@agentprojectcontext/apx 1.50.1 → 1.51.0

package/src/interfaces/web/src/components/voice/VoiceProviderModal.tsx

@@ -1,14 +1,17 @@
 import { useEffect, useState } from "react";
-import { Button, Dialog, Field, Input, Textarea } from "../ui";
+import { Button, Dialog, Field, Input, Switch, Textarea } from "../ui";
 import { UiSelect } from "../UiSelect";
 import { isSecretMarker, secretSuffix } from "../../lib/secrets";
 import {
+  DEFAULT_EMOTION_TAGS,
   ELEVENLABS_MODELS,
   GEMINI_TTS_VOICES,
   OPENAI_TTS_MODELS,
   OPENAI_TTS_VOICES,
   TTS_PROVIDER_META,
+  type CustomTtsConfig,
   type ElevenLabsConfig,
+  type EmotionsConfig,
   type GeminiTtsConfig,
   type OpenAiTtsConfig,
   type PiperConfig,
@@ -39,6 +42,35 @@ function str(v: unknown): string {
   return typeof v === "string" ? v : v == null ? "" : String(v);
 }
 
+// Stable config slug from a display label (e.g. "My QVox 🎙" → "my-qvox").
+function slugify(label: string): string {
+  return label
+    .toLowerCase()
+    .normalize("NFD").replace(/[\u0300-\u036f]/g, "")
+    .replace(/[^a-z0-9]+/g, "-")
+    .replace(/^-+|-+$/g, "")
+    .slice(0, 40);
+}
+
+// Generic inline-emotion-tags control. Enabling it makes the agent emit [tag]
+// markers in voice mode; the daemon teaches the syntax only when this engine
+// will speak, and strips stray tags when it won't (so they're never read aloud).
+function EmotionsField({ on, setOn, tags, setTags }: {
+  on: boolean; setOn: (v: boolean) => void; tags: string; setTags: (v: string) => void;
+}) {
+  return (
+    <div className="rounded-md border border-border/60 p-3 space-y-2">
+      <Switch checked={on} onChange={setOn} label={t("voice_ui.emotions_label")} />
+      <p className="text-xs text-muted-fg">{t("voice_ui.emotions_hint")}</p>
+      {on && (
+        <Field label={t("voice_ui.emotions_tags_label")} hint={t("voice_ui.emotions_tags_hint")}>
+          <Input value={tags} onChange={(e) => setTags(e.target.value)} placeholder={DEFAULT_EMOTION_TAGS.join(", ")} />
+        </Field>
+      )}
+    </div>
+  );
+}
+
 export function VoiceProviderModal({ open, providerId, config, onClose, onSave }: Props) {
   const [busy, setBusy] = useState(false);
   const [error, setError] = useState<string | null>(null);
@@ -46,13 +78,34 @@ export function VoiceProviderModal({ open, providerId, config, onClose, onSave }
   // Field state (the api_key field is always blank on open — typing replaces).
   const [apiKey, setApiKey] = useState("");
   const [f, setF] = useState<Record<string, string>>({});
+  // Generic emotion-tags capability (openai/gemini). Separate state because it
+  // saves as nested keys, not flat strings.
+  const [emoOn, setEmoOn] = useState(false);
+  const [emoTags, setEmoTags] = useState("");
+  const [showAdvanced, setShowAdvanced] = useState(false);
 
   useEffect(() => {
     if (!open || !providerId) return;
     setError(null);
     setApiKey("");
     const c = config || {};
-    if (providerId === "piper") {
+    const emo = (c as { emotions?: EmotionsConfig }).emotions;
+    setEmoOn(!!emo?.enabled);
+    setEmoTags(Array.isArray(emo?.tags) ? emo!.tags!.join(", ") : "");
+    setShowAdvanced(false);
+    const customMode = providerId === "__new__" || providerId.startsWith("custom:");
+    if (customMode) {
+      const p = c as unknown as CustomTtsConfig;
+      setF({
+        label: str(p.label),
+        base_url: str(p.base_url),
+        model: str(p.model),
+        voice: str(p.voice),
+        format: str(p.format),
+        style: str(p.style),
+        temperature: str(p.temperature),
+      });
+    } else if (providerId === "piper") {
       const p = c as PiperConfig;
       setF({ bin: str(p.bin), model: str(p.model), speaker: str(p.speaker) });
     } else if (providerId === "elevenlabs") {
@@ -71,18 +124,35 @@ export function VoiceProviderModal({ open, providerId, config, onClose, onSave }
 
   if (!providerId) return null;
 
+  const isCreate = providerId === "__new__";
+  const isCustom = isCreate || providerId.startsWith("custom:");
   const meta = TTS_PROVIDER_META[providerId];
-  const base = `voice.tts.${providerId}`;
   const up = (patch: Record<string, string>) => setF((s) => ({ ...s, ...patch }));
 
   const hasSecret = providerId !== "piper" && providerId !== "mock";
   const existingKey = hasSecret && isSecretMarker((config as { api_key?: unknown })?.api_key);
   const keyPlaceholder = existingKey ? t("voice_ui.api_key_set", { suffix: secretSuffix((config as { api_key?: unknown })?.api_key) ?? "" }) : t("voice_ui.api_key_label");
 
+  const title = isCreate
+    ? t("voice_ui.new_provider")
+    : isCustom
+      ? (f.label || providerId.slice(7))
+      : meta?.name || providerId;
+
   const submit = async () => {
     setBusy(true);
     setError(null);
     try {
+      // Resolve the config base. Custom providers live under
+      // voice.tts.custom.<slug>; built-ins are flat voice.tts.<id>.
+      const slug = isCreate ? slugify(f.label) : isCustom ? providerId.slice(7) : "";
+      if (isCustom) {
+        if (!f.label.trim()) throw new Error(t("voice_ui.err_label_required"));
+        if (!f.base_url.trim()) throw new Error(t("voice_ui.err_base_url_required"));
+        if (!slug) throw new Error(t("voice_ui.err_label_required"));
+      }
+      const base = isCustom ? `voice.tts.custom.${slug}` : `voice.tts.${providerId}`;
+
       const set: Record<string, unknown> = {};
       const unset: string[] = [];
       const opt = (key: string, val: string) => {
@@ -90,7 +160,17 @@ export function VoiceProviderModal({ open, providerId, config, onClose, onSave }
         else unset.push(`${base}.${key}`);
       };
 
-      if (providerId === "piper") {
+      if (isCustom) {
+        set[`${base}.label`] = f.label.trim();
+        set[`${base}.base_url`] = f.base_url.trim();
+        opt("style", f.style);
+        if (f.temperature.trim() && !Number.isNaN(Number(f.temperature)))
+          set[`${base}.temperature`] = Number(f.temperature);
+        else unset.push(`${base}.temperature`);
+        // Advanced (optional): model/voice for non-QVox OpenAI-compatible servers.
+        opt("model", f.model);
+        opt("voice", f.voice);
+      } else if (providerId === "piper") {
         opt("bin", f.bin);
         opt("model", f.model);
         if (f.speaker.trim()) set[`${base}.speaker`] = f.speaker.trim();
@@ -109,6 +189,14 @@ export function VoiceProviderModal({ open, providerId, config, onClose, onSave }
         opt("style", f.style);
       }
 
+      // Generic emotion-tags capability (engines that parse inline [tags]).
+      if (isCustom || providerId === "gemini") {
+        set[`${base}.emotions.enabled`] = emoOn;
+        const tags = emoTags.split(",").map((t) => t.trim().toLowerCase()).filter(Boolean);
+        if (tags.length) set[`${base}.emotions.tags`] = tags;
+        else unset.push(`${base}.emotions.tags`);
+      }
+
       // Only push a key the user actually typed (blank keeps the stored one).
       if (hasSecret && apiKey.trim()) set[`${base}.api_key`] = apiKey.trim();
 
@@ -125,8 +213,8 @@ export function VoiceProviderModal({ open, providerId, config, onClose, onSave }
     <Dialog
       open={open}
       onClose={onClose}
-      title={t("voice_screen.configure_provider", { name: meta?.name || providerId || "" })}
-      description={meta?.note}
+      title={t("voice_screen.configure_provider", { name: title })}
+      description={isCustom ? t("voice_ui.custom_desc") : meta?.note}
       size="md"
       footer={
         <>
@@ -167,6 +255,7 @@ export function VoiceProviderModal({ open, providerId, config, onClose, onSave }
           </>
         )}
 
+        {/* Built-in OpenAI (cloud only). */}
         {providerId === "openai" && (
           <>
             <Field label={t("voice_ui.api_key_label")} hint={existingKey ? t("voice_ui.api_key_keep_hint") : t("voice_ui.api_key_reuse_hint", { engine: "engines.openai.api_key", env: "OPENAI_API_KEY" })}>
@@ -184,6 +273,47 @@ export function VoiceProviderModal({ open, providerId, config, onClose, onSave }
           </>
         )}
 
+        {/* Custom OpenAI-compatible provider (e.g. a local QVox server). */}
+        {isCustom && (
+          <>
+            <Field label={t("voice_ui.label_label")} hint={t("voice_ui.label_hint")}>
+              <Input value={f.label} onChange={(e) => up({ label: e.target.value })} placeholder="QVox" />
+            </Field>
+            <Field label={t("voice_ui.base_url_req_label")} hint={t("voice_ui.base_url_req_hint")}>
+              <Input value={f.base_url} onChange={(e) => up({ base_url: e.target.value })} placeholder="http://127.0.0.1:5111/v1" />
+            </Field>
+            <Field label={t("voice_ui.api_key_label")} hint={existingKey ? t("voice_ui.api_key_keep_hint") : t("voice_ui.api_key_optional_hint")}>
+              <Input type="password" autoComplete="new-password" value={apiKey} onChange={(e) => setApiKey(e.target.value)} placeholder={keyPlaceholder} />
+            </Field>
+            <Field label={t("voice_ui.style_label")} hint={t("voice_ui.openai_style_hint")}>
+              <Textarea rows={2} value={f.style || ""} onChange={(e) => up({ style: e.target.value })} placeholder={t("voice_ui.style_ph")} />
+            </Field>
+            <Field label={t("voice_ui.temperature_label")} hint={t("voice_ui.temperature_hint")}>
+              <Input value={f.temperature} onChange={(e) => up({ temperature: e.target.value })} inputMode="decimal" placeholder="0.7" />
+            </Field>
+            <EmotionsField on={emoOn} setOn={setEmoOn} tags={emoTags} setTags={setEmoTags} />
+            <div>
+              <button
+                type="button"
+                onClick={() => setShowAdvanced((s) => !s)}
+                className="text-xs text-muted-fg hover:text-fg"
+              >
+                {showAdvanced ? "▾ " : "▸ "}{t("voice_ui.advanced")}
+              </button>
+              {showAdvanced && (
+                <div className="mt-2 space-y-3">
+                  <Field label={t("voice_ui.model_label")} hint={t("voice_ui.custom_model_hint")}>
+                    <Input value={f.model} onChange={(e) => up({ model: e.target.value })} placeholder={t("voice_ui.custom_optional_ph")} />
+                  </Field>
+                  <Field label={t("voice_ui.voice_label")} hint={t("voice_ui.custom_voice_hint")}>
+                    <Input value={f.voice} onChange={(e) => up({ voice: e.target.value })} placeholder={t("voice_ui.custom_optional_ph")} />
+                  </Field>
+                </div>
+              )}
+            </div>
+          </>
+        )}
+
         {providerId === "gemini" && (
           <>
             <Field label={t("voice_ui.api_key_label")} hint={existingKey ? t("voice_ui.api_key_keep_hint") : t("voice_ui.api_key_reuse_hint", { engine: "engines.gemini.api_key", env: "GEMINI_API_KEY" })}>
@@ -198,6 +328,7 @@ export function VoiceProviderModal({ open, providerId, config, onClose, onSave }
             <Field label={t("voice_ui.style_label")} hint={t("voice_ui.style_hint")}>
               <Textarea rows={2} value={f.style || ""} onChange={(e) => up({ style: e.target.value })} placeholder={t("voice_ui.style_ph")} />
             </Field>
+            <EmotionsField on={emoOn} setOn={setEmoOn} tags={emoTags} setTags={setEmoTags} />
           </>
         )}
 

package/src/interfaces/web/src/components/voice/VoiceTestCard.tsx

@@ -1,16 +1,16 @@
 import { useState } from "react";
 import { Play, Square, Volume2 } from "lucide-react";
-import { Button, Field, Input, Textarea } from "../ui";
+import { Button, Field, Textarea } from "../ui";
 import { UiSelect } from "../UiSelect";
 import { useToast } from "../Toast";
 import { useTtsPlayer } from "./useTtsPlayer";
 import { Voice, TTS_PROVIDER_META, type TtsEngineInfo, type TtsMode, type TtsSayResult } from "../../lib/api/voice";
 import { t } from "../../i18n";
 
-// "Decir esto" tester. Lets you pick which engine to synthesize with (overriding
-// the saved default) and add a free-text speaking-style instruction, then plays
-// the resulting audio in-browser via /tts/say. Style only affects engines that
-// support it (today: Gemini); other engines ignore it.
+// "Say this" tester. Lets you pick which engine to synthesize with (overriding
+// the saved default), then plays the resulting audio in-browser via /tts/say.
+// The base voice / emotions are configured per-engine now, so there's no
+// free-text style field here — pick the engine and hear it.
 
 interface Props {
   engines: TtsEngineInfo[];
@@ -25,7 +25,6 @@ export function VoiceTestCard({ engines, defaultProvider, mode }: Props) {
   const [text, setText] = useState(t("voice_ui.test_default_text"));
   // "" = use the saved default; otherwise force a specific engine.
   const [engine, setEngine] = useState("");
-  const [style, setStyle] = useState("");
   const [busy, setBusy] = useState(false);
   const [last, setLast] = useState<TtsSayResult | null>(null);
 
@@ -36,10 +35,15 @@ export function VoiceTestCard({ engines, defaultProvider, mode }: Props) {
 
   const options = [
     { value: "", label: defaultLabel },
-    ...engines.map((e) => ({
-      value: e.id,
-      label: `${TTS_PROVIDER_META[e.id]?.name || e.id}${e.available ? "" : t("voice_ui.test_unavailable_suffix")}`,
-    })),
+    ...engines
+      .filter((e) => e.id !== "mock")
+      .map((e) => ({
+        value: e.id,
+        label: e.custom ? e.label || e.id : TTS_PROVIDER_META[e.id]?.name || e.id,
+        // Unavailable engines stay listed (so the user sees they exist) but
+        // can't be picked.
+        disabled: !e.available,
+      })),
   ];
 
   const say = async () => {
@@ -53,7 +57,6 @@ export function VoiceTestCard({ engines, defaultProvider, mode }: Props) {
       const res = await Voice.say({
         text: txt,
         provider: engine || undefined,
-        style: style.trim() || undefined,
       });
       setLast(res);
       await play(res.audio_path);
@@ -66,19 +69,9 @@ export function VoiceTestCard({ engines, defaultProvider, mode }: Props) {
 
   return (
     <div className="space-y-3">
-      <div className="grid gap-3 sm:grid-cols-2">
-        <Field label={t("voice_ui.test_engine_label")} hint={t("voice_ui.test_engine_hint")}>
-          <UiSelect value={engine} onChange={setEngine} options={options} />
-        </Field>
-        <Field label={t("voice_ui.test_style_label")} hint={t("voice_ui.test_style_hint")}>
-          <Input
-            value={style}
-            onChange={(e) => setStyle(e.target.value)}
-            placeholder={t("voice_ui.style_ph")}
-            data-testid="voice-test-style"
-          />
-        </Field>
-      </div>
+      <Field label={t("voice_ui.test_engine_label")} hint={t("voice_ui.test_engine_hint")}>
+        <UiSelect value={engine} onChange={setEngine} options={options} />
+      </Field>
       <Field label={t("voice_ui.test_text_label")}>
         <Textarea
           rows={2}

package/src/interfaces/web/src/i18n/en.ts

@@ -85,6 +85,7 @@ export const en = {
       desktop: "Desktop",
       deck:  "Deck",
       code:  "Code",
+      web:   "Web",
     },
   },
   topbar: {
@@ -143,6 +144,7 @@ export const en = {
     appearance:  "Appearance",
     light_mode:  "Light",
     dark_mode:   "Dark",
+    system_mode: "System",
     language:    "Language",
     daemon:      "Daemon",
     daemon_sub:  "Status of the local process that serves this web and orchestrates agents.",
@@ -202,7 +204,7 @@ export const en = {
       owner_context_hint: "Who you are, what you work on, what the agent should know about you.",
       language: "Preferred language",
       timezone: "Timezone (IANA)",
-      timezone_hint: "e.g. America/New_York",
+      timezone_hint: "Auto-detected — search to change.",
       saved: "Identity saved.",
     },
 
@@ -1146,9 +1148,20 @@ export const en = {
     voice_id_label:       "Voice ID",
     voice_id_hint:        "ElevenLabs voice id (empty = default).",
     gemini_model_hint:    "Gemini TTS is still in preview.",
+    base_url_label:       "Base URL (optional)",
+    base_url_hint:        "OpenAI-compatible endpoint. Empty = OpenAI. Point it at a local server (e.g. a QVox / Qwen3-TTS daemon) to use that instead.",
+    openai_model_hint:    "tts-1 / tts-1-hd for OpenAI. Leave blank to let a custom server pick.",
+    openai_voice_hint:    "OpenAI preset (alloy…) or a custom server's preset (e.g. custom). Empty = server default.",
+    openai_style_hint:    "Base voice / instruct, used by custom endpoints (the persona kept across the audio). Ignored by stock OpenAI tts-1.",
     style_label:          "Style (how it should speak)",
     style_hint:           "Natural-language instruction. Empty = no style. E.g.: 'speak in a cheerful, unhurried tone'.",
     style_ph:             "speak in a cheerful, energetic tone",
+    temperature_label:    "Temperature (optional)",
+    temperature_hint:     "Sampling temperature for custom endpoints. Empty = server default.",
+    emotions_label:       "Inline emotion tags",
+    emotions_hint:        "When this engine speaks, let the agent drop [happy]/[whisper]-style tags into voice replies to color the delivery. Only enable it if this engine understands the tags (e.g. a QVox/Qwen3-TTS endpoint) — otherwise they're stripped before synthesis.",
+    emotions_tags_label:  "Allowed tags",
+    emotions_tags_hint:   "Comma-separated. Empty = the default set.",
     piper_bin_label:      "Binary (bin)",
     piper_bin_hint:       "Path or name of the piper CLI (PATH).",
     piper_model_label:    "Model (.onnx)",
@@ -1168,8 +1181,24 @@ export const en = {
     badge_unavailable:    "configured, unavailable",
     badge_not_configured: "not configured",
     badge_default:        "default",
+    badge_custom:         "custom",
     set_as_default:       "Set as default",
     configure:            "Configure",
+    remove:               "Remove",
+    remove_confirm:       "Remove this custom provider?",
+    add_provider:         "Add provider",
+    new_provider:         "New provider",
+    custom_note:          "Custom OpenAI-compatible endpoint.",
+    custom_desc:          "Any OpenAI-compatible speech endpoint (e.g. a local QVox / Qwen3-TTS server).",
+    label_label:          "Name",
+    label_hint:           "Display name for this provider.",
+    base_url_req_label:   "Base URL",
+    base_url_req_hint:    "Required. The OpenAI-compatible endpoint, e.g. http://127.0.0.1:5111/v1",
+    api_key_optional_hint:"Optional — only if your server requires a key.",
+    advanced:             "Advanced",
+    custom_model_hint:    "Optional. Most local servers ignore it (e.g. QVox).",
+    custom_voice_hint:    "Optional. A preset your server understands (e.g. custom). Empty = server default.",
+    custom_optional_ph:   "(optional)",
     stt_engine_label:     "Transcription engine",
     stt_engine_hint:      "Local uses faster-whisper (requires python3 + faster-whisper). OpenAI uses the engines.openai key.",
     stt_model_label:      "Local model (whisper)",
@@ -1217,7 +1246,7 @@ export const en = {
     stop:                 "Stop",
     replay:               "Replay",
     engine_result:        "Engine",
-    providers_desc:       "Synthesis engines. Status is reported live by the daemon. Pick which one to use by default.",
+    providers_desc:       "Synthesis engines, in fallback order. Status is reported live by the daemon. Add your own OpenAI-compatible endpoints.",
     providers_load_error: "Could not load providers: {msg}",
     test_desc:            "Pick which engine to synthesize with and, if applicable, how it should speak.",
     stt_desc:             "Speech-to-text engine used by the deck, Telegram, and the CLI when listening.",
@@ -1225,6 +1254,9 @@ export const en = {
     toast_mode_chain:     "Mode: chain with fallback.",
     toast_mode_single:    "Mode: default engine only.",
     toast_config_saved:   "Voice configuration saved.",
+    toast_provider_removed: "Provider removed.",
+    err_label_required:   "A name is required.",
+    err_base_url_required:"A base URL is required.",
     toast_transcription_updated: "Transcription updated.",
   },
 

package/src/interfaces/web/src/i18n/es.ts

@@ -86,6 +86,7 @@ export const es = {
       desktop: "Escritorio",
       deck:  "Deck",
       code:  "Code",
+      web:   "Web",
     },
   },
   topbar: {
@@ -144,6 +145,7 @@ export const es = {
     appearance:  "Apariencia",
     light_mode:  "Claro",
     dark_mode:   "Oscuro",
+    system_mode: "Sistema",
     language:    "Idioma",
     daemon:      "Daemon",
     daemon_sub:  "Estado del proceso local que sirve esta web y orquesta los agentes.",
@@ -203,7 +205,7 @@ export const es = {
       owner_context_hint: "Quién sos, en qué trabajás, qué le interesa al agente saber de vos.",
       language: "Idioma preferido",
       timezone: "Timezone (IANA)",
-      timezone_hint: "ej. America/Argentina/Buenos_Aires",
+      timezone_hint: "Detectado automáticamente — buscá para cambiar.",
       saved: "Identidad guardada.",
     },
 
@@ -1144,9 +1146,20 @@ export const es = {
     voice_id_label:       "Voice ID",
     voice_id_hint:        "Voice id de ElevenLabs (vacío = default).",
     gemini_model_hint:    "El TTS de Gemini todavía está en preview.",
+    base_url_label:       "Base URL (opcional)",
+    base_url_hint:        "Endpoint compatible con OpenAI. Vacío = OpenAI. Apuntalo a un servidor local (ej. un daemon QVox / Qwen3-TTS) para usar ese en su lugar.",
+    openai_model_hint:    "tts-1 / tts-1-hd para OpenAI. Dejalo vacío para que un servidor custom elija.",
+    openai_voice_hint:    "Preset de OpenAI (alloy…) o preset de tu servidor custom (ej. custom). Vacío = default del servidor.",
+    openai_style_hint:    "Voz base / instruct, usada por endpoints custom (la persona que se mantiene en todo el audio). El tts-1 de OpenAI la ignora.",
     style_label:          "Estilo (cómo debería hablar)",
     style_hint:           "Instrucción en lenguaje natural. Vacío = sin estilo. Ej.: 'hablá en un tono alegre y pausado'.",
     style_ph:             "hablá en un tono alegre y enérgico",
+    temperature_label:    "Temperatura (opcional)",
+    temperature_hint:     "Temperatura de sampleo para endpoints custom. Vacío = default del servidor.",
+    emotions_label:       "Tags de emoción inline",
+    emotions_hint:        "Cuando hable este motor, deja que el agente meta tags tipo [happy]/[whisper] en las respuestas de voz para darles color. Activalo solo si este motor entiende los tags (ej. un endpoint QVox/Qwen3-TTS) — si no, se quitan antes de sintetizar.",
+    emotions_tags_label:  "Tags permitidos",
+    emotions_tags_hint:   "Separados por coma. Vacío = el set por defecto.",
     piper_bin_label:      "Binario (bin)",
     piper_bin_hint:       "Ruta o nombre del CLI de piper (PATH).",
     piper_model_label:    "Modelo (.onnx)",
@@ -1166,8 +1179,24 @@ export const es = {
     badge_unavailable:    "configurado, no disponible",
     badge_not_configured: "sin configurar",
     badge_default:        "default",
+    badge_custom:         "custom",
     set_as_default:       "Usar como default",
     configure:            "Configurar",
+    remove:               "Quitar",
+    remove_confirm:       "¿Quitar este proveedor custom?",
+    add_provider:         "Agregar proveedor",
+    new_provider:         "Nuevo proveedor",
+    custom_note:          "Endpoint custom compatible con OpenAI.",
+    custom_desc:          "Cualquier endpoint de voz compatible con OpenAI (ej. un servidor local QVox / Qwen3-TTS).",
+    label_label:          "Nombre",
+    label_hint:           "Nombre para mostrar de este proveedor.",
+    base_url_req_label:   "Base URL",
+    base_url_req_hint:    "Requerido. El endpoint compatible con OpenAI, ej. http://127.0.0.1:5111/v1",
+    api_key_optional_hint:"Opcional — solo si tu servidor pide key.",
+    advanced:             "Avanzado",
+    custom_model_hint:    "Opcional. La mayoría de los servidores locales lo ignoran (ej. QVox).",
+    custom_voice_hint:    "Opcional. Un preset que entienda tu servidor (ej. custom). Vacío = default del servidor.",
+    custom_optional_ph:   "(opcional)",
     stt_engine_label:     "Engine de transcripción",
     stt_engine_hint:      "Local usa faster-whisper (requiere python3 + faster-whisper). OpenAI usa la key de engines.openai.",
     stt_model_label:      "Modelo local (whisper)",
@@ -1215,7 +1244,7 @@ export const es = {
     stop:                 "Detener",
     replay:               "Repetir",
     engine_result:        "Engine",
-    providers_desc:       "Engines de síntesis. El estado lo reporta el daemon en vivo. Elegí cuál usar por defecto.",
+    providers_desc:       "Engines de síntesis, en orden de fallback. El estado lo reporta el daemon en vivo. Agregá tus propios endpoints compatibles con OpenAI.",
     providers_load_error: "No pude cargar los proveedores: {msg}",
     test_desc:            "Elegí con qué engine sintetizar y, si aplica, cómo debería hablar.",
     stt_desc:             "Engine de speech-to-text que usan el deck, Telegram y la CLI al escuchar.",
@@ -1223,6 +1252,9 @@ export const es = {
     toast_mode_chain:     "Modo: cadena con fallback.",
     toast_mode_single:    "Modo: solo engine default.",
     toast_config_saved:   "Configuración de voz guardada.",
+    toast_provider_removed: "Proveedor eliminado.",
+    err_label_required:   "Falta el nombre.",
+    err_base_url_required:"Falta la base URL.",
     toast_transcription_updated: "Transcripción actualizada.",
   },
 

package/src/interfaces/web/src/lib/api/voice.ts

@@ -22,6 +22,9 @@ export interface TtsEngineInfo {
   available: boolean;   // probe says it can synthesize right now
   configured: boolean;  // has a non-empty voice.tts.<id> config block
   enabled: boolean;     // included in the fallback chain (voice.tts.<id>.enabled)
+  custom?: boolean;     // user-added OpenAI-compatible provider ("custom:<slug>")
+  label?: string;       // display name for custom providers
+  note?: string;        // e.g. the custom base_url
 }
 
 export interface TtsProvidersResponse {
@@ -67,11 +70,20 @@ export interface ElevenLabsConfig {
   voice_id?: string;
   output_format?: string;
 }
+/** Inline emotion-tags capability (generic; add it to any tag-aware engine). */
+export interface EmotionsConfig {
+  enabled?: boolean;
+  tags?: string[];      // empty → canonical default set
+}
 export interface OpenAiTtsConfig {
   api_key?: string;
-  model?: string;       // tts-1 | tts-1-hd
-  voice?: string;       // alloy | echo | fable | onyx | nova | shimmer
+  base_url?: string;    // custom OpenAI-compatible endpoint (e.g. a local QVox)
+  model?: string;       // tts-1 | tts-1-hd  (optional for custom endpoints)
+  voice?: string;       // alloy | echo | fable | onyx | nova | shimmer | preset
   format?: string;      // mp3 | opus | aac | flac | wav
+  style?: string;       // base voice / "instruct" (custom endpoints)
+  temperature?: number; // custom endpoints
+  emotions?: EmotionsConfig;
 }
 export interface GeminiTtsConfig {
   api_key?: string;
@@ -79,6 +91,21 @@ export interface GeminiTtsConfig {
   voice?: string;       // e.g. Kore
   style?: string;       // natural-language speaking-style instruction
   enabled?: boolean;
+  emotions?: EmotionsConfig;
+}
+
+/** A user-added OpenAI-compatible provider (voice.tts.custom.<slug>). */
+export interface CustomTtsConfig {
+  label?: string;       // display name
+  base_url: string;     // OpenAI-compatible endpoint (required)
+  api_key?: string;
+  model?: string;
+  voice?: string;
+  format?: string;
+  style?: string;       // base voice / instruct
+  temperature?: number;
+  emotions?: EmotionsConfig;
+  enabled?: boolean;
 }
 
 export interface VoiceTtsConfig {
@@ -89,6 +116,7 @@ export interface VoiceTtsConfig {
   elevenlabs?: ElevenLabsConfig;
   openai?: OpenAiTtsConfig;
   gemini?: GeminiTtsConfig;
+  custom?: Record<string, CustomTtsConfig>;  // user-added providers, by slug
 }
 
 export interface TranscriptionLocalConfig {
@@ -163,6 +191,11 @@ export const OPENAI_TTS_VOICES = ["alloy", "echo", "fable", "onyx", "nova", "shi
 export const GEMINI_TTS_VOICES = ["Kore", "Puck", "Charon", "Fenrir", "Aoede"];
 export const ELEVENLABS_MODELS = ["eleven_multilingual_v2", "eleven_turbo_v2_5", "eleven_flash_v2_5"];
 export const OPENAI_TTS_MODELS = ["tts-1", "tts-1-hd"];
+// Canonical inline emotion-tag set (mirrors the daemon's DEFAULT_EMOTION_TAGS).
+export const DEFAULT_EMOTION_TAGS = [
+  "happy", "sad", "excited", "angry", "calm",
+  "whisper", "shout", "laugh", "cry", "narrator", "neutral",
+];
 export const WHISPER_MODELS = ["tiny", "base", "small", "medium", "large-v2", "large-v3", "large-v3-turbo"];
 
 // Friendly labels + ordering for the provider list. The daemon is the source
@@ -170,7 +203,7 @@ export const WHISPER_MODELS = ["tiny", "base", "small", "medium", "large-v2", "l
 export const TTS_PROVIDER_META: Record<string, { name: string; note: string; local?: boolean }> = {
   piper:      { name: "Piper",       note: "Local, offline (CLI + .onnx model). No API key.", local: true },
   elevenlabs: { name: "ElevenLabs",  note: "Cloud, multilingual. Requires an API key." },
-  openai:     { name: "OpenAI",      note: "Cloud (tts-1 / tts-1-hd). Uses your OpenAI key." },
+  openai:     { name: "OpenAI",      note: "Cloud (tts-1 / tts-1-hd) or any OpenAI-compatible endpoint (set a base URL for a local server, e.g. QVox)." },
   gemini:     { name: "Gemini",      note: "Cloud (preview). Uses your Gemini key." },
   mock:       { name: "Mock",        note: "Silent test engine. Always available as a fallback.", local: true },
 };