feat(dashboard): Faro chat with push-to-talk voice on the Conductor panel

This commit is contained in:
diegosouzapw
2026-07-22 21:11:47 -03:00
parent 9813f35ff0
commit 4b2bd2e37f
6 changed files with 356 additions and 2 deletions

View File

@@ -0,0 +1 @@
- feat(dashboard): Faro chat with voice on the Conductor panel — text via `/api/conductor/ask` (server-side proxy to the spokesperson; hub credential never reaches the browser; `pending` → Sim/Não confirmation buttons) and a guaranteed push-to-talk voice cycle (MediaRecorder → `/api/v1/audio/transcriptions` → ask → `/api/v1/audio/speech` playback), with operator-configurable STT/TTS models

View File

@@ -11,6 +11,8 @@ import { useTranslations } from "next-intl";
import { Badge, Card, ConfirmModal, DataTable, EmptyState, Modal } from "@/shared/components";
import FaroChat from "./FaroChat";
interface FleetRunner {
id: string;
name: string;
@@ -174,6 +176,8 @@ export default function ConductorPageClient() {
</>
)}
<FaroChat />
<Modal isOpen={detail !== null} onClose={() => setDetail(null)} title={t("detailTitle")} size="lg">
{detail && (
<div className="space-y-4 text-sm">

View File

@@ -0,0 +1,272 @@
"use client";
/**
* Faro chat with voice (Conductor PRD RF4). Text: input → /api/conductor/ask
* (server-side proxy — the hub credential never reaches the browser). When the
* answer carries `pending`, Faro is asking for confirmation: the Sim/Não
* buttons just send "sim"/"não" — the safety gate lives in Faro's engine.
*
* Voice (guaranteed cycle, PRD RF4): push-to-talk → MediaRecorder →
* POST /api/v1/audio/transcriptions (multipart) → text → /ask → response →
* POST /api/v1/audio/speech → play the returned audio blob. STT/TTS models are
* operator-configurable (provider/model of THIS OmniRoute install), persisted
* in localStorage.
*/
import { useEffect, useRef, useState } from "react";
import { useTranslations } from "next-intl";
import { Badge, Card } from "@/shared/components";
interface ChatMessage {
role: "user" | "faro";
text: string;
}
type VoiceState = "idle" | "listening" | "thinking" | "speaking";
const STT_KEY = "conductor.sttModel";
const TTS_KEY = "conductor.ttsModel";
function safeGet(key: string, fallback: string): string {
try {
return localStorage.getItem(key) || fallback;
} catch {
return fallback;
}
}
async function errorMessageOf(res: Response, fallback: string): Promise<string> {
try {
const data = await res.json();
return data?.error?.message ?? fallback;
} catch {
return fallback;
}
}
export default function FaroChat() {
const t = useTranslations("conductor");
const [messages, setMessages] = useState<ChatMessage[]>([]);
const [input, setInput] = useState("");
const [pending, setPending] = useState(false);
const [busy, setBusy] = useState(false);
const [voice, setVoice] = useState<VoiceState>("idle");
const [speak, setSpeak] = useState(false);
const [sttModel, setSttModel] = useState("openai/whisper-1");
const [ttsModel, setTtsModel] = useState("openai/tts-1");
const [err, setErr] = useState("");
const recorderRef = useRef<MediaRecorder | null>(null);
const logRef = useRef<HTMLDivElement>(null);
useEffect(() => {
setSttModel(safeGet(STT_KEY, "openai/whisper-1"));
setTtsModel(safeGet(TTS_KEY, "openai/tts-1"));
}, []);
useEffect(() => {
logRef.current?.scrollTo({ top: logRef.current.scrollHeight });
}, [messages]);
const persistModels = (stt: string, tts: string) => {
setSttModel(stt);
setTtsModel(tts);
try {
localStorage.setItem(STT_KEY, stt);
localStorage.setItem(TTS_KEY, tts);
} catch {
// modo privado: segue só em memória
}
};
const playAnswer = async (text: string) => {
setVoice("speaking");
try {
const res = await fetch("/api/v1/audio/speech", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({ model: ttsModel, input: text }),
});
if (!res.ok) {
setErr(await errorMessageOf(res, t("ttsFailed")));
return;
}
const url = URL.createObjectURL(await res.blob());
const audio = new Audio(url);
await audio.play().catch(() => undefined);
audio.onended = () => URL.revokeObjectURL(url);
} finally {
setVoice("idle");
}
};
const send = async (message: string, viaVoice = false) => {
const clean = message.trim();
if (!clean || busy) return;
setErr("");
setBusy(true);
setVoice("thinking");
setMessages((m) => [...m, { role: "user", text: clean }]);
setInput("");
try {
const res = await fetch("/api/conductor/ask", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({ message: clean }),
});
if (!res.ok) {
setErr(await errorMessageOf(res, t("faroOffline")));
return;
}
const data = await res.json();
setMessages((m) => [...m, { role: "faro", text: data.text }]);
setPending(Boolean(data.pending));
if (viaVoice && speak && data.text) await playAnswer(data.text);
} catch {
setErr(t("faroOffline"));
} finally {
setBusy(false);
setVoice((v) => (v === "thinking" ? "idle" : v));
}
};
const startRecording = async () => {
setErr("");
try {
const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
const recorder = new MediaRecorder(stream);
const chunks: Blob[] = [];
recorder.ondataavailable = (e) => e.data.size > 0 && chunks.push(e.data);
recorder.onstop = async () => {
stream.getTracks().forEach((track) => track.stop());
setVoice("thinking");
const blob = new Blob(chunks, { type: recorder.mimeType || "audio/webm" });
const form = new FormData();
form.append("model", sttModel);
form.append("file", new File([blob], "faro-ptt.webm", { type: blob.type }));
try {
// multipart: sem header manual — o browser define o boundary
const res = await fetch("/api/v1/audio/transcriptions", { method: "POST", body: form });
if (!res.ok) {
setErr(await errorMessageOf(res, t("sttFailed")));
setVoice("idle");
return;
}
const data = await res.json();
if (data.text) await send(data.text, true);
else setVoice("idle");
} catch {
setErr(t("sttFailed"));
setVoice("idle");
}
};
recorderRef.current = recorder;
recorder.start();
setVoice("listening");
} catch {
setErr(t("micDenied"));
}
};
const stopRecording = () => {
if (recorderRef.current?.state === "recording") recorderRef.current.stop();
recorderRef.current = null;
};
const voiceLabel: Record<VoiceState, string> = {
idle: t("voiceIdle"),
listening: t("voiceListening"),
thinking: t("voiceThinking"),
speaking: t("voiceSpeaking"),
};
return (
<Card title={t("faroTitle")} subtitle={t("faroSubtitle")}>
<div className="space-y-3">
<div ref={logRef} className="max-h-72 overflow-y-auto space-y-2 text-sm">
{messages.length === 0 && <p className="text-text-muted text-xs">{t("faroEmpty")}</p>}
{messages.map((m, i) => (
<div key={i} className={m.role === "user" ? "text-right" : "text-left"}>
<span
className={
m.role === "user"
? "inline-block rounded px-2 py-1 bg-primary/10"
: "inline-block rounded px-2 py-1 bg-black/5 dark:bg-white/10 whitespace-pre-wrap"
}
>
{m.text}
</span>
</div>
))}
</div>
{err && <Badge variant="error">{err}</Badge>}
{pending && (
<div className="flex items-center gap-2">
<Badge variant="warning" dot>{t("faroPending")}</Badge>
<button type="button" className="text-sm underline" onClick={() => void send("sim")}>
{t("yes")}
</button>
<button type="button" className="text-sm underline" onClick={() => void send("não")}>
{t("no")}
</button>
</div>
)}
<div className="flex items-center gap-2">
<input
className="flex-1 rounded border border-black/10 dark:border-white/10 bg-transparent px-2 py-1 text-sm"
placeholder={t("faroPlaceholder")}
value={input}
onChange={(e) => setInput(e.target.value)}
onKeyDown={(e) => {
if (e.key === "Enter") void send(input);
}}
disabled={busy}
/>
<button type="button" className="text-sm underline" onClick={() => void send(input)} disabled={busy}>
{t("faroSend")}
</button>
<button
type="button"
className={`text-sm px-2 py-1 rounded ${voice === "listening" ? "bg-red-500/20" : "bg-black/5 dark:bg-white/10"}`}
title={t("pushToTalk")}
aria-pressed={voice === "listening"}
onMouseDown={() => void startRecording()}
onMouseUp={stopRecording}
onMouseLeave={stopRecording}
onTouchStart={() => void startRecording()}
onTouchEnd={stopRecording}
>
🎙 {voiceLabel[voice]}
</button>
<label className="flex items-center gap-1 text-xs text-text-muted">
<input type="checkbox" checked={speak} onChange={(e) => setSpeak(e.target.checked)} />
{t("speakAnswers")}
</label>
</div>
<details className="text-xs text-text-muted">
<summary>{t("voiceModels")}</summary>
<div className="flex gap-2 pt-2">
<label className="flex-1">
STT
<input
className="w-full rounded border border-black/10 dark:border-white/10 bg-transparent px-2 py-1"
value={sttModel}
onChange={(e) => persistModels(e.target.value, ttsModel)}
/>
</label>
<label className="flex-1">
TTS
<input
className="w-full rounded border border-black/10 dark:border-white/10 bg-transparent px-2 py-1"
value={ttsModel}
onChange={(e) => persistModels(sttModel, e.target.value)}
/>
</label>
</div>
</details>
</div>
</Card>
);
}

View File

@@ -12116,6 +12116,25 @@
"cancelConfirmMessage": "The hub will abort the execution on the runner. This cannot be undone.",
"cancelFailed": "The hub refused the cancellation",
"close": "Close",
"error": "Error"
"error": "Error",
"faroTitle": "Faro — fleet spokesperson",
"faroSubtitle": "Chat and voice, anchored in real fleet events. Destructive commands always ask for confirmation.",
"faroEmpty": "Ask about the fleet — e.g. \"how is the fleet?\"",
"faroPending": "Faro is asking for confirmation",
"faroPlaceholder": "talk to Faro…",
"faroSend": "send",
"faroOffline": "Faro (spokesperson) is offline",
"yes": "Yes",
"no": "No",
"pushToTalk": "Hold to talk",
"speakAnswers": "speak answers",
"voiceModels": "Voice models (provider/model of this OmniRoute)",
"voiceIdle": "talk",
"voiceListening": "listening…",
"voiceThinking": "thinking…",
"voiceSpeaking": "speaking…",
"sttFailed": "Transcription failed (check the STT model/provider)",
"ttsFailed": "Speech synthesis failed (check the TTS model/provider)",
"micDenied": "Microphone unavailable or permission denied"
}
}

View File

@@ -12116,6 +12116,25 @@
"cancelConfirmMessage": "O hub aborta a execução no runner. Não dá para desfazer.",
"cancelFailed": "O hub recusou o cancelamento",
"close": "Fechar",
"error": "Erro"
"error": "Erro",
"faroTitle": "Faro — porta-voz da frota",
"faroSubtitle": "Chat e voz, ancorados nos eventos reais da frota. Comandos destrutivos sempre pedem confirmação.",
"faroEmpty": "Pergunte sobre a frota — ex.: \"como está a frota?\"",
"faroPending": "O Faro está pedindo confirmação",
"faroPlaceholder": "fale com o Faro…",
"faroSend": "enviar",
"faroOffline": "Faro (porta-voz) fora do ar",
"yes": "Sim",
"no": "Não",
"pushToTalk": "Segure para falar",
"speakAnswers": "falar as respostas",
"voiceModels": "Modelos de voz (provider/model deste OmniRoute)",
"voiceIdle": "falar",
"voiceListening": "ouvindo…",
"voiceThinking": "pensando…",
"voiceSpeaking": "falando…",
"sttFailed": "Transcrição falhou (confira modelo/provider de STT)",
"ttsFailed": "Síntese de voz falhou (confira modelo/provider de TTS)",
"micDenied": "Microfone indisponível ou permissão negada"
}
}

View File

@@ -0,0 +1,39 @@
import test from "node:test";
import assert from "node:assert/strict";
import fs from "node:fs";
import path from "node:path";
// Invariantes de segurança/UX do chat com voz (componentes React: vitest-ui advisory + dashboard-typecheck).
const CHAT = "src/app/(dashboard)/dashboard/conductor/FaroChat.tsx";
function src(): string {
return fs.readFileSync(path.join(process.cwd(), CHAT), "utf8");
}
test("client fala SÓ com o OmniRoute: /api/conductor/ask + /api/v1/audio/* (nunca Faro/hub direto)", () => {
const s = src();
assert.match(s, /"use client"/);
assert.match(s, /\/api\/conductor\/ask/);
assert.match(s, /\/api\/v1\/audio\/transcriptions/);
assert.match(s, /\/api\/v1\/audio\/speech/);
assert.ok(!s.includes(":7920"), "endereço do Faro nunca no client");
assert.ok(!s.includes("CONDUCTOR_"), "nenhuma env do Conductor no client");
});
test("pending do Faro → botões Sim/Não que enviam 'sim'/'não' (trava de confirmação é do motor do Faro)", () => {
const s = src();
assert.match(s, /pending/);
assert.match(s, /"sim"/);
assert.match(s, /"não"/);
});
test("voz: push-to-talk com MediaRecorder/getUserMedia; STT multipart sem Content-Type manual; TTS via Blob com revoke", () => {
const s = src();
assert.match(s, /navigator\.mediaDevices\.getUserMedia/);
assert.match(s, /MediaRecorder/);
assert.match(s, /FormData\(\)/);
assert.ok(!/audio\/transcriptions[\s\S]{0,300}content-type/i.test(s), "multipart deixa o browser definir o boundary");
assert.match(s, /URL\.createObjectURL/);
assert.match(s, /URL\.revokeObjectURL/);
assert.match(s, /useTranslations\("conductor"\)/);
});