iobroker.assistant 1.4.5 → 1.4.7

package/README.md

@@ -57,7 +57,9 @@ the origin in `assistant.0.text.querySource` (satellite name, `chat`, or empty f
 or — if it is a URL/path to an audio file (`.mp3`/`.wav`/…) — played back.
 
 > **Playing audio files needs [ffmpeg](https://ffmpeg.org/) on the host** (Windows and Linux):
-> Linux → `sudo apt install ffmpeg`; Windows → install ffmpeg and add it to the `PATH`.
+> Linux → `sudo apt install ffmpeg`; Windows → install ffmpeg and add it to the `PATH`, or enter its full path
+> under **Path to ffmpeg** on the Voice tab. Without either, **Download ffmpeg** on the Voice tab fetches a static
+> build (ffmpeg-static, 20–30 MB) into the data directory of the instance.
 > Plain-text TTS does not need it — only audio-file playback (mp3/wav/…) does.
 
 ### Start a Hannah satellite pointed at this adapter
@@ -235,6 +237,19 @@ train it twice.
     Placeholder for the next version (at the beginning of the line):
     ### **WORK IN PROGRESS**
 -->
+### 1.4.7 (2026-11-04)
+* (@GermanBluefox) Voice messages: `ask` also takes a recording (`audio` as base64 plus `format`, e.g. `ogg`) instead of text. It is recognised with the speech recognition of the Voice tab and answered like a text question; the reply carries the recognised text as `heard`. Telegram uses it for voice messages
+* (@GermanBluefox) Path to ffmpeg can be set on the Voice tab (empty = from the search path), with a button that checks it
+* (@GermanBluefox) Hosts without ffmpeg: "Download ffmpeg" on the Voice tab fetches a static build of ffmpeg-static (20-30 MB, checked against a fixed SHA-256) into the data directory of the instance; it is used while no path is entered
+* (@GermanBluefox) Pictures: `ask` takes `images` (base64 JPEG/PNG/GIF/WebP, at most 4) and passes them to the cloud model (OpenAI-compatible, Anthropic, pi-agent); without a text the question is "What is in this picture?". Rules and the local model are skipped for them. Telegram uses it for photos
+* (@GermanBluefox) Videos: `ask` takes a reference `video: { adapter, path }` into the file storage (never the content). Videos are not looked at yet: without a question the answer says so, with one the model is told
+* (@GermanBluefox) ESPHome: after a reply that asked something, every later reply re-opened the microphone. The firmware keeps `continue_conversation` until it is told "0", so every reply now says it, like Home Assistant does
+* (@GermanBluefox) Satellites only listen again after a reply that asks something: a [[LISTEN]] marker (pi-agent's confirmations) or a question mark at the end. A question mark in the middle of a reply no longer opens the microphone
+* (@GermanBluefox) ESPHome satellites with a screen: yes/no buttons (event `confirm_answer`) answer the open question like a spoken "yes"/"no" - the microphone closes, the reply is announced on that satellite. A select `display_language` is set to the voice language when the satellite connects
+
+### 1.4.6 (2026-11-03)
+* (@GermanBluefox) Speech recognition: own OpenAI-compatible server (Voice tab, provider OpenAI, "Speech recognition server"), e.g. a local whisper-large-v3-turbo in speaches, faster-whisper-server or LocalAI. It needs no OpenAI key and never gets it; an own key for the server is optional
+
 ### 1.4.5 (2026-11-03)
 * (@GermanBluefox) ESPHome satellites: spoken replies go out in the format the device's media player announces (FLAC at its sample rate, as Home Assistant does; MP3/Opus via ffmpeg). Since ESPHome 2026.10 the player is built with only the codecs its pipelines use, and a WAV reply did not play at all. Devices that announce no formats keep getting WAV
 * (@GermanBluefox) ESPHome satellites: when the device ends the conversation itself (button, or the wake word said again), it now gets its RUN_END. It stayed in the conversation, and every further wake word only sent another stop

package/admin/i18n/de.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "ESPHome-Satelliten",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "ESPHome-Satelliten bekommen die Antwort nicht über ihren API-Socket — sie holen sie als URL ab. Das ist der Port, auf dem dieser kleine HTTP-Server lauscht.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Leer = tts-1. Auch: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Leer = whisper-1. Auch: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Leer = whisper-1. Auch: gpt-4o-transcribe, gpt-4o-mini-transcribe. Mit eigenem Server: das Modell so, wie der Server es nennt, z. B. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Sprache aktivieren (STT/TTS für Satelliten)",
   "Enable voice server (satellites)": "Sprach-Server aktivieren (Satelliten)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Aktiviert Spracherkennung/-synthese. ioBroker-native Satelliten erreichen sie über den Nachrichtenbus — kein Port nötig. Den UDP-Server unten nur für ESP-/Hannah-Satelliten einschalten.",
@@ -222,5 +222,14 @@
   "Person": "Person",
   "Guest": "Gast",
   "Pet": "Haustier",
-  "Group": "Gruppe"
+  "Group": "Gruppe",
+  "Speech recognition server (optional)": "Server für die Spracherkennung (optional)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "OpenAI-kompatibler Server statt OpenAI, z. B. ein lokales Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Leer = OpenAI.",
+  "Key of the speech recognition server": "Schlüssel des Spracherkennungs-Servers",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Nur, wenn der Server einen verlangt. Der OpenAI-Schlüssel wird nie dorthin geschickt.",
+  "Path to ffmpeg": "Pfad zu ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Leer = das heruntergeladene ffmpeg, sonst ffmpeg aus dem Suchpfad. Wird für Timer-/Weckertöne, Telegram-Sprachnachrichten und MP3/Opus-Antworten gebraucht. Beispiel: /usr/bin/ffmpeg oder C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "ffmpeg prüfen",
+  "Download ffmpeg": "ffmpeg herunterladen",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Lädt ein statisches ffmpeg (ffmpeg-static, 20–30 MB) in das Datenverzeichnis der Instanz. Es wird genutzt, wenn kein Pfad eingetragen ist."
 }

package/admin/i18n/en.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "ESPHome satellites",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Enable voice (STT/TTS for satellites)",
   "Enable voice server (satellites)": "Enable voice server (satellites)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.",
@@ -222,5 +222,14 @@
   "Person": "Person",
   "Guest": "Guest",
   "Pet": "Pet",
-  "Group": "Group"
+  "Group": "Group",
+  "Speech recognition server (optional)": "Speech recognition server (optional)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.",
+  "Key of the speech recognition server": "Key of the speech recognition server",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Only if the server wants one. The OpenAI key is never sent to it.",
+  "Path to ffmpeg": "Path to ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "Check ffmpeg",
+  "Download ffmpeg": "Download ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered."
 }

package/admin/i18n/es.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "Satélites ESPHome",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "Los satélites ESPHome no reciben la respuesta por su socket de API: la descargan como URL. Este es el puerto en el que escucha ese pequeño servidor HTTP.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Vacío = tts-1. También: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Vacío = whisper-1. También: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Vacío = whisper-1. También: gpt-4o-transcribe, gpt-4o-mini-transcribe. Con un servidor propio: el modelo tal como lo llama el servidor, p. ej. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Habilitar la voz (STT/TTS para satélites)",
   "Enable voice server (satellites)": "Habilitar servidor de voz (satélites)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Permite el reconocimiento/síntesis de voz. Los satélites nativos de ioBroker se conectan a través del bus de mensajes; no se necesita ningún puerto. Active el servidor UDP que se indica a continuación solo para los satélites ESP/Hannah.",
@@ -222,5 +222,14 @@
   "Person": "Persona",
   "Guest": "Invitado",
   "Pet": "Mascota",
-  "Group": "Grupo"
+  "Group": "Grupo",
+  "Speech recognition server (optional)": "Servidor de reconocimiento de voz (opcional)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "Servidor compatible con OpenAI en lugar de OpenAI, p. ej. un Whisper local: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Vacío = OpenAI.",
+  "Key of the speech recognition server": "Clave del servidor de reconocimiento de voz",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Solo si el servidor la pide. La clave de OpenAI nunca se envía allí.",
+  "Path to ffmpeg": "Ruta de ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Vacío = el ffmpeg descargado, si no, ffmpeg de la ruta de búsqueda. Necesario para los sonidos de temporizador/alarma, los mensajes de voz de Telegram y las respuestas MP3/Opus. Ejemplo: /usr/bin/ffmpeg o C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "Comprobar ffmpeg",
+  "Download ffmpeg": "Descargar ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Descarga un ffmpeg estático (ffmpeg-static, 20-30 MB) en el directorio de datos de la instancia. Se usa cuando no se indica ninguna ruta."
 }

package/admin/i18n/fr.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "Satellites ESPHome",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "Les satellites ESPHome ne reçoivent pas la réponse par leur socket API : ils la récupèrent via une URL. C’est le port sur lequel écoute ce petit serveur HTTP.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Vide = tts-1. Également : tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Vide = whisper-1. Également : gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Vide = whisper-1. Également : gpt-4o-transcribe, gpt-4o-mini-transcribe. Avec un serveur propre : le modèle tel que le serveur le nomme, p. ex. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Activer la voix (STT/TTS pour les satellites)",
   "Enable voice server (satellites)": "Activer le serveur vocal (satellites)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Permet la reconnaissance et la synthèse vocales. Les satellites natifs d'ioBroker y accèdent via le bus de messages ; aucun port n'est nécessaire. Activez le serveur UDP ci-dessous uniquement pour les satellites ESP/Hannah.",
@@ -222,5 +222,14 @@
   "Person": "Personne",
   "Guest": "Invité",
   "Pet": "Animal",
-  "Group": "Groupe"
+  "Group": "Groupe",
+  "Speech recognition server (optional)": "Serveur de reconnaissance vocale (facultatif)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "Serveur compatible OpenAI à la place d'OpenAI, p. ex. un Whisper local : http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Vide = OpenAI.",
+  "Key of the speech recognition server": "Clé du serveur de reconnaissance vocale",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Seulement si le serveur en demande une. La clé OpenAI n'y est jamais envoyée.",
+  "Path to ffmpeg": "Chemin vers ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Vide = le ffmpeg téléchargé, sinon ffmpeg du chemin de recherche. Nécessaire pour les sons de minuterie/réveil, les messages vocaux Telegram et les réponses MP3/Opus. Exemple : /usr/bin/ffmpeg ou C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "Vérifier ffmpeg",
+  "Download ffmpeg": "Télécharger ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Télécharge un ffmpeg statique (ffmpeg-static, 20 à 30 Mo) dans le répertoire de données de l'instance. Il est utilisé quand aucun chemin n'est indiqué."
 }

package/admin/i18n/it.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "Satelliti ESPHome",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "I satelliti ESPHome non ricevono la risposta dal loro socket API: la scaricano come URL. Questa è la porta su cui ascolta quel piccolo server HTTP.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Vuoto = tts-1. Anche: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Vuoto = whisper-1. Anche: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Vuoto = whisper-1. Anche: gpt-4o-transcribe, gpt-4o-mini-transcribe. Con un server proprio: il modello come lo chiama il server, ad es. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Abilita la voce (STT/TTS per i satelliti)",
   "Enable voice server (satellites)": "Abilita il server vocale (satelliti)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Abilita il riconoscimento/sintesi vocale. I satelliti nativi di ioBroker lo raggiungono tramite il bus di messaggi, senza bisogno di porte. Attivare il server UDP qui sotto solo per i satelliti ESP/Hannah.",
@@ -222,5 +222,14 @@
   "Person": "Persona",
   "Guest": "Ospite",
   "Pet": "Animale",
-  "Group": "Gruppo"
+  "Group": "Gruppo",
+  "Speech recognition server (optional)": "Server di riconoscimento vocale (facoltativo)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "Server compatibile con OpenAI al posto di OpenAI, ad es. un Whisper locale: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Vuoto = OpenAI.",
+  "Key of the speech recognition server": "Chiave del server di riconoscimento vocale",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Solo se il server ne richiede una. La chiave OpenAI non viene mai inviata lì.",
+  "Path to ffmpeg": "Percorso di ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Vuoto = l'ffmpeg scaricato, altrimenti ffmpeg dal percorso di ricerca. Necessario per i suoni di timer/sveglia, i messaggi vocali di Telegram e le risposte MP3/Opus. Esempio: /usr/bin/ffmpeg o C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "Verifica ffmpeg",
+  "Download ffmpeg": "Scarica ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Scarica un ffmpeg statico (ffmpeg-static, 20-30 MB) nella directory dei dati dell'istanza. Viene usato quando non è indicato alcun percorso."
 }

package/admin/i18n/nl.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "ESPHome-satellieten",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "ESPHome-satellieten krijgen het antwoord niet via hun API-socket — ze halen het op als URL. Dit is de poort waarop die kleine HTTP-server luistert.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Leeg = tts-1. Ook: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Leeg = whisper-1. Ook: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Leeg = whisper-1. Ook: gpt-4o-transcribe, gpt-4o-mini-transcribe. Met een eigen server: het model zoals de server het noemt, bijv. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Schakel spraak in (STT/TTS voor satellieten)",
   "Enable voice server (satellites)": "Schakel de spraakserver (satellieten) in.",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Maakt spraakherkenning/synthese mogelijk. ioBroker-native satellieten bereiken deze via de berichtenbus — er is geen poort nodig. Schakel de onderstaande UDP-server alleen in voor ESP/Hannah-satellieten.",
@@ -222,5 +222,14 @@
   "Person": "Persoon",
   "Guest": "Gast",
   "Pet": "Huisdier",
-  "Group": "Groep"
+  "Group": "Groep",
+  "Speech recognition server (optional)": "Server voor spraakherkenning (optioneel)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "OpenAI-compatibele server in plaats van OpenAI, bijv. een lokale Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Leeg = OpenAI.",
+  "Key of the speech recognition server": "Sleutel van de spraakherkenningsserver",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Alleen als de server er een vraagt. De OpenAI-sleutel wordt er nooit naartoe gestuurd.",
+  "Path to ffmpeg": "Pad naar ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Leeg = de gedownloade ffmpeg, anders ffmpeg uit het zoekpad. Nodig voor timer-/wekkergeluiden, Telegram-spraakberichten en MP3/Opus-antwoorden. Voorbeeld: /usr/bin/ffmpeg of C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "ffmpeg controleren",
+  "Download ffmpeg": "ffmpeg downloaden",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Downloadt een statische ffmpeg (ffmpeg-static, 20-30 MB) naar de gegevensmap van de instantie. Die wordt gebruikt als er geen pad is ingevuld."
 }

package/admin/i18n/pl.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "Satelity ESPHome",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "Satelity ESPHome nie odbierają odpowiedzi przez swoje gniazdo API — pobierają ją jako URL. To port, na którym nasłuchuje ten mały serwer HTTP.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Pusty = tts-1. Również: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Pusty = szept-1. Również: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Puste = whisper-1. Również: gpt-4o-transcribe, gpt-4o-mini-transcribe. Z własnym serwerem: model tak, jak nazywa go serwer, np. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Włącz głos (STT/TTS dla satelitów)",
   "Enable voice server (satellites)": "Włącz serwer głosowy (satelity)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Umożliwia rozpoznawanie/syntezę mowy. Satelity natywne dla ioBroker osiągają go przez magistralę komunikatów — nie jest potrzebny żaden port. Włącz serwer UDP poniżej tylko dla satelitów ESP/Hannah.",
@@ -222,5 +222,14 @@
   "Person": "Osoba",
   "Guest": "Gość",
   "Pet": "Zwierzę",
-  "Group": "Grupa"
+  "Group": "Grupa",
+  "Speech recognition server (optional)": "Serwer rozpoznawania mowy (opcjonalnie)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "Serwer zgodny z OpenAI zamiast OpenAI, np. lokalny Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Puste = OpenAI.",
+  "Key of the speech recognition server": "Klucz serwera rozpoznawania mowy",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Tylko jeśli serwer go wymaga. Klucz OpenAI nigdy nie jest tam wysyłany.",
+  "Path to ffmpeg": "Ścieżka do ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Puste = pobrany ffmpeg, w przeciwnym razie ffmpeg ze ścieżki wyszukiwania. Potrzebny do dźwięków minutnika/budzika, wiadomości głosowych z Telegrama i odpowiedzi MP3/Opus. Przykład: /usr/bin/ffmpeg lub C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "Sprawdź ffmpeg",
+  "Download ffmpeg": "Pobierz ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Pobiera statyczny ffmpeg (ffmpeg-static, 20–30 MB) do katalogu danych instancji. Jest używany, gdy nie podano ścieżki."
 }

package/admin/i18n/pt.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "Satélites ESPHome",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "Os satélites ESPHome não recebem a resposta pelo seu socket de API — buscam-na como URL. Esta é a porta em que esse pequeno servidor HTTP escuta.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Vazio = tts-1. Também: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Vazio = whisper-1. Também: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Vazio = whisper-1. Também: gpt-4o-transcribe, gpt-4o-mini-transcribe. Com servidor próprio: o modelo como o servidor o chama, p. ex. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Ativar voz (STT/TTS para satélites)",
   "Enable voice server (satellites)": "Ativar servidor de voz (satélites)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Habilita o reconhecimento/síntese de voz. Os satélites nativos do ioBroker acessam o servidor através do barramento de mensagens — sem necessidade de porta. Ative o servidor UDP abaixo apenas para satélites ESP/Hannah.",
@@ -222,5 +222,14 @@
   "Person": "Pessoa",
   "Guest": "Convidado",
   "Pet": "Animal de estimação",
-  "Group": "Grupo"
+  "Group": "Grupo",
+  "Speech recognition server (optional)": "Servidor de reconhecimento de voz (opcional)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "Servidor compatível com OpenAI em vez da OpenAI, p. ex. um Whisper local: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Vazio = OpenAI.",
+  "Key of the speech recognition server": "Chave do servidor de reconhecimento de voz",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Só se o servidor pedir uma. A chave da OpenAI nunca é enviada para lá.",
+  "Path to ffmpeg": "Caminho para o ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Vazio = o ffmpeg transferido, caso contrário ffmpeg do caminho de pesquisa. Necessário para os sons de temporizador/alarme, as mensagens de voz do Telegram e as respostas MP3/Opus. Exemplo: /usr/bin/ffmpeg ou C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "Verificar ffmpeg",
+  "Download ffmpeg": "Transferir ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Transfere um ffmpeg estático (ffmpeg-static, 20-30 MB) para o diretório de dados da instância. É usado quando nenhum caminho é indicado."
 }

package/admin/i18n/ru.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "Сателлиты ESPHome",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "Сателлиты ESPHome получают ответ не через свой сокет API — они забирают его по URL. Это порт, на котором слушает этот небольшой HTTP-сервер.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Пусто = tts-1. Также: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Пусто = whisper-1. Также: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Пусто = whisper-1. Также: gpt-4o-transcribe, gpt-4o-mini-transcribe. Со своим сервером: модель так, как её называет сервер, напр. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Включить голосовую связь (STT/TTS для спутниковой связи)",
   "Enable voice server (satellites)": "Включить голосовой сервер (спутники)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Включает распознавание/синтез речи. Спутники, использующие ioBroker, получают к нему доступ через шину сообщений — порт не требуется. Включите UDP-сервер, указанный ниже, только для спутников ESP/Hannah.",
@@ -222,5 +222,14 @@
   "Person": "Человек",
   "Guest": "Гость",
   "Pet": "Питомец",
-  "Group": "Группа"
+  "Group": "Группа",
+  "Speech recognition server (optional)": "Сервер распознавания речи (опционально)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "OpenAI-совместимый сервер вместо OpenAI, напр. локальный Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Пусто = OpenAI.",
+  "Key of the speech recognition server": "Ключ сервера распознавания речи",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Только если сервер его требует. Ключ OpenAI туда никогда не отправляется.",
+  "Path to ffmpeg": "Путь к ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Пусто = загруженный ffmpeg, иначе ffmpeg из пути поиска (PATH). Нужен для звуков таймера/будильника, голосовых сообщений Telegram и ответов в MP3/Opus. Пример: /usr/bin/ffmpeg или C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "Проверить ffmpeg",
+  "Download ffmpeg": "Загрузить ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Загружает статический ffmpeg (ffmpeg-static, 20–30 МБ) в каталог данных экземпляра. Используется, если путь не указан."
 }

package/admin/i18n/uk.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "Сателіти ESPHome",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "Сателіти ESPHome отримують відповідь не через свій сокет API — вони забирають її за URL. Це порт, на якому слухає цей невеликий HTTP-сервер.",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "Порожнє = tts-1. Також: tts-1-hd, gpt-4o-mini-tts.",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "Порожнє = шепіт-1. Також: gpt-4o-trancribe, gpt-4o-mini-trancribe.",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "Порожньо = whisper-1. Також: gpt-4o-transcribe, gpt-4o-mini-transcribe. З власним сервером: модель так, як її називає сервер, напр. Systran/faster-whisper-large-v3-turbo.",
   "Enable voice (STT/TTS for satellites)": "Увімкнути голосовий зв'язок (STT/TTS для супутників)",
   "Enable voice server (satellites)": "Увімкнути голосовий сервер (супутники)",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "Дозволяє розпізнавання/синтез мовлення. Супутники, сумісні з ioBroker, досягають його через шину повідомлень — порт не потрібен. Увімкніть UDP-сервер нижче лише для супутників ESP / Hannah.",
@@ -222,5 +222,14 @@
   "Person": "Людина",
   "Guest": "Гість",
   "Pet": "Улюбленець",
-  "Group": "Група"
+  "Group": "Група",
+  "Speech recognition server (optional)": "Сервер розпізнавання мовлення (необов'язково)",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "OpenAI-сумісний сервер замість OpenAI, напр. локальний Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Порожньо = OpenAI.",
+  "Key of the speech recognition server": "Ключ сервера розпізнавання мовлення",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "Лише якщо сервер його вимагає. Ключ OpenAI туди ніколи не надсилається.",
+  "Path to ffmpeg": "Шлях до ffmpeg",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "Порожньо = завантажений ffmpeg, інакше ffmpeg зі шляху пошуку (PATH). Потрібен для звуків таймера/будильника, голосових повідомлень Telegram і відповідей у MP3/Opus. Приклад: /usr/bin/ffmpeg або C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "Перевірити ffmpeg",
+  "Download ffmpeg": "Завантажити ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "Завантажує статичний ffmpeg (ffmpeg-static, 20–30 МБ) до каталогу даних екземпляра. Використовується, якщо шлях не вказано."
 }

package/admin/i18n/zh-cn.json

@@ -58,7 +58,7 @@
   "ESPHome satellites": "ESPHome 卫星",
   "ESPHome satellites do not get the reply over their API socket — they fetch it as a URL. This is the port that small HTTP server listens on.": "ESPHome 卫星不会通过其 API 套接字接收回复，而是按 URL 获取。这里就是那个小型 HTTP 服务器监听的端口。",
   "Empty = tts-1. Also: tts-1-hd, gpt-4o-mini-tts.": "空 = tts-1。另见：tts-1-hd、gpt-4o-mini-tts。",
-  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.": "空 = whisper-1。另见：gpt-4o-transcribe、gpt-4o-mini-transcribe。",
+  "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.": "空 = whisper-1。另见：gpt-4o-transcribe、gpt-4o-mini-transcribe。使用自己的服务器时：按服务器的名称填写模型，例如 Systran/faster-whisper-large-v3-turbo。",
   "Enable voice (STT/TTS for satellites)": "启用语音功能（卫星语音转语音/文本转语音）",
   "Enable voice server (satellites)": "启用语音服务器（卫星）",
   "Enables speech recognition/synthesis. ioBroker-native satellites reach it over the message bus — no port needed. Turn on the UDP server below only for ESP / Hannah satellites.": "启用语音识别/合成功能。ioBroker 原生卫星通过消息总线与其通信，无需端口。仅当 ESP/Hannah 卫星连接时，才需要启用下方的 UDP 服务器。",
@@ -222,5 +222,14 @@
   "Person": "人员",
   "Guest": "访客",
   "Pet": "宠物",
-  "Group": "组"
+  "Group": "组",
+  "Speech recognition server (optional)": "语音识别服务器（可选）",
+  "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.": "代替 OpenAI 的 OpenAI 兼容服务器，例如本地 Whisper：http://192.168.1.10:8000/v1（speaches、faster-whisper-server、LocalAI）。空 = OpenAI。",
+  "Key of the speech recognition server": "语音识别服务器的密钥",
+  "Only if the server wants one. The OpenAI key is never sent to it.": "仅当服务器需要时填写。OpenAI 密钥永远不会发送到该服务器。",
+  "Path to ffmpeg": "ffmpeg 路径",
+  "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe": "留空 = 使用已下载的 ffmpeg，否则使用搜索路径中的 ffmpeg。计时器/闹钟提示音、Telegram 语音消息以及 MP3/Opus 回复需要它。示例：/usr/bin/ffmpeg 或 C:\\ffmpeg\\bin\\ffmpeg.exe",
+  "Check ffmpeg": "检查 ffmpeg",
+  "Download ffmpeg": "下载 ffmpeg",
+  "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.": "将静态 ffmpeg（ffmpeg-static，20–30 MB）下载到实例的数据目录。未填写路径时使用它。"
 }

package/admin/jsonConfig.json

@@ -1184,8 +1184,30 @@
                     "newLine": true,
                     "type": "text",
                     "label": "OpenAI STT model (optional)",
-                    "help": "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe.",
-                    "hidden": "!data.voiceEnabled || data.sttProvider !== 'openai'",
+                    "help": "Empty = whisper-1. Also: gpt-4o-transcribe, gpt-4o-mini-transcribe. With an own server: the model as that server names it, e.g. Systran/faster-whisper-large-v3-turbo.",
+                    "hidden": "!data.voiceEnabled || (data.sttProvider !== 'openai' && data.sttFallback !== 'openai')",
+                    "xs": 12,
+                    "sm": 6,
+                    "md": 4,
+                    "lg": 4,
+                    "xl": 4
+                },
+                "sttBaseUrl": {
+                    "type": "text",
+                    "label": "Speech recognition server (optional)",
+                    "help": "OpenAI-compatible server instead of OpenAI, e.g. a local Whisper: http://192.168.1.10:8000/v1 (speaches, faster-whisper-server, LocalAI). Empty = OpenAI.",
+                    "hidden": "!data.voiceEnabled || (data.sttProvider !== 'openai' && data.sttFallback !== 'openai')",
+                    "xs": 12,
+                    "sm": 6,
+                    "md": 4,
+                    "lg": 4,
+                    "xl": 4
+                },
+                "sttServerKey": {
+                    "type": "password",
+                    "label": "Key of the speech recognition server",
+                    "help": "Only if the server wants one. The OpenAI key is never sent to it.",
+                    "hidden": "!data.voiceEnabled || (data.sttProvider !== 'openai' && data.sttFallback !== 'openai') || !data.sttBaseUrl",
                     "xs": 12,
                     "sm": 6,
                     "md": 4,
@@ -1488,6 +1510,47 @@
                     "lg": 5,
                     "xl": 5
                 },
+                "ffmpegPath": {
+                    "newLine": true,
+                    "type": "text",
+                    "label": "Path to ffmpeg",
+                    "help": "Empty = the downloaded ffmpeg, otherwise ffmpeg from the search path. Needed for timer/alarm sounds, Telegram voice messages and MP3/Opus replies. Example: /usr/bin/ffmpeg or C:\\ffmpeg\\bin\\ffmpeg.exe",
+                    "default": "",
+                    "hidden": "!data.voiceEnabled",
+                    "xs": 12,
+                    "sm": 8,
+                    "md": 5,
+                    "lg": 5,
+                    "xl": 5
+                },
+                "_testFfmpeg": {
+                    "type": "sendTo",
+                    "command": "testFfmpeg",
+                    "label": "Check ffmpeg",
+                    "variant": "outlined",
+                    "jsonData": "{\"path\":\"${data.ffmpegPath}\"}",
+                    "hidden": "!data.voiceEnabled",
+                    "xs": 12,
+                    "sm": 4,
+                    "md": 3,
+                    "lg": 3,
+                    "xl": 3
+                },
+                "_downloadFfmpeg": {
+                    "type": "sendTo",
+                    "command": "downloadFfmpeg",
+                    "label": "Download ffmpeg",
+                    "title": "Downloads a static ffmpeg (ffmpeg-static, 20-30 MB) into the data directory of the instance. It is used when no path is entered.",
+                    "variant": "outlined",
+                    "showProcess": true,
+                    "jsonData": "{\"path\":\"${data.ffmpegPath}\"}",
+                    "hidden": "!data.voiceEnabled",
+                    "xs": 12,
+                    "sm": 4,
+                    "md": 3,
+                    "lg": 3,
+                    "xl": 3
+                },
                 "_voiceTest": {
                     "newLine": true,
                     "type": "custom",

package/build/lib/ask.js

@@ -14,10 +14,20 @@
  */
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.PendingQuestions = exports.ASK_TIMEOUT_MS = exports.ANY_SOURCE = void 0;
+exports.endsWithQuestion = endsWithQuestion;
 /** Key for a question that any source may answer (a broadcast question). */
 exports.ANY_SOURCE = '*';
 /** Default time to wait for an answer before giving up. */
 exports.ASK_TIMEOUT_MS = 60_000;
+/**
+ * Whether a reply ends with a question, so a satellite should re-open its mic for the answer. Only the end counts:
+ * "Wie spät es ist? Es ist zwölf." waits for nothing, and a mic opened for nothing hears the room and ends with
+ * "No text recognized" (field report 2026-11-04). A model that asks in the middle ("… setzen? Sag ja oder nein.")
+ * marks its reply with [[LISTEN]] instead — pi-agent does that for its confirmations.
+ */
+function endsWithQuestion(reply) {
+    return /[?？؟][\s"'»«“”„)\]]*$/u.test(reply || '');
+}
 class PendingQuestions {
     defaultTimeoutMs;
     /** Every armed key points at its entry; several keys can share one (a question asked in two rooms). */

package/build/lib/ask.js.map

@@ -1 +1 @@
-{"version":3,"file":"ask.js","sourceRoot":"","sources":["../../src/lib/ask.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;GAYG;;;AAEH,4EAA4E;AAC/D,QAAA,UAAU,GAAG,GAAG,CAAC;AAE9B,2DAA2D;AAC9C,QAAA,cAAc,GAAG,MAAM,CAAC;AAoBrC,MAAa,gBAAgB;IAII;IAH7B,uGAAuG;IACtF,KAAK,GAAG,IAAI,GAAG,EAAiB,CAAC;IAElD,YAA6B,mBAAmB,sBAAc;QAAjC,qBAAgB,GAAhB,gBAAgB,CAAiB;IAAG,CAAC;IAElE;;;;OAIG;IACH,GAAG,CAAC,IAAc,EAAE,QAAgB,EAAE,SAAkB;QACpD,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACnF,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YACjB,MAAM,CAAC,IAAI,CAAC,kBAAU,CAAC,CAAC;QAC5B,CAAC;QACD,KAAK,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;YACvB,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACrB,CAAC;QACD,OAAO,IAAI,OAAO,CAAgB,OAAO,CAAC,EAAE;YACxC,MAAM,KAAK,GAAU;gBACjB,IAAI,EAAE,MAAM;gBACZ,QAAQ;gBACR,KAAK,EAAE,IAAI,CAAC,GAAG,EAAE;gBACjB,KAAK,EAAE,IAAI;gBACX,IAAI,EAAE,KAAK;gBACX,MAAM,EAAE,OAAO;aAClB,CAAC;YACF,8FAA8F;YAC9F,8FAA8F;YAC9F,KAAK,CAAC,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,SAAS,IAAI,IAAI,CAAC,gBAAgB,CAAC,CAAC;YAC7F,KAAK,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;gBACvB,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YAC/B,CAAC;QACL,CAAC,CAAC,CAAC;IACP,CAAC;IAED;;;;OAIG;IACH,OAAO,CAAC,MAAc,EAAE,IAAY;QAChC,IAAI,CAAC,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;YAC7B,OAAO,KAAK,CAAC;QACjB,CAAC;QACD,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAChC,IAAI,CAAC,KAAK,EAAE,CAAC;YACT,OAAO,KAAK,CAAC;QACjB,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QACzB,OAAO,IAAI,CAAC;IAChB,CAAC;IAED,4EAA4E;IAC5E,GAAG,CAAC,MAAc;QACd,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC/B,CAAC;IAED,2EAA2E;IAC3E,QAAQ,CAAC,MAAc;QACnB,OAAO,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,CAAC;IAED,yEAAyE;IACzE,MAAM,CAAC,GAAW;QACd,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAClC,IAAI,CAAC,KAAK,EAAE,CAAC;YACT,OAAO,KAAK,CAAC;QACjB,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QACzB,OAAO,IAAI,CAAC;IAChB,CAAC;IAED,gFAAgF;IAChF,SAAS;QACL,KAAK,MAAM,KAAK,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE,CAAC;YACpD,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QAC7B,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;IACvB,CAAC;IAED,gEAAgE;IAChE,IAAI;QACA,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAC/C,QAAQ,EAAE,CAAC,CAAC,QAAQ;YACpB,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC;YACjB,KAAK,EAAE,CAAC,CAAC,KAAK;SACjB,CAAC,CAAC,CAAC;IACR,CAAC;IAED,wFAAwF;IAChF,IAAI,CAAC,MAAc;QACvB,MAAM,GAAG,GAAG,MAAM,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,kBAAU,CAAC,CAAC;IACjF,CAAC;IAEO,MAAM,CAAC,KAAY,EAAE,MAAqB;QAC9C,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;YACb,OAAO;QACX,CAAC;QACD,KAAK,CAAC,IAAI,GAAG,IAAI,CAAC;QAClB,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;YACd,YAAY,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;YAC1B,KAAK,CAAC,KAAK,GAAG,IAAI,CAAC;QACvB,CAAC;QACD,KAAK,MAAM,GAAG,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;YAC3B,yFAAyF;YACzF,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,KAAK,EAAE,CAAC;gBAChC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC3B,CAAC;QACL,CAAC;QACD,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IACzB,CAAC;CACJ;AAlHD,4CAkHC","sourcesContent":["/**\n * Pending questions — the other direction of a conversation: the adapter asks and waits for the answer.\n *\n * Everywhere else the user starts the exchange. Here a script, a timer or (later) a trigger asks\n * something (\"The fryer has been on for 5 hours. Shall I switch it off?\") and the *next* utterance from\n * that source is the answer: it must not go to the NLU or the LLM, which would treat a bare \"yes\" as a\n * command and swallow it.\n *\n * A question is armed for one or more **sources** — the same key the answer pipeline sees: a satellite's\n * device name, `'chat'`, `'telegram:<user>'`, … — or for {@link ANY_SOURCE} when it was broadcast and\n * whoever hears it may answer. If nothing arrives within the timeout the promise resolves with `null`,\n * so a question can never block the pipeline for good.\n */\n\n/** Key for a question that any source may answer (a broadcast question). */\nexport const ANY_SOURCE = '*';\n\n/** Default time to wait for an answer before giving up. */\nexport const ASK_TIMEOUT_MS = 60_000;\n\ninterface Entry {\n    keys: string[];\n    question: string;\n    asked: number;\n    timer: NodeJS.Timeout | null;\n    done: boolean;\n    settle: (answer: string | null) => void;\n}\n\n/** One open question, for logging and diagnostics. */\nexport interface PendingInfo {\n    question: string;\n    /** Sources that may answer it (`'*'` = any). */\n    keys: string[];\n    /** When it was asked (ms epoch). */\n    asked: number;\n}\n\nexport class PendingQuestions {\n    /** Every armed key points at its entry; several keys can share one (a question asked in two rooms). */\n    private readonly byKey = new Map<string, Entry>();\n\n    constructor(private readonly defaultTimeoutMs = ASK_TIMEOUT_MS) {}\n\n    /**\n     * Arm `question` for `keys` (empty → {@link ANY_SOURCE}) and resolve with the first answer from one of\n     * them, or with `null` on timeout. Arming a key that already holds a question cancels that older one:\n     * the newest question is the one the user just heard, so it has to be the one that wins.\n     */\n    ask(keys: string[], question: string, timeoutMs?: number): Promise<string | null> {\n        const unique = [...new Set(keys.map(k => String(k ?? '').trim()).filter(Boolean))];\n        if (!unique.length) {\n            unique.push(ANY_SOURCE);\n        }\n        for (const key of unique) {\n            this.cancel(key);\n        }\n        return new Promise<string | null>(resolve => {\n            const entry: Entry = {\n                keys: unique,\n                question,\n                asked: Date.now(),\n                timer: null,\n                done: false,\n                settle: resolve,\n            };\n            // Deliberately not unref'd: the timeout is what releases the caller, so it has to fire. It is\n            // cleared on an answer, on cancel, and by cancelAll() on unload, so nothing is held for long.\n            entry.timer = setTimeout(() => this.finish(entry, null), timeoutMs ?? this.defaultTimeoutMs);\n            for (const key of unique) {\n                this.byKey.set(key, entry);\n            }\n        });\n    }\n\n    /**\n     * Consume an utterance as the answer to an open question. Returns true when it was taken — the caller\n     * must then stop processing it. Blank text is never an answer (a dropped utterance would otherwise\n     * silently satisfy the question).\n     */\n    deliver(source: string, text: string): boolean {\n        if (!String(text ?? '').trim()) {\n            return false;\n        }\n        const entry = this.find(source);\n        if (!entry) {\n            return false;\n        }\n        this.finish(entry, text);\n        return true;\n    }\n\n    /** Whether an open question would take the next utterance from `source`. */\n    has(source: string): boolean {\n        return !!this.find(source);\n    }\n\n    /** The open question `source` is expected to answer, if any (for logs). */\n    question(source: string): string {\n        return this.find(source)?.question ?? '';\n    }\n\n    /** Give up on the question armed for `key` (resolves it with `null`). */\n    cancel(key: string): boolean {\n        const entry = this.byKey.get(key);\n        if (!entry) {\n            return false;\n        }\n        this.finish(entry, null);\n        return true;\n    }\n\n    /** Give up on every open question — on unload, so no caller is left hanging. */\n    cancelAll(): void {\n        for (const entry of [...new Set(this.byKey.values())]) {\n            this.finish(entry, null);\n        }\n        this.byKey.clear();\n    }\n\n    /** All open questions (one entry per question, not per key). */\n    list(): PendingInfo[] {\n        return [...new Set(this.byKey.values())].map(e => ({\n            question: e.question,\n            keys: [...e.keys],\n            asked: e.asked,\n        }));\n    }\n\n    /** The entry responsible for `source`: armed for it by name, else the broadcast one. */\n    private find(source: string): Entry | undefined {\n        const key = String(source ?? '').trim();\n        return (key ? this.byKey.get(key) : undefined) ?? this.byKey.get(ANY_SOURCE);\n    }\n\n    private finish(entry: Entry, answer: string | null): void {\n        if (entry.done) {\n            return;\n        }\n        entry.done = true;\n        if (entry.timer) {\n            clearTimeout(entry.timer);\n            entry.timer = null;\n        }\n        for (const key of entry.keys) {\n            // Only drop keys still pointing here — a newer question may already have taken one over.\n            if (this.byKey.get(key) === entry) {\n                this.byKey.delete(key);\n            }\n        }\n        entry.settle(answer);\n    }\n}\n"]}
+{"version":3,"file":"ask.js","sourceRoot":"","sources":["../../src/lib/ask.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;GAYG;;;AAcH,4CAEC;AAdD,4EAA4E;AAC/D,QAAA,UAAU,GAAG,GAAG,CAAC;AAE9B,2DAA2D;AAC9C,QAAA,cAAc,GAAG,MAAM,CAAC;AAErC;;;;;GAKG;AACH,SAAgB,gBAAgB,CAAC,KAAa;IAC1C,OAAO,wBAAwB,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;AACtD,CAAC;AAoBD,MAAa,gBAAgB;IAII;IAH7B,uGAAuG;IACtF,KAAK,GAAG,IAAI,GAAG,EAAiB,CAAC;IAElD,YAA6B,mBAAmB,sBAAc;QAAjC,qBAAgB,GAAhB,gBAAgB,CAAiB;IAAG,CAAC;IAElE;;;;OAIG;IACH,GAAG,CAAC,IAAc,EAAE,QAAgB,EAAE,SAAkB;QACpD,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACnF,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YACjB,MAAM,CAAC,IAAI,CAAC,kBAAU,CAAC,CAAC;QAC5B,CAAC;QACD,KAAK,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;YACvB,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACrB,CAAC;QACD,OAAO,IAAI,OAAO,CAAgB,OAAO,CAAC,EAAE;YACxC,MAAM,KAAK,GAAU;gBACjB,IAAI,EAAE,MAAM;gBACZ,QAAQ;gBACR,KAAK,EAAE,IAAI,CAAC,GAAG,EAAE;gBACjB,KAAK,EAAE,IAAI;gBACX,IAAI,EAAE,KAAK;gBACX,MAAM,EAAE,OAAO;aAClB,CAAC;YACF,8FAA8F;YAC9F,8FAA8F;YAC9F,KAAK,CAAC,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,SAAS,IAAI,IAAI,CAAC,gBAAgB,CAAC,CAAC;YAC7F,KAAK,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;gBACvB,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YAC/B,CAAC;QACL,CAAC,CAAC,CAAC;IACP,CAAC;IAED;;;;OAIG;IACH,OAAO,CAAC,MAAc,EAAE,IAAY;QAChC,IAAI,CAAC,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;YAC7B,OAAO,KAAK,CAAC;QACjB,CAAC;QACD,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAChC,IAAI,CAAC,KAAK,EAAE,CAAC;YACT,OAAO,KAAK,CAAC;QACjB,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QACzB,OAAO,IAAI,CAAC;IAChB,CAAC;IAED,4EAA4E;IAC5E,GAAG,CAAC,MAAc;QACd,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC/B,CAAC;IAED,2EAA2E;IAC3E,QAAQ,CAAC,MAAc;QACnB,OAAO,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,CAAC;IAED,yEAAyE;IACzE,MAAM,CAAC,GAAW;QACd,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAClC,IAAI,CAAC,KAAK,EAAE,CAAC;YACT,OAAO,KAAK,CAAC;QACjB,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QACzB,OAAO,IAAI,CAAC;IAChB,CAAC;IAED,gFAAgF;IAChF,SAAS;QACL,KAAK,MAAM,KAAK,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE,CAAC;YACpD,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QAC7B,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;IACvB,CAAC;IAED,gEAAgE;IAChE,IAAI;QACA,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAC/C,QAAQ,EAAE,CAAC,CAAC,QAAQ;YACpB,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC;YACjB,KAAK,EAAE,CAAC,CAAC,KAAK;SACjB,CAAC,CAAC,CAAC;IACR,CAAC;IAED,wFAAwF;IAChF,IAAI,CAAC,MAAc;QACvB,MAAM,GAAG,GAAG,MAAM,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,kBAAU,CAAC,CAAC;IACjF,CAAC;IAEO,MAAM,CAAC,KAAY,EAAE,MAAqB;QAC9C,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;YACb,OAAO;QACX,CAAC;QACD,KAAK,CAAC,IAAI,GAAG,IAAI,CAAC;QAClB,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;YACd,YAAY,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;YAC1B,KAAK,CAAC,KAAK,GAAG,IAAI,CAAC;QACvB,CAAC;QACD,KAAK,MAAM,GAAG,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;YAC3B,yFAAyF;YACzF,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,KAAK,EAAE,CAAC;gBAChC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC3B,CAAC;QACL,CAAC;QACD,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IACzB,CAAC;CACJ;AAlHD,4CAkHC","sourcesContent":["/**\n * Pending questions — the other direction of a conversation: the adapter asks and waits for the answer.\n *\n * Everywhere else the user starts the exchange. Here a script, a timer or (later) a trigger asks\n * something (\"The fryer has been on for 5 hours. Shall I switch it off?\") and the *next* utterance from\n * that source is the answer: it must not go to the NLU or the LLM, which would treat a bare \"yes\" as a\n * command and swallow it.\n *\n * A question is armed for one or more **sources** — the same key the answer pipeline sees: a satellite's\n * device name, `'chat'`, `'telegram:<user>'`, … — or for {@link ANY_SOURCE} when it was broadcast and\n * whoever hears it may answer. If nothing arrives within the timeout the promise resolves with `null`,\n * so a question can never block the pipeline for good.\n */\n\n/** Key for a question that any source may answer (a broadcast question). */\nexport const ANY_SOURCE = '*';\n\n/** Default time to wait for an answer before giving up. */\nexport const ASK_TIMEOUT_MS = 60_000;\n\n/**\n * Whether a reply ends with a question, so a satellite should re-open its mic for the answer. Only the end counts:\n * \"Wie spät es ist? Es ist zwölf.\" waits for nothing, and a mic opened for nothing hears the room and ends with\n * \"No text recognized\" (field report 2026-11-04). A model that asks in the middle (\"… setzen? Sag ja oder nein.\")\n * marks its reply with [[LISTEN]] instead — pi-agent does that for its confirmations.\n */\nexport function endsWithQuestion(reply: string): boolean {\n    return /[?？؟][\\s\"'»«“”„)\\]]*$/u.test(reply || '');\n}\n\ninterface Entry {\n    keys: string[];\n    question: string;\n    asked: number;\n    timer: NodeJS.Timeout | null;\n    done: boolean;\n    settle: (answer: string | null) => void;\n}\n\n/** One open question, for logging and diagnostics. */\nexport interface PendingInfo {\n    question: string;\n    /** Sources that may answer it (`'*'` = any). */\n    keys: string[];\n    /** When it was asked (ms epoch). */\n    asked: number;\n}\n\nexport class PendingQuestions {\n    /** Every armed key points at its entry; several keys can share one (a question asked in two rooms). */\n    private readonly byKey = new Map<string, Entry>();\n\n    constructor(private readonly defaultTimeoutMs = ASK_TIMEOUT_MS) {}\n\n    /**\n     * Arm `question` for `keys` (empty → {@link ANY_SOURCE}) and resolve with the first answer from one of\n     * them, or with `null` on timeout. Arming a key that already holds a question cancels that older one:\n     * the newest question is the one the user just heard, so it has to be the one that wins.\n     */\n    ask(keys: string[], question: string, timeoutMs?: number): Promise<string | null> {\n        const unique = [...new Set(keys.map(k => String(k ?? '').trim()).filter(Boolean))];\n        if (!unique.length) {\n            unique.push(ANY_SOURCE);\n        }\n        for (const key of unique) {\n            this.cancel(key);\n        }\n        return new Promise<string | null>(resolve => {\n            const entry: Entry = {\n                keys: unique,\n                question,\n                asked: Date.now(),\n                timer: null,\n                done: false,\n                settle: resolve,\n            };\n            // Deliberately not unref'd: the timeout is what releases the caller, so it has to fire. It is\n            // cleared on an answer, on cancel, and by cancelAll() on unload, so nothing is held for long.\n            entry.timer = setTimeout(() => this.finish(entry, null), timeoutMs ?? this.defaultTimeoutMs);\n            for (const key of unique) {\n                this.byKey.set(key, entry);\n            }\n        });\n    }\n\n    /**\n     * Consume an utterance as the answer to an open question. Returns true when it was taken — the caller\n     * must then stop processing it. Blank text is never an answer (a dropped utterance would otherwise\n     * silently satisfy the question).\n     */\n    deliver(source: string, text: string): boolean {\n        if (!String(text ?? '').trim()) {\n            return false;\n        }\n        const entry = this.find(source);\n        if (!entry) {\n            return false;\n        }\n        this.finish(entry, text);\n        return true;\n    }\n\n    /** Whether an open question would take the next utterance from `source`. */\n    has(source: string): boolean {\n        return !!this.find(source);\n    }\n\n    /** The open question `source` is expected to answer, if any (for logs). */\n    question(source: string): string {\n        return this.find(source)?.question ?? '';\n    }\n\n    /** Give up on the question armed for `key` (resolves it with `null`). */\n    cancel(key: string): boolean {\n        const entry = this.byKey.get(key);\n        if (!entry) {\n            return false;\n        }\n        this.finish(entry, null);\n        return true;\n    }\n\n    /** Give up on every open question — on unload, so no caller is left hanging. */\n    cancelAll(): void {\n        for (const entry of [...new Set(this.byKey.values())]) {\n            this.finish(entry, null);\n        }\n        this.byKey.clear();\n    }\n\n    /** All open questions (one entry per question, not per key). */\n    list(): PendingInfo[] {\n        return [...new Set(this.byKey.values())].map(e => ({\n            question: e.question,\n            keys: [...e.keys],\n            asked: e.asked,\n        }));\n    }\n\n    /** The entry responsible for `source`: armed for it by name, else the broadcast one. */\n    private find(source: string): Entry | undefined {\n        const key = String(source ?? '').trim();\n        return (key ? this.byKey.get(key) : undefined) ?? this.byKey.get(ANY_SOURCE);\n    }\n\n    private finish(entry: Entry, answer: string | null): void {\n        if (entry.done) {\n            return;\n        }\n        entry.done = true;\n        if (entry.timer) {\n            clearTimeout(entry.timer);\n            entry.timer = null;\n        }\n        for (const key of entry.keys) {\n            // Only drop keys still pointing here — a newer question may already have taken one over.\n            if (this.byKey.get(key) === entry) {\n                this.byKey.delete(key);\n            }\n        }\n        entry.settle(answer);\n    }\n}\n"]}