES[0];
+        return 'af_heart';
     let hash = 0;
     for (let i = 0; i < name.length; i++) {
         hash = (hash << 5) - hash + name.charCodeAt(i);
@@ -157,22 +156,13 @@ export function fallbackVoice(name) {
 }
 /** The voice to read a character's replies in. Always returns a valid voice. */
 export const voiceOf = (name) => roleOf(name)?.voice || fallbackVoice(name);
-export const VOICE_PROFILES = {
-    ash: { voice: 'ash', pitch: 1.0, rate: 1.0, description: 'Clear, direct, technical', gender: 'male' },
-    echo: { voice: 'echo', pitch: 1.1, rate: 1.05, description: 'Adaptive, creative, resonant', gender: 'female' },
-    sage: { voice: 'sage', pitch: 0.95, rate: 1.05, description: 'Authoritative, calm, systems-minded', gender: 'male' },
-    nova: { voice: 'nova', pitch: 1.1, rate: 1.0, description: 'Energetic, articulate, engaging', gender: 'female' },
-    onyx: { voice: 'onyx', pitch: 0.85, rate: 1.0, description: 'Deep, measured, quantitative', gender: 'male' },
-    fable: { voice: 'fable', pitch: 1.05, rate: 1.12, description: 'Warm, enthusiastic, pedagogical', gender: 'male' },
-    shimmer: { voice: 'shimmer', pitch: 1.1, rate: 0.98, description: 'Precise, professional, clear', gender: 'female' },
-    alloy: { voice: 'alloy', pitch: 0.95, rate: 1.05, description: 'Visionary, forward-leaning, focused', gender: 'male' },
-    verse: { voice: 'verse', pitch: 0.9, rate: 0.95, description: 'Intentional, iconic, design-first', gender: 'male' },
-    ballad: { voice: 'ballad', pitch: 0.92, rate: 1.0, description: 'Thoughtful, deep, expansive', gender: 'male' },
-    coral: { voice: 'coral', pitch: 1.15, rate: 1.05, description: 'Friendly, supportive, organized', gender: 'female' },
-};
 export function voiceProfileOf(name) {
-    const v = voiceOf(name);
-    return VOICE_PROFILES[v] || VOICE_PROFILES.ash;
+    const voice = voiceOf(name);
+    return {
+        voice,
+        gender: voice[1] === 'm' ? 'male' : 'female',
+        accent: voice[0] === 'b' ? 'british' : 'american',
+    };
 }
 /** Initials, for anyone with no likeness. Two at most: a disc this size holds
  *  two letters and turns three into a smudge. */

package/dist/lib/limits.d.ts

@@ -47,6 +47,15 @@ export declare const spendShown: ({ limits, answered }: {
 export declare const left: (s: Span) => number;
 /** Whether a model is a Hanzo SKU — an Enso or Zen id — the only models a consumer surface offers. */
 export declare const sku: (id: string) => boolean;
+/**
+ * Whether a model can answer a chat turn: it names no outputs, or names text
+ * among them. `zen-scribe` (a transcript) and `zen-voice-mini` (audio) are
+ * Zen ids the catalog lists, and neither answers a question.
+ */
+export declare const chats: (model: {
+    id: string;
+    outputs?: unknown;
+}) => boolean;
 /** A reset as a clock reads it in this browser — "17:00" — or '' for none. */
 export declare function clock(iso: string | null | undefined): string;
 /** A period's end as a sentence states it — "Oct 30, 2026" — or '' for none. Periods end on UTC days. */

package/dist/lib/limits.js

@@ -102,6 +102,12 @@ export const spendShown = ({ limits, answered }) => answered && !limits?.plan;
 export const left = (s) => 100 - s.percent;
 /** Whether a model is a Hanzo SKU — an Enso or Zen id — the only models a consumer surface offers. */
 export const sku = (id) => /^(hanzo\/)?(enso|zen)/i.test(id);
+/**
+ * Whether a model can answer a chat turn: it names no outputs, or names text
+ * among them. `zen-scribe` (a transcript) and `zen-voice-mini` (audio) are
+ * Zen ids the catalog lists, and neither answers a question.
+ */
+export const chats = (model) => !Array.isArray(model.outputs) || model.outputs.includes('text');
 /** A reset as a clock reads it in this browser — "17:00" — or '' for none. */
 export function clock(iso) {
     const t = iso ? Date.parse(iso) : NaN;

package/dist/speech.d.ts

@@ -1,16 +1,37 @@
 import React from 'react';
+import { type Refusal } from '@hanzo/voice';
+import { type VoiceProfile } from './cast.js';
 /** Strips code blocks, links, headers, and markdown markup for natural audio speech. */
 export declare function cleanForSpeech(raw: string): string;
+/**
+ * The browser voice that reads when the platform refuses: English, the cast
+ * voice's accent where this browser has it, its register where a name says so.
+ * Read per sentence, because the browser loads its voice list asynchronously.
+ */
+export declare function standIn({ gender, accent }: VoiceProfile, voices: {
+    name: string;
+    lang: string;
+}[]): string | undefined;
 export declare function stopAgentSpeech(): void;
 export interface SpeakOptions {
     agent?: string;
+    /** The reader's IAM bearer (`useIamToken`). Without one the platform refuses
+     *  and the browser reads, which is reported through `onRefusal`. */
+    token?: string | null;
+    /** Speaking began. Fires at once: the request for the audio is part of it. */
     onStart?: () => void;
+    /** Speaking stopped, for any reason: played out, stopped, or replaced. */
     onEnd?: () => void;
     onError?: (err: unknown) => void;
+    /** The platform refused, and whether the browser stood in. */
+    onRefusal?: (refusal: Refusal) => void;
 }
 /**
- * Speak text in the assigned voice of an agent.
+ * Speak text in the assigned voice of an agent, through the platform.
  * Returns a cancel function.
+ *
+ * Call it from the click that asks for it: the player is made inside that
+ * gesture, which is what lets Safari play the reply when it arrives.
  */
 export declare function speakAgent(text: string, options?: SpeakOptions): () => void;
 /**
@@ -19,6 +40,7 @@ export declare function speakAgent(text: string, options?: SpeakOptions): () =>
 export declare function useAgentSpeech(defaultAgent?: string): {
     speaking: boolean;
     currentText: string;
+    refusal: Refusal;
     speak: (text: string, agentName?: string) => void;
     stop: () => void;
     toggle: (text: string, agentName?: string) => void;

package/dist/speech.js

@@ -1,11 +1,15 @@
 'use client';
 import { Fragment as _Fragment, jsx as _jsx, jsxs as _jsxs } from "react/jsx-runtime";
-// Agent Speech & Voice Synthesis
+// Agent Speech
 //
 // Every agent in the Hanzo ecosystem has an assigned voice (`voiceOf(name)` from
-// `cast.tsx`). This module provides speech synthesis for agents across Chat,
-// Talk channels, and Meet video calls:
-// - Direct browser SpeechSynthesis with agent-tailored pitch, rate, and voice matching
+// `cast.tsx`), and the platform reads it: `/v1/audio/speech` through
+// @hanzo/voice's `speech()` transport and `mouth()`, in that voice. The
+// browser's own speechSynthesis reads only when the platform refuses, in a
+// voice of the same register and accent, and the refusal is reported rather
+// than passed off as the platform's voice.
+//
+// - The Listen button on a message, the Meet speaker test and its briefing
 // - Audio wave visualization during playback
 // - Clean cancellation and state management
 // - Text normalization (strips code fences, markdown syntax so speech sounds natural)
@@ -13,7 +17,10 @@ import { useEffect, useRef, useState, useCallback } from 'react';
 import { Volume2, Square } from 'lucide-react';
 import { View, XStack } from '@hanzo/gui';
 import { Button } from '@hanzo/ui';
+import { useIamToken } from '@hanzo/iam/react';
+import { mouth, speech, REFUSED } from '@hanzo/voice';
 import { GOOD } from './lib/mix.js';
+import { base } from './lib/ai.js';
 import { voiceOf, voiceProfileOf } from './cast.js';
 /** Strips code blocks, links, headers, and markdown markup for natural audio speech. */
 export function cleanForSpeech(raw) {
@@ -41,24 +48,41 @@ export function cleanForSpeech(raw) {
         .replace(/\s+/g, ' ')
         .trim();
 }
-/** Currently active speech synthesis utterance tracker. */
-let activeUtterance = null;
+/** Browser voice names by register, for the stand-in. Word-bounded: "Female" holds "male". */
+const REGISTER = {
+    female: /\b(female|samantha|victoria|karen|zira|serena|moira|tessa)\b/i,
+    male: /\b(male|daniel|alex|fred|david|arthur|oliver)\b/i,
+};
+/**
+ * The browser voice that reads when the platform refuses: English, the cast
+ * voice's accent where this browser has it, its register where a name says so.
+ * Read per sentence, because the browser loads its voice list asynchronously.
+ */
+export function standIn({ gender, accent }, voices) {
+    const english = voices.filter((v) => v.lang.toLowerCase().startsWith('en'));
+    const local = english.filter((v) => v.lang.toLowerCase().replace('_', '-') === (accent === 'british' ? 'en-gb' : 'en-us'));
+    const pick = (list) => list.find((v) => REGISTER[gender].test(v.name));
+    return (pick(local) ?? pick(english) ?? local[0] ?? english[0] ?? voices[0])?.name;
+}
+/** The reply being read aloud on this page. One at a time: a new one stops the last. */
+let active = null;
 const activeListeners = new Set();
+const notify = () => {
+    for (const listener of activeListeners)
+        listener();
+};
 export function stopAgentSpeech() {
-    if (typeof window !== 'undefined' && 'speechSynthesis' in window) {
-        try {
-            window.speechSynthesis.cancel();
-        }
-        catch { }
-    }
-    activeUtterance = null;
-    for (const notify of activeListeners) {
-        notify();
-    }
+    const lips = active;
+    active = null;
+    lips?.hush();
+    notify();
 }
 /**
- * Speak text in the assigned voice of an agent.
+ * Speak text in the assigned voice of an agent, through the platform.
  * Returns a cancel function.
+ *
+ * Call it from the click that asks for it: the player is made inside that
+ * gesture, which is what lets Safari play the reply when it arrives.
  */
 export function speakAgent(text, options = {}) {
     if (typeof window === 'undefined')
@@ -67,77 +91,38 @@ export function speakAgent(text, options = {}) {
     const clean = cleanForSpeech(text);
     if (!clean)
         return () => { };
-    if (!('speechSynthesis' in window)) {
-        options.onError?.(new Error('SpeechSynthesis not supported in this browser'));
-        return () => { };
-    }
     const profile = voiceProfileOf(options.agent);
-    const utterance = new SpeechSynthesisUtterance(clean);
-    utterance.pitch = profile.pitch;
-    utterance.rate = profile.rate;
-    // Attempt to select best matching system voice based on gender / voice tone
-    try {
-        const voices = window.speechSynthesis.getVoices();
-        if (voices.length > 0) {
-            const isFemale = profile.gender === 'female';
-            const candidate = voices.find((v) => {
-                const name = v.name.toLowerCase();
-                const lang = v.lang.toLowerCase();
-                if (!lang.startsWith('en'))
-                    return false;
-                if (isFemale) {
-                    return name.includes('female') || name.includes('samantha') || name.includes('victoria') || name.includes('karen') || name.includes('zira');
-                }
-                else {
-                    return name.includes('male') || name.includes('daniel') || name.includes('alex') || name.includes('fred') || name.includes('david');
-                }
-            }) || voices.find((v) => v.lang.toLowerCase().startsWith('en')) || voices[0];
-            if (candidate) {
-                utterance.voice = candidate;
-            }
-        }
-    }
-    catch { }
-    utterance.onstart = () => {
-        activeUtterance = utterance;
-        options.onStart?.();
-    };
-    utterance.onend = () => {
-        if (activeUtterance === utterance) {
-            activeUtterance = null;
-        }
+    const lips = mouth({
+        speech: speech({ ...base(), ...(options.token ? { token: options.token } : {}) }),
+        voice: () => standIn(profile, window.speechSynthesis?.getVoices() ?? []),
+        refused: options.onRefusal,
+    });
+    active = lips;
+    options.onStart?.();
+    lips.say(clean, profile.voice).then(() => {
+        if (active === lips)
+            active = null;
         options.onEnd?.();
-        for (const notify of activeListeners) {
-            notify();
-        }
-    };
-    utterance.onerror = (e) => {
-        if (activeUtterance === utterance) {
-            activeUtterance = null;
-        }
-        options.onError?.(e);
-        for (const notify of activeListeners) {
-            notify();
-        }
-    };
-    try {
-        window.speechSynthesis.speak(utterance);
-    }
-    catch (err) {
+        notify();
+    }, (err) => {
+        if (active === lips)
+            active = null;
         options.onError?.(err);
-    }
+        notify();
+    });
     return () => {
-        if (activeUtterance === utterance) {
+        if (active === lips)
             stopAgentSpeech();
-        }
     };
 }
 /**
  * Hook for managing agent speech playback in UI components.
  */
 export function useAgentSpeech(defaultAgent) {
+    const { token } = useIamToken();
     const [speaking, setSpeaking] = useState(false);
     const [currentText, setCurrentText] = useState(null);
+    const [refusal, setRefusal] = useState(null);
     const cancelRef = useRef(null);
     useEffect(() => {
         const listener = () => {
@@ -160,8 +145,10 @@ export function useAgentSpeech(defaultAgent) {
         stop();
         setCurrentText(text);
         setSpeaking(true);
+        setRefusal(null);
         cancelRef.current = speakAgent(text, {
             agent: agentName || defaultAgent,
+            token,
             onStart: () => setSpeaking(true),
             onEnd: () => {
                 setSpeaking(false);
@@ -171,8 +158,9 @@ export function useAgentSpeech(defaultAgent) {
                 setSpeaking(false);
                 setCurrentText(null);
             },
+            onRefusal: setRefusal,
         });
-    }, [defaultAgent, stop]);
+    }, [defaultAgent, stop, token]);
     const toggle = useCallback((text, agentName) => {
         if (speaking && currentText === text) {
             stop();
@@ -181,7 +169,7 @@ export function useAgentSpeech(defaultAgent) {
             speak(text, agentName);
         }
     }, [speaking, currentText, speak, stop]);
-    return { speaking, currentText, speak, stop, toggle };
+    return { speaking, currentText, refusal, speak, stop, toggle };
 }
 /** The five bars' resting heights, as a share of the 12px the wave stands in. */
 const BARS = [0.4, 0.9, 0.6, 1, 0.5];
@@ -219,11 +207,14 @@ export function AudioWave({ active = true, color = 'var(--state-success)' }) {
  * Allows one-click audio playback in the agent's voice.
  */
 export function SpeakButton({ text, agentName, compact = false, }) {
-    const { speaking, toggle } = useAgentSpeech(agentName);
+    const { speaking, refusal, toggle } = useAgentSpeech(agentName);
     const voice = voiceOf(agentName);
+    // A browser voice standing in for a refused platform sounds like success;
+    // the label is the one place that says otherwise.
+    const stood = refusal ? ` ${REFUSED[refusal.covered ? 'covered' : 'lost']}` : '';
     // An @hanzo/ui Button: ghost at rest, the quiet pressed ground while it
     // speaks. The word for speaking is the one status here, so it alone takes
     // the success ink (GOOD holds 4.5:1 on either theme); the icons follow the
     // label through currentColor.
-    return (_jsx(Button, { variant: speaking ? 'secondary' : 'ghost', size: compact ? 'icon-sm' : 'sm', color: speaking ? GOOD : '$soft', onClick: () => toggle(text, agentName), "aria-label": speaking ? 'Stop speaking' : `Listen to ${agentName || 'agent'} (${voice})`, "aria-pressed": speaking, title: speaking ? 'Stop speaking' : `Listen in ${agentName || 'agent'}'s voice (${voice})`, children: speaking ? (_jsxs(_Fragment, { children: [_jsx(AudioWave, { active: true, color: GOOD }), !compact && 'Speaking…', _jsx(Square, { size: 10, fill: "currentColor" })] })) : (_jsxs(_Fragment, { children: [_jsx(Volume2, { size: 13 }), !compact && 'Listen'] })) }));
+    return (_jsx(Button, { variant: speaking ? 'secondary' : 'ghost', size: compact ? 'icon-sm' : 'sm', color: speaking ? GOOD : '$soft', onClick: () => toggle(text, agentName), "aria-label": (speaking ? 'Stop speaking' : `Listen to ${agentName || 'agent'} (${voice})`) + stood, "aria-pressed": speaking, "data-refusal": refusal ? refusal.service : undefined, title: (speaking ? 'Stop speaking' : `Listen in ${agentName || 'agent'}'s voice (${voice})`) + stood, children: speaking ? (_jsxs(_Fragment, { children: [_jsx(AudioWave, { active: true, color: GOOD }), !compact && 'Speaking…', _jsx(Square, { size: 10, fill: "currentColor" })] })) : (_jsxs(_Fragment, { children: [_jsx(Volume2, { size: 13 }), !compact && 'Listen'] })) }));
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hanzo/rooms",
-  "version": "0.1.3",
+  "version": "0.1.4",
   "description": "The workspace rooms — Home, Chat, Inbox, Contacts, Meet, Cal, Drive, Board, Work, Bots — and the frame that holds them, once, for every Hanzo surface.",
   "type": "module",
   "main": "./dist/index.js",
@@ -56,7 +56,7 @@
     "@hanzo/plans": ">=1.8.9",
     "@hanzo/ui": ">=8.27.51",
     "@hanzo/usage": ">=0.1.17",
-    "@hanzo/voice": ">=0.1.8",
+    "@hanzo/voice": ">=0.1.10",
     "@hanzogui/shell": "^8.3.20",
     "lucide-react": ">=1.42.0",
     "react": ">=19.2.8",
@@ -67,6 +67,7 @@
     "@hanzo/appearance": "0.2.6",
     "@hanzo/build": "0.2.34",
     "@hanzo/composer": "0.2.0",
+    "@hanzo/design": "0.5.33",
     "@hanzo/event": "0.3.56",
     "@hanzo/gui": "8.3.5",
     "@hanzo/iam": "0.21.28",
@@ -74,9 +75,10 @@
     "@hanzo/plans": "1.8.9",
     "@hanzo/ui": "8.27.51",
     "@hanzo/usage": "0.1.17",
-    "@hanzo/voice": "0.1.8",
+    "@hanzo/voice": "0.1.10",
     "@hanzogui/config": "8.3.5",
     "@hanzogui/core": "8.3.5",
+    "@hanzogui/lucide-icons-2": "8.3.5",
     "@hanzogui/shell": "8.3.20",
     "@hanzogui/web": "8.3.5",
     "@types/node": "26.5.0",
@@ -87,9 +89,7 @@
     "react-dom": "19.2.8",
     "react-native-web": "0.21.2",
     "typescript": "7.0.2",
-    "vitest": "4.1.11",
-    "@hanzo/design": "0.5.33",
-    "@hanzogui/lucide-icons-2": "8.3.5"
+    "vitest": "4.1.11"
   },
   "repository": {
     "type": "git",