@wildix/wilma-voice-client 0.0.51 → 0.0.53

package/dist-types/models/models_0.d.ts

@@ -1,5 +1,5 @@
 import type { DocumentType as __DocumentType, StreamingBlobTypes } from "@smithy/types";
-import type { MediaAudioBeginFrameType, MediaAudioEncoding, MediaAudioEndFrameType, MediaDtmfFrameType, MediaDuckFrameType, MediaErrorFrameType, MediaFinishedFrameType, MediaFinishFrameType, MediaFinishReason, MediaHangupFrameType, MediaLanguageFrameType, MediaParty, MediaPlaybackEvent, MediaPlaybackFrameType, MediaPlaybackStopCause, MediaRestoreFrameType, MediaRunPurpose, MediaStartedFrameType, MediaStartFrameType, MediaStopFrameType, MediaTranscriptionFrameType, MediaTransferFrameType, MediaVadFrameType, VoiceAudioEncoding, VoiceCallEndReason, VoiceCallToolPhase, VoiceCommandType, VoiceControlAction, VoiceEventType, VoiceInputMode, VoicePronunciationEncoding, VoiceTranscriptionDropReason } from "./enums";
+import type { VoiceCallClientEndReason, VoiceCallClientFrameType, VoiceCallEndReason, VoiceCallParty, VoiceCallPlaybackPhase, VoiceCallServerFrameType, VoiceCommandType, VoiceEventType, VoiceInputMode, VoicePronunciationEncoding, VoiceToolPhase, VoiceTranscriptionDropReason } from "./enums";
 /**
  * A pronunciation rule for one phrase. Belongs to a LANGUAGE, not to the agent: 'Wildix' is
  *     respelled differently for an Italian and a German voice, so one list shared ACROSS languages
@@ -25,968 +25,656 @@ export interface VoicePronunciation {
     encoding?: VoicePronunciationEncoding | undefined;
 }
 /**
- * A recogniser result for the agent's own leg — what the caller hears. Sent by the PBX for new-stack agents only, from the same dual-channel stream that transcribes the caller. Never starts a turn.
+ * Caller voice detection on the PBX.
  * @public
  */
-export interface AgentTranscriptionEvent {
+export interface VoiceDetectionSettings {
     /**
-     * Discriminator for the inner payload of an `event` frame.
+     * Silence in ms that ends the caller speech. Absent: 500.
      * @public
      */
-    type: VoiceEventType;
-    /**
-     * Identifier of a transcription chunk within a session.
-     * @public
-     */
-    id: string;
-    text: string;
+    silenceMs?: number | undefined;
     /**
-     * Transcription confidence score (accuracy level) in range from 0.0 to 1.0. Value -1.0 indicates confidence is unavailable.
+     * Mean frame level in dBFS below which caller audio is silence. Absent: -36.
      * @public
      */
-    confidence?: number | undefined;
+    silenceLevelDb?: number | undefined;
     /**
-     * True when this is the final transcript for the utterance rather than an interim hypothesis. An absent value is read as final, as on `TranscriptionEvent`.
+     * Mean frame level in dBFS from which caller audio is speech. Absent: -30.
      * @public
      */
-    final?: boolean | undefined;
-    /**
-     * Language the recogniser identified for this chunk. Recorded in the trace only — it drives no behavior.
-     * @public
-     */
-    language?: string | undefined;
-    time: number;
+    speechLevelDb?: number | undefined;
 }
 /**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+ * A transcription chunk of the agent speech, as the caller hears it. It does not start a turn.
  * @public
  */
-export interface AgentTranscriptionWebsocketEvent {
-    type: string;
+export interface VoiceAgentTranscriptionEvent {
     /**
-     * A recogniser result for the agent's own leg — what the caller hears. Sent by the PBX for new-stack agents only, from the same dual-channel stream that transcribes the caller. Never starts a turn.
-     * @public
-     */
-    event: AgentTranscriptionEvent;
-}
-/**
- * A chunk of raw caller audio, base64-encoded in the session's negotiated input format
- *     (inputAudio from the connect handshake). Used by web/app clients in AUDIO input mode; the
- *     service buffers and transcribes it server-side. Not used in TRANSCRIPTION mode.
- * @public
- */
-export interface AudioEvent {
-    /**
-     * Discriminator for the inner payload of an `event` frame.
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
     type: VoiceEventType;
-    data: string;
     /**
-     * Optional monotonic chunk sequence number, for gap detection.
+     * Id of the utterance. Absent: the service makes a new id.
      * @public
      */
-    seq?: number | undefined;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface AudioWebsocketEvent {
-    type: string;
+    id?: string | undefined;
     /**
-     * A chunk of raw caller audio, base64-encoded in the session's negotiated input format
-     *     (inputAudio from the connect handshake). Used by web/app clients in AUDIO input mode; the
-     *     service buffers and transcribes it server-side. Not used in TRANSCRIPTION mode.
+     * Text of the chunk. Must not be empty.
      * @public
      */
-    event: AudioEvent;
-}
-/**
- * The call/conversation completed (e.g. the caller hung up or the call was transferred away).
- * @public
- */
-export interface CompleteEvent {
-    /**
-     * Discriminator for the inner payload of an `event` frame.
-     * @public
-     */
-    type: VoiceEventType;
-    reason: string;
-    time: number;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface CompleteWebsocketEvent {
-    type: string;
-    /**
-     * The call/conversation completed (e.g. the caller hung up or the call was transferred away).
-     * @public
-     */
-    event: CompleteEvent;
-}
-/**
- * A lightweight control frame from the client (AUDIO mode).
- * @public
- */
-export interface ControlEvent {
-    /**
-     * Discriminator for the inner payload of an `event` frame.
-     * @public
-     */
-    type: VoiceEventType;
-    /**
-     * A client control action on the audio input (AUDIO mode).
-     * @public
-     */
-    action: VoiceControlAction;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface ControlWebsocketEvent {
-    type: string;
+    text: string;
     /**
-     * A lightweight control frame from the client (AUDIO mode).
+     * Confidence of the recogniser. The service does not record it.
      * @public
      */
-    event: ControlEvent;
-}
-/**
- * @public
- */
-export interface DescribeVoiceProtocolInput {
-}
-/**
- * A recogniser result for the AGENT's own leg — what the caller actually heard. Same semantics as VoiceCallTranscriptionEvent; an interrupted reply simply stops.
- * @public
- */
-export interface VoiceCallAgentTranscriptionEvent {
+    confidence?: number | undefined;
     /**
-     * Identifier of a transcription chunk within a session.
+     * True when the chunk is the final transcript of the utterance. False for an interim result. Absent: true.
      * @public
      */
-    id: string;
-    text: string;
-    final: boolean;
+    final?: boolean | undefined;
     /**
-     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
-     *     constrained string rather than an enum: the offerable set is a property of the selected
-     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
-     *     when the model was released.
+     * Language that the recogniser identified for the chunk. The service records it and does not change the conversation language.
      * @public
      */
     language?: string | undefined;
-    time?: number | undefined;
-}
-/**
- * The session ended. After `transfer` another session of the same call may follow with its own `started`.
- * @public
- */
-export interface VoiceCallEndedEvent {
-    /**
-     * Caller-facing voice session identifier, unique per connection. Supplied by the client at connect (the PBX call/session id, or a client-generated id for web sessions).
-     * @public
-     */
-    sessionId: string;
-    reason: VoiceCallEndReason;
-}
-/**
- * The caller talked over the agent. `playedText` is what they heard before the cut.
- * @public
- */
-export interface VoiceCallInterruptionEvent {
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * Time of the chunk, in epoch ms. Absent: the time when the service received it.
      * @public
      */
-    replyId?: string | undefined;
-    playedText?: string | undefined;
-    originalText?: string | undefined;
+    time?: number | undefined;
 }
 /**
- * The conversation language changed on a recognised utterance.
+ * The call or the conversation is complete, for example because the caller hung up. The service ends the session. The service also accepts `type` `ended` with the same meaning.
  * @public
  */
-export interface VoiceCallLanguageEvent {
+export interface VoiceCompleteEvent {
     /**
-     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
-     *     constrained string rather than an enum: the offerable set is a property of the selected
-     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
-     *     when the model was released.
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
-    from?: string | undefined;
+    type: VoiceEventType;
     /**
-     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
-     *     constrained string rather than an enum: the offerable set is a property of the selected
-     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
-     *     when the model was released.
+     * Why the call is complete. Absent: the value of `type`.
      * @public
      */
-    to: string;
-    reason: string;
-}
-/**
- * The PBX finished (or abandoned) playing a reply.
- * @public
- */
-export interface VoiceCallPlaybackEvent {
+    reason?: string | undefined;
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * Time of the completion, in epoch ms. The service ignores it.
      * @public
      */
-    replyId: string;
-    complete: boolean;
+    time?: number | undefined;
 }
 /**
- * The caller's completed utterance that started a turn — after the runtime's filler and duplicate filter.
+ * End the session and hang up the call.
  * @public
  */
-export interface VoiceCallReplyEvent {
+export interface VoiceHangupCommand {
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * Why the session ended.
      * @public
      */
-    id: string;
-    text: string;
+    reason?: string | undefined;
     /**
-     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
-     *     constrained string rather than an enum: the offerable set is a property of the selected
-     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
-     *     when the model was released.
+     * Value of `type` in the `command` of a command frame.
      * @public
      */
-    language?: string | undefined;
+    type: VoiceCommandType;
 }
 /**
- * One sentence the agent speaks, as it is handed to the PBX (after playback sanitizing). Sentences of one reply share `replyId`.
+ * Liveness check of the service by the client: `\{"type": "healthcheck"\}`, with no envelope. The service sends no answer.
  * @public
  */
-export interface VoiceCallSayEvent {
-    /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
-     * @public
-     */
-    replyId: string;
-    text: string;
-    interruptible: boolean;
+export interface VoiceHealthcheckCommand {
+    type: string;
 }
 /**
- * The session that answered the call, sent once per session of the call (a transfer starts another).
+ * The caller spoke over a reply while the client played it.
  * @public
  */
-export interface VoiceCallStartedEvent {
+export interface VoiceInterruptionEvent {
     /**
-     * Caller-facing voice session identifier, unique per connection. Supplied by the client at connect (the PBX call/session id, or a client-generated id for web sessions).
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
-    sessionId: string;
+    type: VoiceEventType;
     /**
-     * The session's trace, for the LLM-level detail this stream deliberately omits. Absent for a built-in agent, whose trace is not the tenant's to read.
+     * Reply that the caller interrupted.
      * @public
      */
-    traceId?: string | undefined;
-    agentId: string;
-    agentName?: string | undefined;
+    replyId?: string | undefined;
     /**
-     * Published agent version number. Monotonic within an agent's version history, starting at 1.
+     * Part of the reply that the client played before the interruption.
      * @public
      */
-    agentVersion?: number | undefined;
+    playedText?: string | undefined;
     /**
-     * True when the session runs the agent's draft.
+     * Full text of the reply.
      * @public
      */
-    agentDraft?: boolean | undefined;
+    originalText?: string | undefined;
     /**
-     * BCP-47 language the call started in.
+     * Time of the interruption, in epoch ms on the client clock. The service ignores 0.
      * @public
      */
-    language?: string | undefined;
+    time?: number | undefined;
 }
 /**
- * A tool call the agent made during a turn. Name and phase only — arguments and results live in the trace. Sent to a caller signed in to the system; never to a guest.
+ * Progress of the playback of a reply on the client.
  * @public
  */
-export interface VoiceCallToolEvent {
+export interface VoicePlaybackEvent {
     /**
-     * Identifies the tool call inside the turn — the model's own call id, not the PBX call this stream follows.
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
-    toolCallId: string;
-    toolName: string;
-    phase: VoiceCallToolPhase;
+    type: VoiceEventType;
     /**
-     * The phrase spoken while the tool runs, when the capability configures one.
+     * Reply that the client plays.
      * @public
      */
-    startMessage?: string | undefined;
-}
-/**
- * A recogniser result for the caller's speech. Interim results (`final: false`) are replaced by the final one with the same `id`.
- * @public
- */
-export interface VoiceCallTranscriptionEvent {
+    replyId?: string | undefined;
     /**
-     * Identifier of a transcription chunk within a session.
+     * Text of the reply that the client played.
      * @public
      */
-    id: string;
-    text: string;
-    final: boolean;
+    text?: string | undefined;
     /**
-     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
-     *     constrained string rather than an enum: the offerable set is a property of the selected
-     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
-     *     when the model was released.
+     * True when the client played the reply to its end.
      * @public
      */
-    language?: string | undefined;
+    complete: boolean;
     /**
-     * Milliseconds since the call started, as the PBX stamps it.
+     * Start of the playback, in epoch ms on the client clock. The service ignores 0.
      * @public
      */
-    time?: number | undefined;
+    startTime?: number | undefined;
     /**
-     * Present on a final that did not go to the agent as a caller turn. It came while the agent spoke, and its recognition confidence was below the minimum of the call or the recogniser labelled it with another speaker.
+     * End of the playback, in epoch ms on the client clock. The service ignores 0.
      * @public
      */
-    dropped?: VoiceTranscriptionDropReason | undefined;
+    endTime?: number | undefined;
 }
 /**
- * One event on a live call. The SSE frame's `event:` is the member name and `data:` this whole union document, so a frame reads `data: \{"say": \{...\}\}` — never the member's payload on its own.
- * @public
- */
-export type VoiceCallEvent = VoiceCallEvent.AgentTranscriptionMember | VoiceCallEvent.EndedMember | VoiceCallEvent.InterruptionMember | VoiceCallEvent.LanguageMember | VoiceCallEvent.PlaybackMember | VoiceCallEvent.ReplyMember | VoiceCallEvent.SayMember | VoiceCallEvent.StartedMember | VoiceCallEvent.ToolMember | VoiceCallEvent.TranscriptionMember | VoiceCallEvent.$UnknownMember;
-/**
+ * A final caller utterance. It starts an agent turn.
  * @public
  */
-export declare namespace VoiceCallEvent {
-    /**
-     * The session that answered the call, sent once per session of the call (a transfer starts another).
-     * @public
-     */
-    interface StartedMember {
-        started: VoiceCallStartedEvent;
-        transcription?: never;
-        agentTranscription?: never;
-        reply?: never;
-        say?: never;
-        playback?: never;
-        interruption?: never;
-        language?: never;
-        tool?: never;
-        ended?: never;
-        $unknown?: never;
-    }
+export interface VoiceReplyEvent {
     /**
-     * A recogniser result for the caller's speech. Interim results (`final: false`) are replaced by the final one with the same `id`.
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
-    interface TranscriptionMember {
-        started?: never;
-        transcription: VoiceCallTranscriptionEvent;
-        agentTranscription?: never;
-        reply?: never;
-        say?: never;
-        playback?: never;
-        interruption?: never;
-        language?: never;
-        tool?: never;
-        ended?: never;
-        $unknown?: never;
-    }
-    /**
-     * A recogniser result for the AGENT's own leg — what the caller actually heard. Same semantics as VoiceCallTranscriptionEvent; an interrupted reply simply stops.
-     * @public
-     */
-    interface AgentTranscriptionMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription: VoiceCallAgentTranscriptionEvent;
-        reply?: never;
-        say?: never;
-        playback?: never;
-        interruption?: never;
-        language?: never;
-        tool?: never;
-        ended?: never;
-        $unknown?: never;
-    }
-    /**
-     * The caller's completed utterance that started a turn — after the runtime's filler and duplicate filter.
-     * @public
-     */
-    interface ReplyMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription?: never;
-        reply: VoiceCallReplyEvent;
-        say?: never;
-        playback?: never;
-        interruption?: never;
-        language?: never;
-        tool?: never;
-        ended?: never;
-        $unknown?: never;
-    }
-    /**
-     * One sentence the agent speaks, as it is handed to the PBX (after playback sanitizing). Sentences of one reply share `replyId`.
-     * @public
-     */
-    interface SayMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription?: never;
-        reply?: never;
-        say: VoiceCallSayEvent;
-        playback?: never;
-        interruption?: never;
-        language?: never;
-        tool?: never;
-        ended?: never;
-        $unknown?: never;
-    }
-    /**
-     * The PBX finished (or abandoned) playing a reply.
-     * @public
-     */
-    interface PlaybackMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription?: never;
-        reply?: never;
-        say?: never;
-        playback: VoiceCallPlaybackEvent;
-        interruption?: never;
-        language?: never;
-        tool?: never;
-        ended?: never;
-        $unknown?: never;
-    }
-    /**
-     * The caller talked over the agent. `playedText` is what they heard before the cut.
-     * @public
-     */
-    interface InterruptionMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription?: never;
-        reply?: never;
-        say?: never;
-        playback?: never;
-        interruption: VoiceCallInterruptionEvent;
-        language?: never;
-        tool?: never;
-        ended?: never;
-        $unknown?: never;
-    }
+    type: VoiceEventType;
     /**
-     * The conversation language changed on a recognised utterance.
+     * Id of the utterance. Absent: the service makes a new id.
      * @public
      */
-    interface LanguageMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription?: never;
-        reply?: never;
-        say?: never;
-        playback?: never;
-        interruption?: never;
-        language: VoiceCallLanguageEvent;
-        tool?: never;
-        ended?: never;
-        $unknown?: never;
-    }
+    id?: string | undefined;
     /**
-     * A tool call the agent made during a turn. Name and phase only — arguments and results live in the trace. Sent to a caller signed in to the system; never to a guest.
+     * Text of the utterance. Must not be empty.
      * @public
      */
-    interface ToolMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription?: never;
-        reply?: never;
-        say?: never;
-        playback?: never;
-        interruption?: never;
-        language?: never;
-        tool: VoiceCallToolEvent;
-        ended?: never;
-        $unknown?: never;
-    }
+    text: string;
     /**
-     * The session ended. After `transfer` another session of the same call may follow with its own `started`.
+     * Time of the utterance, in epoch ms. Absent: the time when the service received it.
      * @public
      */
-    interface EndedMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription?: never;
-        reply?: never;
-        say?: never;
-        playback?: never;
-        interruption?: never;
-        language?: never;
-        tool?: never;
-        ended: VoiceCallEndedEvent;
-        $unknown?: never;
-    }
+    time?: number | undefined;
     /**
+     * Language that the recogniser identified for the utterance. A value that is different from the conversation language changes the conversation language. When the utterance has segments in more than one language, send the language of the last segment.
      * @public
      */
-    interface $UnknownMember {
-        started?: never;
-        transcription?: never;
-        agentTranscription?: never;
-        reply?: never;
-        say?: never;
-        playback?: never;
-        interruption?: never;
-        language?: never;
-        tool?: never;
-        ended?: never;
-        $unknown: [string, any];
-    }
-    /**
-     * @deprecated unused in schema-serde mode.
-     *
-     */
-    interface Visitor<T> {
-        started: (value: VoiceCallStartedEvent) => T;
-        transcription: (value: VoiceCallTranscriptionEvent) => T;
-        agentTranscription: (value: VoiceCallAgentTranscriptionEvent) => T;
-        reply: (value: VoiceCallReplyEvent) => T;
-        say: (value: VoiceCallSayEvent) => T;
-        playback: (value: VoiceCallPlaybackEvent) => T;
-        interruption: (value: VoiceCallInterruptionEvent) => T;
-        language: (value: VoiceCallLanguageEvent) => T;
-        tool: (value: VoiceCallToolEvent) => T;
-        ended: (value: VoiceCallEndedEvent) => T;
-        _: (name: string, value: any) => T;
-    }
-}
-/**
- * A PCM/compressed audio format description.
- * @public
- */
-export interface VoiceAudioFormat {
+    language?: string | undefined;
     /**
-     * Encoding of raw audio carried over the connection (client `audio` chunks and server `play` frames).
+     * Confidence of the recogniser for the utterance. The service ignores a value outside 0 to 1.
      * @public
      */
-    encoding: VoiceAudioEncoding;
+    confidence?: number | undefined;
     /**
-     * Samples per second, e.g. 8000, 16000 or 24000.
+     * Start of the speech, in epoch ms on the client clock. The service ignores 0.
      * @public
      */
-    sampleRate: number;
+    startTime?: number | undefined;
     /**
-     * Channel count. Defaults to 1 (mono).
+     * End of the speech, in epoch ms on the client clock. The service ignores 0.
      * @public
      */
-    channels?: number | undefined;
+    endTime?: number | undefined;
 }
 /**
- * Parameters supplied on the WebSocket connect (upgrade) request, as query-string parameters.
+ * A transcription chunk of caller speech. A client that does its own speech-to-text sends it. When the service does the speech-to-text, the service sends it to the client.
  * @public
  */
-export interface VoiceConnectParams {
-    /**
-     * Protocol version advertised by the client (query parameter `v`). 0 (or absent) is the legacy PBX protocol; >= 1 enables the sessionStarted event; >= 2 additionally means the client WAITS for `started` before opening its recognition stream and honours the `speech` payload on it.
-     * @public
-     */
-    protocolVersion?: number | undefined;
-    /**
-     * Caller-supplied session id, unique per connection.
-     * @public
-     */
-    sessionId: string;
-    /**
-     * Attach the running session `sessionId` again after a close with code `1001` or a lost connection. The session keeps its history and does not play the greeting again.
-     * @public
-     */
-    resume?: boolean | undefined;
-    /**
-     * The client connects again with `resume` after a close with code `1001`. Only then can the service move a PBX text-socket call to another server while the call runs; without it the socket stays open until the call ends or the server stops.
-     * @public
-     */
-    reconnect?: boolean | undefined;
+export interface VoiceTranscriptionEvent {
     /**
-     * Agent to run on this session.
-     * @public
-     */
-    agentId: string;
-    /**
-     * Published agent version to run. Mutually exclusive with agentDraft. When neither is set, the agent's deployment configuration selects the version. The PBX sends it only for a preview number (`<agentId>-vN`, WMS-27502); a customer call never carries it.
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
-    agentVersion?: number | undefined;
+    type: VoiceEventType;
     /**
-     * Run the agent's current draft (for testing). Mutually exclusive with agentVersion.
+     * Id of the utterance. The interim chunks and the final chunk of one utterance have the same id. Absent: the service makes a new id.
      * @public
      */
-    agentDraft?: boolean | undefined;
+    id?: string | undefined;
     /**
-     * PBX call id, when the session is bridged to a call.
+     * Text of the chunk. Must not be empty.
      * @public
      */
-    callId?: string | undefined;
+    text: string;
     /**
-     * Leg of the PBX call inside `callId` (pbx-data-engine's flow index). Kept on the session as metadata beside `callId` and `dialedNumber`; never interpreted by the runtime.
+     * Confidence of the recogniser. The service ignores a value outside 0 to 1.
      * @public
      */
-    flowIndex?: number | undefined;
+    confidence?: number | undefined;
     /**
-     * The number the caller dialled to reach the agent. Kept on the session as metadata beside `callId` and `flowIndex`.
+     * True when the chunk is the final transcript of the utterance. False for an interim result. Absent: true.
      * @public
      */
-    dialedNumber?: string | undefined;
+    final?: boolean | undefined;
     /**
-     * BCP-47 language of the conversation, e.g. 'en-US'.
+     * Language that the recogniser identified for the chunk. The service records it and does not change the conversation language.
      * @public
      */
     language?: string | undefined;
-    callerName?: string | undefined;
-    callerNumber?: string | undefined;
-    callerEmail?: string | undefined;
-    callerCompany?: string | undefined;
-    /**
-     * Override the TTS voice for this session, as a voice URI, e.g. 'elevenlabs://<voiceId>?apiKey=...'. Defaults to the agent's configured voice.
-     * @public
-     */
-    voice?: string | undefined;
     /**
-     * First message spoken to the caller. Overrides the agent's configured greeting.
+     * Time of the chunk, in epoch ms. Absent: the time when the service received it.
      * @public
      */
-    initialMessage?: string | undefined;
+    time?: number | undefined;
+}
+/**
+ * The session ended. This is the last event of the session.
+ * @public
+ */
+export interface VoiceSessionEndedEvent {
     /**
-     * Base64-encoded JSON object of agent variables for this session, e.g. base64('\{"orderId":"123"\}').
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
-    variables?: string | undefined;
+    type: VoiceEventType;
     /**
-     * How caller speech reaches the service. Defaults to TRANSCRIPTION (PBX). Web/app clients use AUDIO.
+     * Why the session ended.
      * @public
      */
-    inputMode?: VoiceInputMode | undefined;
+    reason?: string | undefined;
     /**
-     * Format of client `audio` chunks in AUDIO mode.
+     * Present when the session ended because of an error.
      * @public
      */
-    inputAudio?: VoiceAudioFormat | undefined;
+    error?: string | undefined;
     /**
-     * Preferred format for server-synthesized `play` frames.
+     * Time of the event, in epoch ms.
      * @public
      */
-    outputAudio?: VoiceAudioFormat | undefined;
+    time: number;
 }
 /**
- * An error surfaced on the connection. Non-terminal unless followed by a SessionEndedEvent.
+ * An error on the connection. The session continues, unless a `ended` event follows.
  * @public
  */
 export interface VoiceErrorEvent {
     /**
-     * Discriminator for the inner payload of an `event` frame.
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
     type: VoiceEventType;
     /**
-     * Stable machine-readable error code.
+     * Error code. The value is stable.
      * @public
      */
     code: string;
+    /**
+     * Description of the error.
+     * @public
+     */
     message: string;
-    time: number;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface ErrorWebsocketEvent {
-    type: string;
     /**
-     * An error surfaced on the connection. Non-terminal unless followed by a SessionEndedEvent.
+     * Time of the event, in epoch ms.
      * @public
      */
-    event: VoiceErrorEvent;
+    time: number;
 }
 /**
- * Terminate the session and hang up the call.
+ * A chunk of synthesized audio of a reply. `data` is base64 of PCM audio at the sample rate of the `say` command. An empty `data` marks the end of the stream.
  * @public
  */
-export interface HangupCommandData {
-    reason?: string | undefined;
+export interface VoicePlayCommand {
+    replyId: string;
+    streamId: string;
+    data: string;
     /**
-     * Discriminator for the inner payload of a `command` frame.
+     * Value of `type` in the `command` of a command frame.
      * @public
      */
     type: VoiceCommandType;
 }
 /**
- * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
+ * Audio of a reply that the service synthesizes. `play` commands with this stream id follow the `say` command.
  * @public
  */
-export interface HangupWebsocketCommand {
-    type: string;
+export interface VoiceSayCommandAudioOptions {
+    streamId: string;
     /**
-     * Terminate the session and hang up the call.
+     * Sample rate of the `pcm_s16le` mono audio of the `play` commands, in Hz: 8000, 12000, 16000, 24000, 32000, 44100 or 48000.
      * @public
      */
-    command: HangupCommandData;
+    sampleRate: number;
 }
 /**
- * The caller interrupted (barged in over) a reply that was being played.
+ * Play a reply to the caller. When `audio` is present, the service sends the audio of the reply in `play` commands. When `audio` is absent, the client shows `text` or synthesizes it itself.
  * @public
  */
-export interface InterruptionEvent {
+export interface VoiceSayCommand {
     /**
-     * Discriminator for the inner payload of an `event` frame.
+     * Text of the reply.
      * @public
      */
-    type: VoiceEventType;
+    text: string;
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * Id of the reply.
      * @public
      */
     replyId: string;
     /**
-     * Portion of the reply that had been played before the interruption.
+     * True when the caller can interrupt the reply.
      * @public
      */
-    playedText: string;
+    interruptible?: boolean | undefined;
     /**
-     * The full reply text that was being played.
+     * Audio of a reply that the service synthesizes. `play` commands with this stream id follow the `say` command.
      * @public
      */
-    originalText: string;
-    time: number;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface InterruptionWebsocketEvent {
-    type: string;
+    audio?: VoiceSayCommandAudioOptions | undefined;
     /**
-     * The caller interrupted (barged in over) a reply that was being played.
+     * Value of `type` in the `command` of a command frame.
      * @public
      */
-    event: InterruptionEvent;
+    type: VoiceCommandType;
 }
 /**
- * Opens an agent audio stream. `MediaCloudAudio` messages of the stream follow until `audio.end`. Streams do not interleave.
+ * Speech recognition settings of the call. The client applies them to its recogniser.
  * @public
  */
-export interface MediaAudioBeginFrame {
-    type: MediaAudioBeginFrameType;
+export interface VoiceSpeechPayload {
     /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
+     * Recogniser URI. Always present.
      * @public
      */
-    replyId: string;
+    engine: string;
     /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
+     * All configured languages. These are the candidates for language identification.
      * @public
      */
-    streamId: string;
+    languages: string[];
     /**
-     * Sample rate of the stream's `pcm_s16le` mono audio in Hz: 8000, 16000, 22050 or 24000.
+     * Language at the start of the call. Always a member of `languages`.
      * @public
      */
-    sampleRate: number;
-}
-/**
- * Closes an agent audio stream.
- * @public
- */
-export interface MediaAudioEndFrame {
-    type: MediaAudioEndFrameType;
+    preferredLanguage: string;
     /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
+     * True to identify the spoken language of each utterance. False when `languages` has fewer languages than the recogniser needs, or when the recogniser cannot identify languages.
      * @public
      */
-    streamId: string;
+    autoDetect: boolean;
 }
 /**
- * Lower the volume of a playing reply.
+ * The session is ready. Sent once after the handshake, on the PBX voice socket only when the client sends protocol version 1 or more.
  * @public
  */
-export interface MediaDuckFrame {
-    type: MediaDuckFrameType;
+export interface VoiceSessionStartedEvent {
     /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
+     * Value of `type` in the `event` of an event frame.
      * @public
      */
-    replyId: string;
+    type: VoiceEventType;
     /**
-     * Target gain in dB.
+     * Time of the event, in epoch ms.
      * @public
      */
-    volumeDb: number;
+    time: number;
     /**
-     * Ramp duration in ms.
+     * Id of the session.
      * @public
      */
-    rampMs: number;
+    sessionId: string;
+    /**
+     * Agent of the session.
+     * @public
+     */
+    agentId: string;
+    /**
+     * Published agent version of the session. Absent when the session runs the agent draft.
+     * @public
+     */
+    agentVersion?: number | undefined;
+    /**
+     * Trace of the session. It does not change during the session.
+     * @public
+     */
+    traceId?: string | undefined;
+    /**
+     * Display name of the agent.
+     * @public
+     */
+    agentName?: string | undefined;
+    /**
+     * Category of the agent.
+     * @public
+     */
+    category?: string | undefined;
+    /**
+     * PBX group of the agent, for data access control. Empty when the agent has no group.
+     * @public
+     */
+    groupId?: string | undefined;
+    /**
+     * Speech recognition settings of the call. Present only at `protocolVersion` 2 or more and only when the agent has speech settings. Absent: use a single-language recogniser on the language of the channel.
+     * @public
+     */
+    speech?: VoiceSpeechPayload | undefined;
+    /**
+     * Music-on-hold class to play as background sound of the call. Absent: no background sound.
+     * @public
+     */
+    backgroundSound?: string | undefined;
 }
 /**
- * An error on the run. When `retryable` is false, the socket closes after it.
+ * Stop the audio of the agent.
  * @public
  */
-export interface MediaErrorFrame {
-    type: MediaErrorFrameType;
+export interface VoiceStopCommand {
     /**
-     * Stable error code, for example `START_TIMEOUT`, `PROTOCOL_ERROR` or `STT_FAILED`.
+     * Value of `type` in the `command` of a command frame.
      * @public
      */
-    code: string;
-    retryable: boolean;
-    message: string;
+    type: VoiceCommandType;
 }
 /**
- * Answer to `finish`: the cloud processed all audio. The socket closes with 1000 after it.
+ * Transfer the call to a PBX destination and end the session. Only a PBX client does the transfer.
  * @public
  */
-export interface MediaFinishedFrame {
-    type: MediaFinishedFrameType;
+export interface VoiceTransferCommand {
+    /**
+     * Dialplan context of the destination.
+     * @public
+     */
+    context: string;
+    /**
+     * Extension of the destination.
+     * @public
+     */
+    extension: string;
+    /**
+     * Value of `type` in the `command` of a command frame.
+     * @public
+     */
+    type: VoiceCommandType;
 }
 /**
- * Hang up the call after the queued agent audio plays. Terminal: the PBX answers with `finish`.
  * @public
  */
-export interface MediaHangupFrame {
-    type: MediaHangupFrameType;
+export interface DescribeVoiceProtocolInput {
 }
 /**
- * The call continues in another language.
+ * A recogniser result for the AGENT's own leg — what the caller actually heard. Same semantics as VoiceCallTranscriptionEvent; an interrupted reply simply stops.
  * @public
  */
-export interface MediaLanguageFrame {
-    type: MediaLanguageFrameType;
+export interface VoiceCallAgentTranscriptionEvent {
+    id: string;
+    text: string;
+    final: boolean;
     /**
-     * Language that the call continues in.
+     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
+     *     constrained string rather than an enum: the offerable set is a property of the selected
+     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
+     *     when the model was released.
      * @public
      */
-    language: string;
+    language?: string | undefined;
+    time?: number | undefined;
 }
 /**
- * Return a reply to full volume.
+ * The session ended. After `transfer` another session of the same call may follow with its own `started`.
  * @public
  */
-export interface MediaRestoreFrame {
-    type: MediaRestoreFrameType;
-    /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
-     * @public
-     */
-    replyId: string;
+export interface VoiceCallEndedEvent {
     /**
-     * Ramp duration in ms.
+     * Id of the voice session.
      * @public
      */
-    rampMs: number;
+    sessionId: string;
+    reason: VoiceCallEndReason;
 }
 /**
- * The agent that answers the run.
+ * The caller talked over the agent. `playedText` is what they heard before the cut.
  * @public
  */
-export interface MediaStartedAgent {
-    name: string;
+export interface VoiceCallInterruptionEvent {
+    replyId?: string | undefined;
+    playedText?: string | undefined;
+    originalText?: string | undefined;
 }
 /**
- * Caller voice detection of a run, with every value resolved.
+ * The conversation language changed on a recognised utterance.
  * @public
  */
-export interface MediaVoiceDetection {
+export interface VoiceCallLanguageEvent {
     /**
-     * Silence in ms that ends the caller speech.
+     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
+     *     constrained string rather than an enum: the offerable set is a property of the selected
+     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
+     *     when the model was released.
      * @public
      */
-    silenceMs: number;
+    from?: string | undefined;
     /**
-     * Mean frame level in dBFS below which caller audio is silence.
+     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
+     *     constrained string rather than an enum: the offerable set is a property of the selected
+     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
+     *     when the model was released.
      * @public
      */
-    silenceLevelDb: number;
+    to: string;
+    reason: string;
+}
+/**
+ * The PBX finished (or abandoned) playing a reply.
+ * @public
+ */
+export interface VoiceCallPlaybackEvent {
+    replyId: string;
+    complete: boolean;
+}
+/**
+ * The caller's completed utterance that started a turn — after the runtime's filler and duplicate filter.
+ * @public
+ */
+export interface VoiceCallReplyEvent {
+    id: string;
+    text: string;
     /**
-     * Mean frame level in dBFS from which caller audio is speech.
+     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
+     *     constrained string rather than an enum: the offerable set is a property of the selected
+     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
+     *     when the model was released.
      * @public
      */
-    speechLevelDb: number;
+    language?: string | undefined;
+}
+/**
+ * One sentence the agent speaks, as it is handed to the PBX (after playback sanitizing). Sentences of one reply share `replyId`.
+ * @public
+ */
+export interface VoiceCallSayEvent {
+    replyId: string;
+    text: string;
+    interruptible: boolean;
 }
 /**
- * The cloud accepted `start`. Sent once per connection.
+ * The session that answered the call, sent once per session of the call (a transfer starts another).
  * @public
  */
-export interface MediaStartedFrame {
-    type: MediaStartedFrameType;
+export interface VoiceCallStartedEvent {
     /**
-     * Id of the cloud stream of this run. Cloud logs carry it; use it to match PBX and cloud logs.
+     * Id of the voice session.
      * @public
      */
-    streamId: string;
+    sessionId: string;
     /**
-     * Present when the run purpose is `agent`.
+     * The session's trace, for the LLM-level detail this stream deliberately omits. Absent for a built-in agent, whose trace is not the tenant's to read.
      * @public
      */
-    agent?: MediaStartedAgent | undefined;
+    traceId?: string | undefined;
+    agentId: string;
+    agentName?: string | undefined;
     /**
-     * Group of the agent. Absent when the agent has none.
+     * Published agent version number. Monotonic within an agent's version history, starting at 1.
      * @public
      */
-    groupId?: string | undefined;
+    agentVersion?: number | undefined;
     /**
-     * Music-on-hold class that the PBX plays after each caller turn until agent audio starts. Absent: none.
+     * True when the session runs the agent's draft.
      * @public
      */
-    backgroundSound?: string | undefined;
+    agentDraft?: boolean | undefined;
     /**
-     * Caller voice detection that the PBX uses for the `vad` frames of this run. Absent: the PBX keeps its defaults.
+     * BCP-47 language the call started in.
      * @public
      */
-    voiceDetection?: MediaVoiceDetection | undefined;
-}
-/**
- * Stop every agent audio stream of the run: the one that plays and the ones that wait.
- * @public
- */
-export interface MediaStopFrame {
-    type: MediaStopFrameType;
+    language?: string | undefined;
 }
 /**
- * Words of one party. Partials and the final of one utterance share `id`. For the agent, the words that played.
+ * A tool call the agent made during a turn. Name and phase only — arguments and results live in the trace. Sent to a caller signed in to the system; never to a guest.
  * @public
  */
-export interface MediaTranscriptionFrame {
-    type: MediaTranscriptionFrameType;
-    /**
-     * Party that spoke. The agent's own words are `callee`.
-     * @public
-     */
-    party: MediaParty;
+export interface VoiceCallToolEvent {
     /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
+     * Identifies the tool call inside the turn — the model's own call id, not the PBX call this stream follows.
      * @public
      */
-    id: string;
-    final: boolean;
-    text: string;
+    toolCallId: string;
+    toolName: string;
     /**
-     * Run-clock time or duration in ms of PBX audio since the run started. It continues across reconnects.
+     * Phase of a tool call of the agent.
      * @public
      */
-    startMs: number;
+    phase: VoiceToolPhase;
     /**
-     * Run-clock time or duration in ms of PBX audio since the run started. It continues across reconnects.
+     * The phrase spoken while the tool runs, when the capability configures one.
      * @public
      */
-    endMs: number;
+    startMessage?: string | undefined;
+}
+/**
+ * A recogniser result for the caller's speech. Interim results (`final: false`) are replaced by the final one with the same `id`.
+ * @public
+ */
+export interface VoiceCallTranscriptionEvent {
+    id: string;
+    text: string;
+    final: boolean;
     /**
      * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
      *     constrained string rather than an enum: the offerable set is a property of the selected
@@ -994,147 +682,277 @@ export interface MediaTranscriptionFrame {
      *     when the model was released.
      * @public
      */
-    language: string;
-    /**
-     * Recogniser confidence from 0 to 1. Absent when the text is not from a recogniser, or the engine gives none.
-     * @public
-     */
-    confidence?: number | undefined;
-}
-/**
- * Transfer the call after the queued agent audio plays. Terminal: the PBX answers with `finish`.
- * @public
- */
-export interface MediaTransferFrame {
-    type: MediaTransferFrameType;
+    language?: string | undefined;
     /**
-     * Dialplan context of the target.
+     * Milliseconds since the call started, as the PBX stamps it.
      * @public
      */
-    context: string;
+    time?: number | undefined;
     /**
-     * Extension of the target.
+     * Present on a final that did not go to the agent as a caller turn. It came while the agent spoke, and its recognition confidence was below the minimum of the call or the recogniser labelled it with another speaker.
      * @public
      */
-    extension: string;
+    finalDropReason?: VoiceTranscriptionDropReason | undefined;
 }
 /**
- * One example of each message that the cloud sends.
+ * One event on a live call. The SSE frame's `event:` is the member name and `data:` this whole union document, so a frame reads `data: \{"say": \{...\}\}` — never the member's payload on its own.
+ * @public
+ */
+export type VoiceCallEvent = VoiceCallEvent.AgentTranscriptionMember | VoiceCallEvent.EndedMember | VoiceCallEvent.InterruptionMember | VoiceCallEvent.LanguageMember | VoiceCallEvent.PlaybackMember | VoiceCallEvent.ReplyMember | VoiceCallEvent.SayMember | VoiceCallEvent.StartedMember | VoiceCallEvent.ToolMember | VoiceCallEvent.TranscriptionMember | VoiceCallEvent.$UnknownMember;
+/**
  * @public
  */
-export interface MediaCloudFrames {
-    /**
-     * The cloud accepted `start`. Sent once per connection.
-     * @public
-     */
-    started?: MediaStartedFrame | undefined;
+export declare namespace VoiceCallEvent {
     /**
-     * Words of one party. Partials and the final of one utterance share `id`. For the agent, the words that played.
+     * The session that answered the call, sent once per session of the call (a transfer starts another).
      * @public
      */
-    transcription?: MediaTranscriptionFrame | undefined;
+    interface StartedMember {
+        started: VoiceCallStartedEvent;
+        transcription?: never;
+        agentTranscription?: never;
+        reply?: never;
+        say?: never;
+        playback?: never;
+        interruption?: never;
+        language?: never;
+        tool?: never;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * Opens an agent audio stream. `MediaCloudAudio` messages of the stream follow until `audio.end`. Streams do not interleave.
+     * A recogniser result for the caller's speech. Interim results (`final: false`) are replaced by the final one with the same `id`.
      * @public
      */
-    audioBegin?: MediaAudioBeginFrame | undefined;
+    interface TranscriptionMember {
+        started?: never;
+        transcription: VoiceCallTranscriptionEvent;
+        agentTranscription?: never;
+        reply?: never;
+        say?: never;
+        playback?: never;
+        interruption?: never;
+        language?: never;
+        tool?: never;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * One binary audio message. In the describe answer it is base64; on the socket it is raw bytes.
+     * A recogniser result for the AGENT's own leg — what the caller actually heard. Same semantics as VoiceCallTranscriptionEvent; an interrupted reply simply stops.
      * @public
      */
-    audio?: Uint8Array | undefined;
+    interface AgentTranscriptionMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription: VoiceCallAgentTranscriptionEvent;
+        reply?: never;
+        say?: never;
+        playback?: never;
+        interruption?: never;
+        language?: never;
+        tool?: never;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * Closes an agent audio stream.
+     * The caller's completed utterance that started a turn — after the runtime's filler and duplicate filter.
      * @public
      */
-    audioEnd?: MediaAudioEndFrame | undefined;
+    interface ReplyMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription?: never;
+        reply: VoiceCallReplyEvent;
+        say?: never;
+        playback?: never;
+        interruption?: never;
+        language?: never;
+        tool?: never;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * Stop every agent audio stream of the run: the one that plays and the ones that wait.
+     * One sentence the agent speaks, as it is handed to the PBX (after playback sanitizing). Sentences of one reply share `replyId`.
      * @public
      */
-    stop?: MediaStopFrame | undefined;
+    interface SayMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription?: never;
+        reply?: never;
+        say: VoiceCallSayEvent;
+        playback?: never;
+        interruption?: never;
+        language?: never;
+        tool?: never;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * Lower the volume of a playing reply.
+     * The PBX finished (or abandoned) playing a reply.
      * @public
      */
-    duck?: MediaDuckFrame | undefined;
+    interface PlaybackMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription?: never;
+        reply?: never;
+        say?: never;
+        playback: VoiceCallPlaybackEvent;
+        interruption?: never;
+        language?: never;
+        tool?: never;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * Return a reply to full volume.
+     * The caller talked over the agent. `playedText` is what they heard before the cut.
      * @public
      */
-    restore?: MediaRestoreFrame | undefined;
+    interface InterruptionMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription?: never;
+        reply?: never;
+        say?: never;
+        playback?: never;
+        interruption: VoiceCallInterruptionEvent;
+        language?: never;
+        tool?: never;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * The call continues in another language.
+     * The conversation language changed on a recognised utterance.
      * @public
      */
-    language?: MediaLanguageFrame | undefined;
+    interface LanguageMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription?: never;
+        reply?: never;
+        say?: never;
+        playback?: never;
+        interruption?: never;
+        language: VoiceCallLanguageEvent;
+        tool?: never;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * Transfer the call after the queued agent audio plays. Terminal: the PBX answers with `finish`.
+     * A tool call the agent made during a turn. Name and phase only — arguments and results live in the trace. Sent to a caller signed in to the system; never to a guest.
      * @public
      */
-    transfer?: MediaTransferFrame | undefined;
+    interface ToolMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription?: never;
+        reply?: never;
+        say?: never;
+        playback?: never;
+        interruption?: never;
+        language?: never;
+        tool: VoiceCallToolEvent;
+        ended?: never;
+        $unknown?: never;
+    }
     /**
-     * Hang up the call after the queued agent audio plays. Terminal: the PBX answers with `finish`.
+     * The session ended. After `transfer` another session of the same call may follow with its own `started`.
      * @public
      */
-    hangup?: MediaHangupFrame | undefined;
+    interface EndedMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription?: never;
+        reply?: never;
+        say?: never;
+        playback?: never;
+        interruption?: never;
+        language?: never;
+        tool?: never;
+        ended: VoiceCallEndedEvent;
+        $unknown?: never;
+    }
     /**
-     * An error on the run. When `retryable` is false, the socket closes after it.
      * @public
      */
-    error?: MediaErrorFrame | undefined;
+    interface $UnknownMember {
+        started?: never;
+        transcription?: never;
+        agentTranscription?: never;
+        reply?: never;
+        say?: never;
+        playback?: never;
+        interruption?: never;
+        language?: never;
+        tool?: never;
+        ended?: never;
+        $unknown: [string, any];
+    }
     /**
-     * Answer to `finish`: the cloud processed all audio. The socket closes with 1000 after it.
-     * @public
+     * @deprecated unused in schema-serde mode.
+     *
      */
-    finished?: MediaFinishedFrame | undefined;
+    interface Visitor<T> {
+        started: (value: VoiceCallStartedEvent) => T;
+        transcription: (value: VoiceCallTranscriptionEvent) => T;
+        agentTranscription: (value: VoiceCallAgentTranscriptionEvent) => T;
+        reply: (value: VoiceCallReplyEvent) => T;
+        say: (value: VoiceCallSayEvent) => T;
+        playback: (value: VoiceCallPlaybackEvent) => T;
+        interruption: (value: VoiceCallInterruptionEvent) => T;
+        language: (value: VoiceCallLanguageEvent) => T;
+        tool: (value: VoiceCallToolEvent) => T;
+        ended: (value: VoiceCallEndedEvent) => T;
+        _: (name: string, value: any) => T;
+    }
 }
 /**
  * DTMF input that the PBX collected. The cloud takes `digits` as the caller's reply.
  * @public
  */
-export interface MediaDtmfFrame {
-    type: MediaDtmfFrameType;
+export interface VoiceCallDtmfFrame {
     /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
+     * Value of `type` in a text frame from the client (the PBX).
      * @public
      */
+    type: VoiceCallClientFrameType;
     id: string;
     digits: string;
 }
 /**
- * The run ended on the PBX. The PBX sends no audio after it and waits up to 3 s for `finished`.
+ * The run ended on the PBX. The PBX sends no audio after it and waits up to 3 s for `ended`.
  * @public
  */
-export interface MediaFinishFrame {
-    type: MediaFinishFrameType;
+export interface VoiceCallEndFrame {
+    /**
+     * Value of `type` in a text frame from the client (the PBX).
+     * @public
+     */
+    type: VoiceCallClientFrameType;
     /**
-     * Who ended the run on the PBX.
+     * Who ended the run on the PBX. The live events of the call report it as `VoiceCallEndReason`: `caller-hangup` as `hangup`, `agent-hangup` as `complete`, `agent-transfer` as `transfer`, `timeout` as `timeout`.
      * @public
      */
-    reason: MediaFinishReason;
+    reason: VoiceCallClientEndReason;
 }
 /**
  * An agent audio stream changed state on the PBX.
  * @public
  */
-export interface MediaPlaybackFrame {
-    type: MediaPlaybackFrameType;
+export interface VoiceCallPlaybackFrame {
     /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
+     * Value of `type` in a text frame from the client (the PBX).
      * @public
      */
+    type: VoiceCallClientFrameType;
     replyId: string;
-    /**
-     * Id of a reply, a stream, an utterance or a DTMF input, unique in the run.
-     * @public
-     */
     streamId: string;
     /**
      * Event of an agent audio stream.
      * @public
      */
-    event: MediaPlaybackEvent;
+    event: VoiceCallPlaybackPhase;
     /**
      * Run-clock time or duration in ms of PBX audio since the run started. It continues across reconnects.
      * @public
@@ -1146,16 +964,16 @@ export interface MediaPlaybackFrame {
      */
     playedMs?: number | undefined;
     /**
-     * Present with `stopped`.
+     * True when the PBX dropped the stream, for example at its queue limit. Only with `stopped`. Absent: false.
      * @public
      */
-    cause?: MediaPlaybackStopCause | undefined;
+    dropped?: boolean | undefined;
 }
 /**
- * The agent of a media run. At most one of `draft` and `version` is set.
+ * The agent of the run. At most one of `draft` and `version` is set.
  * @public
  */
-export interface MediaStartAgent {
+export interface VoiceCallStartAgent {
     /**
      * Agent id.
      * @public
@@ -1173,31 +991,10 @@ export interface MediaStartAgent {
     version?: number | undefined;
 }
 /**
- * Format of the binary PBX audio (`MediaPbxAudio`). Channel 0 is the caller; channel 1, when present, is the callee. Samples interleave. The cloud accepts only 8000 Hz and one channel at this time; another format closes the socket with 4400.
- * @public
- */
-export interface MediaAudioFormat {
-    /**
-     * Sample encoding of PBX audio.
-     * @public
-     */
-    encoding: MediaAudioEncoding;
-    /**
-     * Sample rate in Hz.
-     * @public
-     */
-    sampleRate: number;
-    /**
-     * Number of channels.
-     * @public
-     */
-    channels: number;
-}
-/**
- * Caller identity. Each member is absent, null or empty when unknown.
+ * Caller identity from the lookup application of the PBX: an external lookup or the local phonebook. Each member is absent, null or empty when unknown.
  * @public
  */
-export interface MediaStartCaller {
+export interface VoiceCallStartCaller {
     name?: string | undefined;
     number?: string | undefined;
     email?: string | undefined;
@@ -1207,13 +1004,12 @@ export interface MediaStartCaller {
  * First text frame of a socket, sent within 10 s of the upgrade. Opens a run, or attaches a reconnect to it.
  * @public
  */
-export interface MediaStartFrame {
-    type: MediaStartFrameType;
+export interface VoiceCallStartFrame {
     /**
-     * Purpose of a media run.
+     * Value of `type` in a text frame from the client (the PBX).
      * @public
      */
-    purpose: MediaRunPurpose;
+    type: VoiceCallClientFrameType;
     /**
      * Id of the PBX call. Use it to correlate the call events of the run.
      * @public
@@ -1240,35 +1036,20 @@ export interface MediaStartFrame {
      */
     resumeAtMs: number;
     /**
-     * Format of the binary PBX audio (`MediaPbxAudio`). Channel 0 is the caller; channel 1, when present, is the callee. Samples interleave. The cloud accepts only 8000 Hz and one channel at this time; another format closes the socket with 4400.
-     * @public
-     */
-    audio: MediaAudioFormat;
-    /**
-     * Present when `purpose` is `agent`.
-     * @public
-     */
-    agent?: MediaStartAgent | undefined;
-    /**
-     * Languages to recognise when `purpose` is `transcription`. May be empty only when `autoDetect` is true. Ignored when `purpose` is `agent`: the agent settings give the languages.
+     * The agent of the run. At most one of `draft` and `version` is set.
      * @public
      */
-    languages?: string[] | undefined;
+    agent: VoiceCallStartAgent;
     /**
-     * Language of the call. Required when `purpose` is `agent`: the agent starts in its configured language that matches it. With `transcription`, absent or null means the first of `languages`.
+     * Language of the call, for example `en-US`. The agent starts in its configured language that matches it.
      * @public
      */
-    preferredLanguage?: string | undefined;
-    /**
-     * True to identify the caller's language among `languages` when `purpose` is `transcription`. Absent: false. Ignored when `purpose` is `agent`.
-     * @public
-     */
-    autoDetect?: boolean | undefined;
+    language: string;
     /**
-     * Caller identity. Each member is absent, null or empty when unknown.
+     * Caller identity from the lookup application of the PBX: an external lookup or the local phonebook. Each member is absent, null or empty when unknown.
      * @public
      */
-    caller: MediaStartCaller;
+    caller: VoiceCallStartCaller;
     /**
      * Number that the caller dialled. Absent, null or empty when unknown.
      * @public
@@ -1281,18 +1062,17 @@ export interface MediaStartFrame {
     variables?: string | undefined;
 }
 /**
- * Voice activity of a party as the PBX detects it. Sent when the state changes, and every 100 ms while agent audio plays.
+ * Voice activity of the caller as the PBX detects it. Sent when the state changes, and every 100 ms while agent audio plays.
  * @public
  */
-export interface MediaVadFrame {
-    type: MediaVadFrameType;
+export interface VoiceCallVadFrame {
     /**
-     * Party whose audio the report describes.
+     * Value of `type` in a text frame from the client (the PBX).
      * @public
      */
-    party: MediaParty;
+    type: VoiceCallClientFrameType;
     /**
-     * True while the party speaks.
+     * True while the caller speaks.
      * @public
      */
     speaking: boolean;
@@ -1308,571 +1088,564 @@ export interface MediaVadFrame {
     atMs: number;
 }
 /**
- * One example of each message that the PBX sends.
+ * One example of each text message that the client (the PBX) sends on the call socket.
  * @public
  */
-export interface MediaPbxFrames {
+export interface VoiceCallClientFrames {
     /**
      * First text frame of a socket, sent within 10 s of the upgrade. Opens a run, or attaches a reconnect to it.
      * @public
      */
-    start?: MediaStartFrame | undefined;
+    start?: VoiceCallStartFrame | undefined;
     /**
-     * One binary audio message. In the describe answer it is base64; on the socket it is raw bytes.
+     * Voice activity of the caller as the PBX detects it. Sent when the state changes, and every 100 ms while agent audio plays.
      * @public
      */
-    audio?: Uint8Array | undefined;
+    vad?: VoiceCallVadFrame | undefined;
     /**
-     * Voice activity of a party as the PBX detects it. Sent when the state changes, and every 100 ms while agent audio plays.
+     * DTMF input that the PBX collected. The cloud takes `digits` as the caller's reply.
      * @public
      */
-    vad?: MediaVadFrame | undefined;
+    dtmf?: VoiceCallDtmfFrame | undefined;
     /**
-     * DTMF input that the PBX collected. The cloud takes `digits` as the caller's reply.
+     * An agent audio stream changed state on the PBX.
      * @public
      */
-    dtmf?: MediaDtmfFrame | undefined;
+    playback?: VoiceCallPlaybackFrame | undefined;
     /**
-     * An agent audio stream changed state on the PBX.
+     * The run ended on the PBX. The PBX sends no audio after it and waits up to 3 s for `ended`.
+     * @public
+     */
+    end?: VoiceCallEndFrame | undefined;
+}
+/**
+ * Opens an agent audio stream. Binary audio messages of the stream follow until `audio.end`. Streams do not interleave.
+ * @public
+ */
+export interface VoiceCallAudioBeginFrame {
+    /**
+     * Value of `type` in a text frame from the server (the cloud).
+     * @public
+     */
+    type: VoiceCallServerFrameType;
+    replyId: string;
+    streamId: string;
+    /**
+     * Sample rate of the stream's `pcm_s16le` mono audio in Hz: 8000, 12000, 16000, 24000, 32000, 44100 or 48000.
      * @public
      */
-    playback?: MediaPlaybackFrame | undefined;
+    sampleRate: number;
+}
+/**
+ * Closes an agent audio stream.
+ * @public
+ */
+export interface VoiceCallAudioEndFrame {
+    /**
+     * Value of `type` in a text frame from the server (the cloud).
+     * @public
+     */
+    type: VoiceCallServerFrameType;
+    streamId: string;
+}
+/**
+ * Change the volume of a playing reply. `volumeDb` 0 returns it to full volume.
+ * @public
+ */
+export interface VoiceCallDuckFrame {
+    /**
+     * Value of `type` in a text frame from the server (the cloud).
+     * @public
+     */
+    type: VoiceCallServerFrameType;
+    replyId: string;
+    /**
+     * Target gain in dB.
+     * @public
+     */
+    volumeDb: number;
+    /**
+     * Ramp duration in ms.
+     * @public
+     */
+    rampMs: number;
+}
+/**
+ * Answer to `end`: the cloud processed all audio. The socket closes with close code 1000 after it.
+ * @public
+ */
+export interface VoiceCallEndedFrame {
     /**
-     * The run ended on the PBX. The PBX sends no audio after it and waits up to 3 s for `finished`.
+     * Value of `type` in a text frame from the server (the cloud).
      * @public
      */
-    finish?: MediaFinishFrame | undefined;
+    type: VoiceCallServerFrameType;
 }
 /**
- * One example of each message of the PBX media socket.
+ * An error on the run. When `retryable` is false, the socket closes after it.
  * @public
  */
-export interface MediaProtocolFrames {
+export interface VoiceCallErrorFrame {
     /**
-     * Messages that the PBX sends.
+     * Value of `type` in a text frame from the server (the cloud).
      * @public
      */
-    pbx?: MediaPbxFrames | undefined;
+    type: VoiceCallServerFrameType;
     /**
-     * Messages that the cloud sends.
+     * Stable error code, for example `START_TIMEOUT`, `PROTOCOL_ERROR` or `STT_FAILED`.
      * @public
      */
-    cloud?: MediaCloudFrames | undefined;
+    code: string;
+    retryable: boolean;
+    message: string;
 }
 /**
- * Client-reported progress of an in-flight reply playback.
+ * Hang up the call after the queued agent audio plays. Terminal: the PBX answers with `end`.
  * @public
  */
-export interface PlaybackEvent {
-    /**
-     * Discriminator for the inner payload of an `event` frame.
-     * @public
-     */
-    type: VoiceEventType;
+export interface VoiceCallHangupFrame {
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * Value of `type` in a text frame from the server (the cloud).
      * @public
      */
-    replyId: string;
-    text: string;
-    complete: boolean;
-    startTime: number;
-    endTime?: number | undefined;
+    type: VoiceCallServerFrameType;
 }
 /**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+ * The call continues in another language.
  * @public
  */
-export interface PlaybackWebsocketEvent {
-    type: string;
+export interface VoiceCallLanguageFrame {
+    /**
+     * Value of `type` in a text frame from the server (the cloud).
+     * @public
+     */
+    type: VoiceCallServerFrameType;
     /**
-     * Client-reported progress of an in-flight reply playback.
+     * Language that the call continues in.
      * @public
      */
-    event: PlaybackEvent;
+    language: string;
 }
 /**
- * A chunk of synthesized reply audio to play. `data` is base64, PCM-encoded at the sample rate declared in the matching `say` command; an empty `data` marks the end of the stream.
+ * The agent that answers the run.
  * @public
  */
-export interface PlayCommandData {
+export interface VoiceCallStartedAgent {
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * Agent id, as in `start.agent.id`.
      * @public
      */
-    replyId: string;
+    id: string;
     /**
-     * Identifies one synthesized audio stream within a session; carried by a `say` command and the `play` frames that render it.
+     * Display name of the agent.
      * @public
      */
-    streamId: string;
-    data: string;
+    name: string;
     /**
-     * Discriminator for the inner payload of a `command` frame.
+     * Group of the agent. Absent when the agent has none.
      * @public
      */
-    type: VoiceCommandType;
+    groupId?: string | undefined;
 }
 /**
- * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
+ * Answer to `start`: the session of the run is ready on this connection. Sent once per connection, also after a reconnect. Example: `\{"type":"started","connectionId":"0f6d…","agent":\{"id":"a_qqyHI7xWBH","name":"Support"\}\}`.
  * @public
  */
-export interface PlayWebsocketCommand {
-    type: string;
+export interface VoiceCallStartedFrame {
     /**
-     * A chunk of synthesized reply audio to play. `data` is base64, PCM-encoded at the sample rate declared in the matching `say` command; an empty `data` marks the end of the stream.
+     * Value of `type` in a text frame from the server (the cloud).
      * @public
      */
-    command: PlayCommandData;
-}
-/**
- * A final caller phrase that should trigger an agent turn (PBX / TRANSCRIPTION mode).
- * @public
- */
-export interface ReplyEvent {
+    type: VoiceCallServerFrameType;
     /**
-     * Discriminator for the inner payload of an `event` frame.
+     * Id of this connection. Each connection of the run has its own.
      * @public
      */
-    type: VoiceEventType;
+    connectionId: string;
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * The agent that answers the run.
      * @public
      */
-    id: string;
-    text: string;
+    agent: VoiceCallStartedAgent;
     /**
-     * Language the recogniser identified for this completed utterance. THIS is the one that drives the language switch. When the recogniser reports several languages inside one utterance, the PBX sends the LAST labelled segment's language — it is the one the caller is speaking now.
+     * Music-on-hold class that the PBX plays after each caller turn until agent audio starts. Absent: none.
      * @public
      */
-    language?: string | undefined;
-    startTime: number;
-    endTime: number;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface ReplyWebsocketEvent {
-    type: string;
+    backgroundSound?: string | undefined;
     /**
-     * A final caller phrase that should trigger an agent turn (PBX / TRANSCRIPTION mode).
+     * Caller voice detection that the PBX uses for the `vad` frames of this run. The cloud sends every member. Absent: the PBX keeps its defaults.
      * @public
      */
-    event: ReplyEvent;
+    voiceDetection?: VoiceDetectionSettings | undefined;
 }
 /**
- * Audio-stream parameters when the server synthesizes the reply itself (TTS provider configured on the agent): the `say` is followed by a stream of `play` frames on this stream id.
+ * Stop every agent audio stream of the run: the one that plays and the ones that wait.
  * @public
  */
-export interface SayCommandAudioOptions {
+export interface VoiceCallStopFrame {
     /**
-     * Identifies one synthesized audio stream within a session; carried by a `say` command and the `play` frames that render it.
+     * Value of `type` in a text frame from the server (the cloud).
      * @public
      */
-    streamId: string;
-    sampleRate: number;
+    type: VoiceCallServerFrameType;
 }
 /**
- * Play a reply to the caller. When `audio` is absent the client performs text-to-speech from `text`; when present the server streams synthesized `play` frames for this reply.
+ * Words of one party. Partials and the final of one utterance share `id`. For the agent, the words that played.
  * @public
  */
-export interface SayCommandData {
-    text: string;
+export interface VoiceCallTranscriptionFrame {
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * Value of `type` in a text frame from the server (the cloud).
      * @public
      */
-    replyId: string;
+    type: VoiceCallServerFrameType;
     /**
-     * Whether the caller may barge in over this reply.
+     * Party that spoke. The agent's own words are `callee`.
      * @public
      */
-    interruptible?: boolean | undefined;
+    party: VoiceCallParty;
+    id: string;
+    final: boolean;
+    text: string;
     /**
-     * Bare Amazon Polly voice id, e.g. 'Joanna' — NOT a URI. The PBX consumes this verbatim as the
-     *     AMI `Voice=` action parameter, so it takes exactly what Polly names the voice, with the
-     *     `amazon_polly://` scheme and any query parameters already stripped by whoever builds this
-     *     command out of the agent's stored `VoiceUri`.
-     *
-     *     Set only for a provider whose synthesis runs ON THE PBX (Amazon Polly); ElevenLabs and Google
-     *     are synthesized in the cloud and streamed as PCM through `play` instead, and leave this
-     *     absent.
+     * Run-clock time or duration in ms of PBX audio since the run started. It continues across reconnects.
      * @public
      */
-    voice?: string | undefined;
+    startMs: number;
     /**
-     * Audio-stream parameters when the server synthesizes the reply itself (TTS provider configured on the agent): the `say` is followed by a stream of `play` frames on this stream id.
+     * Run-clock time or duration in ms of PBX audio since the run started. It continues across reconnects.
      * @public
      */
-    audio?: SayCommandAudioOptions | undefined;
+    endMs: number;
     /**
-     * Discriminator for the inner payload of a `command` frame.
+     * BCP-47 language tag, optionally with a region subtag: 'en', 'en-US', 'pt-BR'. Deliberately a
+     *     constrained string rather than an enum: the offerable set is a property of the selected
+     *     recogniser and changes with it, so an enumeration here would cap the set at whatever was known
+     *     when the model was released.
      * @public
      */
-    type: VoiceCommandType;
-}
-/**
- * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
- * @public
- */
-export interface SayWebsocketCommand {
-    type: string;
+    language: string;
     /**
-     * Play a reply to the caller. When `audio` is absent the client performs text-to-speech from `text`; when present the server streams synthesized `play` frames for this reply.
+     * Recogniser confidence from 0 to 1. Absent when the text is not from a recogniser, or the engine gives none.
      * @public
      */
-    command: SayCommandData;
+    confidence?: number | undefined;
 }
 /**
- * Terminal server event: the session ended, optionally with an error.
+ * Transfer the call after the queued agent audio plays. Terminal: the PBX answers with `end`.
  * @public
  */
-export interface SessionEndedEvent {
+export interface VoiceCallTransferFrame {
     /**
-     * Discriminator for the inner payload of an `event` frame.
+     * Dialplan context of the destination.
      * @public
      */
-    type: VoiceEventType;
-    reason?: string | undefined;
+    context: string;
     /**
-     * Present when the session ended because of an error.
+     * Extension of the destination.
      * @public
      */
-    error?: string | undefined;
-    time: number;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface SessionEndedWebsocketEvent {
-    type: string;
+    extension: string;
     /**
-     * Terminal server event: the session ended, optionally with an error.
+     * Value of `type` in a text frame from the server (the cloud).
      * @public
      */
-    event: SessionEndedEvent;
+    type: VoiceCallServerFrameType;
 }
 /**
- * Recognition intent for the call, sent on `started` at protocolVersion >= 2. NOT recogniser
- *     parameters: the PBX translates it into its provider's API. Against Amazon Transcribe with
- *     autoDetect true that is identify-multiple-languages plus language-options and
- *     preferred-language; with autoDetect false, a single-language stream on preferredLanguage.
- *
- *     Both members the server can get wrong are RESOLVED here rather than forwarded: `engine` is
- *     always concrete (an absent one would let the PBX apply a different default than the console
- *     built its language list for), and `autoDetect` is clamped to false whenever the language count
- *     is below the recogniser's minimum or the recogniser cannot identify languages at all
- *     (forwarding true with one candidate makes Amazon reject the stream outright — a dead call).
+ * One example of each text message that the server (the cloud) sends on the call socket.
  * @public
  */
-export interface VoiceSpeechPayload {
+export interface VoiceCallServerFrames {
     /**
-     * Concrete recogniser URI, resolved from the agent or from the platform default; never absent. See TranscriptionUri.
+     * Answer to `start`: the session of the run is ready on this connection. Sent once per connection, also after a reconnect. Example: `\{"type":"started","connectionId":"0f6d…","agent":\{"id":"a_qqyHI7xWBH","name":"Support"\}\}`.
      * @public
      */
-    engine: string;
+    started?: VoiceCallStartedFrame | undefined;
     /**
-     * Every configured language, as candidates for identification.
+     * Words of one party. Partials and the final of one utterance share `id`. For the agent, the words that played.
      * @public
      */
-    languages: string[];
+    transcription?: VoiceCallTranscriptionFrame | undefined;
     /**
-     * The language the call starts in, resolved from the channel. Always a member of `languages`.
+     * Opens an agent audio stream. Binary audio messages of the stream follow until `audio.end`. Streams do not interleave.
      * @public
      */
-    preferredLanguage: string;
+    audioBegin?: VoiceCallAudioBeginFrame | undefined;
     /**
-     * Whether to identify the spoken language per utterance. Already clamped by the server — see the shape's documentation.
+     * Closes an agent audio stream.
      * @public
      */
-    autoDetect: boolean;
-}
-/**
- * Emitted once right after the handshake to confirm the session and the agent it is bound to. Sent only when the client advertises protocol version >= 1.
- * @public
- */
-export interface SessionStartedEvent {
+    audioEnd?: VoiceCallAudioEndFrame | undefined;
     /**
-     * Discriminator for the inner payload of an `event` frame.
+     * Stop every agent audio stream of the run: the one that plays and the ones that wait.
      * @public
      */
-    type: VoiceEventType;
-    time: number;
+    stop?: VoiceCallStopFrame | undefined;
     /**
-     * Caller-facing voice session identifier, unique per connection. Supplied by the client at connect (the PBX call/session id, or a client-generated id for web sessions).
+     * Change the volume of a playing reply. `volumeDb` 0 returns it to full volume.
      * @public
      */
-    sessionId: string;
+    duck?: VoiceCallDuckFrame | undefined;
     /**
-     * Agent bound to the session.
+     * The call continues in another language.
      * @public
      */
-    agentId: string;
+    language?: VoiceCallLanguageFrame | undefined;
     /**
-     * Published agent version resolved at connect; absent when running the agent draft.
+     * Transfer the call after the queued agent audio plays. Terminal: the PBX answers with `end`.
      * @public
      */
-    agentVersion?: number | undefined;
+    transfer?: VoiceCallTransferFrame | undefined;
     /**
-     * Trace recording this voice session in wilma-traces — the caller's transcribed phrases, the
-     *     model calls they drove and the tools the agent ran. Read it with the wilma-traces GetTrace
-     *     operation, or tail it live with StreamTrace to follow a call while it is in progress.
-     *
-     *     Sent once at handshake and stable for the connection: a call is one trace. A caller who
-     *     redials gets a new session and therefore a new trace, since the trace is keyed on the session
-     *     id supplied at connect.
+     * Hang up the call after the queued agent audio plays. Terminal: the PBX answers with `end`.
      * @public
      */
-    traceId?: string | undefined;
+    hangup?: VoiceCallHangupFrame | undefined;
     /**
-     * Agent display name.
+     * An error on the run. When `retryable` is false, the socket closes after it.
      * @public
      */
-    agentName?: string | undefined;
+    error?: VoiceCallErrorFrame | undefined;
     /**
-     * Agent category.
+     * Answer to `end`: the cloud processed all audio. The socket closes with close code 1000 after it.
      * @public
      */
-    category?: string | undefined;
+    ended?: VoiceCallEndedFrame | undefined;
+}
+/**
+ * The text messages of the call socket `/v2/voice/calls/ws`. The socket has no connect parameters: the first text frame is `start`.
+ * @public
+ */
+export interface VoiceCallProtocolFrames {
     /**
-     * PBX group used for data ACL; empty when the agent is not assigned to a group.
+     * One example of each text message that the client (the PBX) sends on the call socket.
      * @public
      */
-    groupId?: string | undefined;
+    client?: VoiceCallClientFrames | undefined;
     /**
-     * Recognition intent for the call. Sent only at protocolVersion >= 2 and only when the agent has `speech` configured. A v2 client that receives `started` WITHOUT this member falls back to exactly its pre-v2 behavior: a single-language stream on the channel's own `language`.
+     * One example of each text message that the server (the cloud) sends on the call socket.
      * @public
      */
-    speech?: VoiceSpeechPayload | undefined;
+    server?: VoiceCallServerFrames | undefined;
+}
+/**
+ * One example of each message that the client sends on the text socket. An event goes in `\{"type": "event", "event": \{...\}\}`; `healthcheck` has no envelope.
+ * @public
+ */
+export interface VoiceTextClientFrames {
     /**
-     * Music-on-hold class the PBX should play under the call as ambient background.
-     *
-     *     A BARE NAME, not the configuration object `VoiceBackgroundSoundSettings`: this is the RESOLVED
-     *     outcome. The server has already applied `enabled` and the class default, so the PBX gets a name
-     *     to play or nothing at all and needs no resolution rule of its own.
-     *
-     *     TOP LEVEL, NOT INSIDE `speech`, because it is not recognition intent and must not inherit
-     *     `speech`'s conditions: a call with no configured languages carries no `speech` payload and can
-     *     still have background audio.
-     *
-     *     ABSENT MEANS OFF, and absence is the only way to say it — there is no value for 'none'. A bot
-     *     whose administrator never touched the setting sends nothing here, which is exactly what every
-     *     call sent before it existed, so the PBX needs no new case.
+     * Liveness check of the service by the client: `\{"type": "healthcheck"\}`, with no envelope. The service sends no answer.
      * @public
      */
-    backgroundSound?: string | undefined;
+    healthcheck?: VoiceHealthcheckCommand | undefined;
     /**
-     * Negotiated format of caller `audio` chunks the client should send (AUDIO mode).
+     * A transcription chunk of caller speech. A client that does its own speech-to-text sends it. When the service does the speech-to-text, the service sends it to the client.
      * @public
      */
-    inputAudio?: VoiceAudioFormat | undefined;
+    transcription?: VoiceTranscriptionEvent | undefined;
     /**
-     * Format of the synthesized `play` frames the server will send when it performs TTS.
+     * A transcription chunk of the agent speech, as the caller hears it. It does not start a turn.
      * @public
      */
-    outputAudio?: VoiceAudioFormat | undefined;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface SessionStartedWebsocketEvent {
-    type: string;
+    agentTranscription?: VoiceAgentTranscriptionEvent | undefined;
     /**
-     * Emitted once right after the handshake to confirm the session and the agent it is bound to. Sent only when the client advertises protocol version >= 1.
+     * A final caller utterance. It starts an agent turn.
      * @public
      */
-    event: SessionStartedEvent;
-}
-/**
- * @public
- */
-export interface StopCommandData {
+    reply?: VoiceReplyEvent | undefined;
     /**
-     * Discriminator for the inner payload of a `command` frame.
+     * Progress of the playback of a reply on the client.
      * @public
      */
-    type: VoiceCommandType;
-}
-/**
- * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
- * @public
- */
-export interface StopWebsocketCommand {
-    type: string;
-    command: StopCommandData;
+    playback?: VoicePlaybackEvent | undefined;
+    /**
+     * The caller spoke over a reply while the client played it.
+     * @public
+     */
+    interruption?: VoiceInterruptionEvent | undefined;
+    /**
+     * The call or the conversation is complete, for example because the caller hung up. The service ends the session. The service also accepts `type` `ended` with the same meaning.
+     * @public
+     */
+    complete?: VoiceCompleteEvent | undefined;
 }
 /**
- * A transcription chunk of caller speech, provided by a client that does its own speech-to-text (PBX / TRANSCRIPTION mode), or emitted by the service (AUDIO mode) and echoed to the client.
+ * Parameters supplied on the WebSocket connect (upgrade) request, as query-string parameters.
  * @public
  */
-export interface TranscriptionEvent {
+export interface VoiceConnectParams {
     /**
-     * Discriminator for the inner payload of an `event` frame.
+     * Protocol version advertised by the client (query parameter `v`). 0 (or absent) is the legacy PBX protocol; >= 1 enables the sessionStarted event; >= 2 additionally means the client WAITS for `started` before opening its recognition stream and honours the `speech` payload on it; >= 3 additionally means the client connects again with `resume` after close code `1001`, so the service can move the call to another server while it runs.
      * @public
      */
-    type: VoiceEventType;
+    protocolVersion?: number | undefined;
     /**
-     * Identifier of a transcription chunk within a session.
+     * Caller-supplied session id, unique per connection.
      * @public
      */
-    id: string;
-    text: string;
+    sessionId: string;
     /**
-     * Transcription confidence score (accuracy level) in range from 0.0 to 1.0. Value -1.0 indicates confidence is unavailable.
+     * True when this connection attaches the running session `sessionId` again, after a close with close code `1001` or a lost connection. The session keeps its history and does not play the greeting again.
      * @public
      */
-    confidence?: number | undefined;
+    resume?: boolean | undefined;
     /**
-     * True when this is the final transcript for the utterance rather than an interim hypothesis. An absent value is read as final: a client that does not distinguish interim results sends none.
+     * Agent to run on this session.
      * @public
      */
-    final?: boolean | undefined;
+    agentId: string;
     /**
-     * Language the recogniser identified for this chunk. Recorded in the trace only — it drives no behavior, so an operator can see what the recogniser thought during an utterance that never became a turn.
+     * Published agent version to run. Mutually exclusive with agentDraft. When neither is set, the agent's deployment configuration selects the version. The PBX sends it only for a preview number (`<agentId>-vN`, WMS-27502); a customer call never carries it.
      * @public
      */
-    language?: string | undefined;
-    time: number;
-}
-/**
- * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
- * @public
- */
-export interface TranscriptionWebsocketEvent {
-    type: string;
+    agentVersion?: number | undefined;
     /**
-     * A transcription chunk of caller speech, provided by a client that does its own speech-to-text (PBX / TRANSCRIPTION mode), or emitted by the service (AUDIO mode) and echoed to the client.
+     * Run the agent's current draft (for testing). Mutually exclusive with agentVersion.
      * @public
      */
-    event: TranscriptionEvent;
-}
-/**
- * Transfer the call to a PBX destination and terminate the session. Honoured on PBX clients.
- * @public
- */
-export interface TransferCommandData {
-    context: string;
-    extension: string;
+    agentDraft?: boolean | undefined;
     /**
-     * Discriminator for the inner payload of a `command` frame.
+     * PBX call id, when the session is bridged to a call.
      * @public
      */
-    type: VoiceCommandType;
-}
-/**
- * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
- * @public
- */
-export interface TransferWebsocketCommand {
-    type: string;
+    callId?: string | undefined;
     /**
-     * Transfer the call to a PBX destination and terminate the session. Honoured on PBX clients.
+     * Leg of the PBX call inside `callId` (pbx-data-engine's flow index). Kept on the session as metadata beside `callId` and `dialedNumber`; never interpreted by the runtime.
      * @public
      */
-    command: TransferCommandData;
-}
-/**
- * @public
- */
-export interface DescribeVoiceProtocolOutput {
+    flowIndex?: number | undefined;
+    /**
+     * The number the caller dialled to reach the agent. Kept on the session as metadata beside `callId` and `flowIndex`.
+     * @public
+     */
+    dialedNumber?: string | undefined;
     /**
-     * Parameters supplied on the WebSocket connect (upgrade) request, as query-string parameters.
+     * BCP-47 language of the conversation, e.g. 'en-US'.
      * @public
      */
-    connectParams?: VoiceConnectParams | undefined;
+    language?: string | undefined;
+    callerName?: string | undefined;
+    callerNumber?: string | undefined;
+    callerEmail?: string | undefined;
+    callerCompany?: string | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * Override the TTS voice for this session, as a voice URI, e.g. 'elevenlabs://<voiceId>?apiKey=...'. Defaults to the agent's configured voice.
      * @public
      */
-    audioWebsocketEvent?: AudioWebsocketEvent | undefined;
+    voice?: string | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * First message spoken to the caller. Overrides the agent's configured greeting.
      * @public
      */
-    transcriptionWebsocketEvent?: TranscriptionWebsocketEvent | undefined;
+    initialMessage?: string | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * Base64-encoded JSON object of agent variables for this session, e.g. base64('\{"orderId":"123"\}').
      * @public
      */
-    agentTranscriptionWebsocketEvent?: AgentTranscriptionWebsocketEvent | undefined;
+    variables?: string | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * How caller speech reaches the service. Absent: `transcription`. The service refuses another value with HTTP 400.
      * @public
      */
-    replyWebsocketEvent?: ReplyWebsocketEvent | undefined;
+    inputMode?: VoiceInputMode | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * Owner user of the session. Absent: no user. The service ignores it for a PBX token with `callId`.
      * @public
      */
-    playbackWebsocketEvent?: PlaybackWebsocketEvent | undefined;
+    user?: string | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * Tenant of the session. Required for a service token, which carries no tenant. For another token a value different from the token tenant gets HTTP 401.
      * @public
      */
-    interruptionWebsocketEvent?: InterruptionWebsocketEvent | undefined;
+    company?: string | undefined;
+}
+/**
+ * One example of each message that the server sends on the text socket. An event goes in `\{"type": "event", "event": \{...\}\}`, a command in `\{"type": "command", "command": \{...\}\}`.
+ * @public
+ */
+export interface VoiceTextServerFrames {
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * The session is ready. Sent once after the handshake, on the PBX voice socket only when the client sends protocol version 1 or more.
      * @public
      */
-    completeWebsocketEvent?: CompleteWebsocketEvent | undefined;
+    started?: VoiceSessionStartedEvent | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * The session ended. This is the last event of the session.
      * @public
      */
-    controlWebsocketEvent?: ControlWebsocketEvent | undefined;
+    ended?: VoiceSessionEndedEvent | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * An error on the connection. The session continues, unless a `ended` event follows.
      * @public
      */
-    sessionStartedWebsocketEvent?: SessionStartedWebsocketEvent | undefined;
+    error?: VoiceErrorEvent | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * Play a reply to the caller. When `audio` is present, the service sends the audio of the reply in `play` commands. When `audio` is absent, the client shows `text` or synthesizes it itself.
      * @public
      */
-    sessionEndedWebsocketEvent?: SessionEndedWebsocketEvent | undefined;
+    say?: VoiceSayCommand | undefined;
     /**
-     * Envelope of a WebSocket event frame: `\{ "type": "event", "event": \{ ... \} \}`.
+     * A chunk of synthesized audio of a reply. `data` is base64 of PCM audio at the sample rate of the `say` command. An empty `data` marks the end of the stream.
      * @public
      */
-    errorWebsocketEvent?: ErrorWebsocketEvent | undefined;
+    play?: VoicePlayCommand | undefined;
     /**
-     * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
+     * Stop the audio of the agent.
      * @public
      */
-    sayWebsocketCommand?: SayWebsocketCommand | undefined;
+    stop?: VoiceStopCommand | undefined;
     /**
-     * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
+     * End the session and hang up the call.
      * @public
      */
-    playWebsocketCommand?: PlayWebsocketCommand | undefined;
+    hangup?: VoiceHangupCommand | undefined;
     /**
-     * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
+     * Transfer the call to a PBX destination and end the session. Only a PBX client does the transfer.
      * @public
      */
-    stopWebsocketCommand?: StopWebsocketCommand | undefined;
+    transfer?: VoiceTransferCommand | undefined;
+}
+/**
+ * The connect parameters and the messages of the text socket `/v2/voice/ws`.
+ * @public
+ */
+export interface VoiceTextProtocolFrames {
     /**
-     * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
+     * Query parameters of the upgrade request.
      * @public
      */
-    hangupWebsocketCommand?: HangupWebsocketCommand | undefined;
+    connectParams?: VoiceConnectParams | undefined;
     /**
-     * Envelope of a WebSocket command frame: `\{ "type": "command", "command": \{ ... \} \}`.
+     * Messages that the client sends.
      * @public
      */
-    transferWebsocketCommand?: TransferWebsocketCommand | undefined;
+    client?: VoiceTextClientFrames | undefined;
     /**
-     * One event of the StreamCallEvents SSE stream (`event:` is the member name, `data:` its JSON). Listed here so the generated SDKs carry the union; the stream body itself is an opaque byte stream.
+     * Messages that the server sends.
      * @public
      */
-    callEvent?: VoiceCallEvent | undefined;
+    server?: VoiceTextServerFrames | undefined;
+}
+/**
+ * @public
+ */
+export interface DescribeVoiceProtocolOutput {
+    /**
+     * Text socket `/v2/voice/ws`.
+     * @public
+     */
+    textSocket?: VoiceTextProtocolFrames | undefined;
+    /**
+     * Call socket `/v2/voice/calls/ws`.
+     * @public
+     */
+    callSocket?: VoiceCallProtocolFrames | undefined;
     /**
-     * Messages of the PBX media socket `/v2/voice/media/ws` (subprotocol `wilma-media-v2`): one example of each. The binary audio messages are base64 here and raw bytes on the socket.
+     * One event of the StreamCallEvents SSE stream (`event:` is the member name, `data:` its JSON). The stream body itself is an opaque byte stream.
      * @public
      */
-    media?: MediaProtocolFrames | undefined;
+    callEvent?: VoiceCallEvent | undefined;
 }
 /**
  * @public
@@ -1936,13 +1709,17 @@ export interface ListVoicesOutput {
     voices: Voice[];
 }
 /**
- * Terminate the session and hang up the call.
+ * End the session and hang up the call.
  * @public
  */
 export interface SendHangupInput {
+    /**
+     * Why the session ended.
+     * @public
+     */
     reason?: string | undefined;
     /**
-     * Caller-facing voice session identifier, unique per connection. Supplied by the client at connect (the PBX call/session id, or a client-generated id for web sessions).
+     * Id of the voice session.
      * @public
      */
     sessionId: string;
@@ -1953,23 +1730,15 @@ export interface SendHangupInput {
 export interface SendHangupOutput {
 }
 /**
- * A chunk of synthesized reply audio to play. `data` is base64, PCM-encoded at the sample rate declared in the matching `say` command; an empty `data` marks the end of the stream.
+ * A chunk of synthesized audio of a reply. `data` is base64 of PCM audio at the sample rate of the `say` command. An empty `data` marks the end of the stream.
  * @public
  */
 export interface SendPlayInput {
-    /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
-     * @public
-     */
     replyId: string;
-    /**
-     * Identifies one synthesized audio stream within a session; carried by a `say` command and the `play` frames that render it.
-     * @public
-     */
     streamId: string;
     data: string;
     /**
-     * Caller-facing voice session identifier, unique per connection. Supplied by the client at connect (the PBX call/session id, or a client-generated id for web sessions).
+     * Id of the voice session.
      * @public
      */
     sessionId: string;
@@ -1980,40 +1749,32 @@ export interface SendPlayInput {
 export interface SendPlayOutput {
 }
 /**
- * Play a reply to the caller. When `audio` is absent the client performs text-to-speech from `text`; when present the server streams synthesized `play` frames for this reply.
+ * Play a reply to the caller. When `audio` is present, the service sends the audio of the reply in `play` commands. When `audio` is absent, the client shows `text` or synthesizes it itself.
  * @public
  */
 export interface SendSayInput {
-    text: string;
     /**
-     * Correlates a caller phrase, the agent reply it produced, and the playback/interruption of that reply.
+     * Text of the reply.
      * @public
      */
-    replyId: string;
+    text: string;
     /**
-     * Whether the caller may barge in over this reply.
+     * Id of the reply.
      * @public
      */
-    interruptible?: boolean | undefined;
+    replyId: string;
     /**
-     * Bare Amazon Polly voice id, e.g. 'Joanna' — NOT a URI. The PBX consumes this verbatim as the
-     *     AMI `Voice=` action parameter, so it takes exactly what Polly names the voice, with the
-     *     `amazon_polly://` scheme and any query parameters already stripped by whoever builds this
-     *     command out of the agent's stored `VoiceUri`.
-     *
-     *     Set only for a provider whose synthesis runs ON THE PBX (Amazon Polly); ElevenLabs and Google
-     *     are synthesized in the cloud and streamed as PCM through `play` instead, and leave this
-     *     absent.
+     * True when the caller can interrupt the reply.
      * @public
      */
-    voice?: string | undefined;
+    interruptible?: boolean | undefined;
     /**
-     * Audio-stream parameters when the server synthesizes the reply itself (TTS provider configured on the agent): the `say` is followed by a stream of `play` frames on this stream id.
+     * Audio of a reply that the service synthesizes. `play` commands with this stream id follow the `say` command.
      * @public
      */
-    audio?: SayCommandAudioOptions | undefined;
+    audio?: VoiceSayCommandAudioOptions | undefined;
     /**
-     * Caller-facing voice session identifier, unique per connection. Supplied by the client at connect (the PBX call/session id, or a client-generated id for web sessions).
+     * Id of the voice session.
      * @public
      */
     sessionId: string;
@@ -2028,7 +1789,7 @@ export interface SendSayOutput {
  */
 export interface SendStopInput {
     /**
-     * Caller-facing voice session identifier, unique per connection. Supplied by the client at connect (the PBX call/session id, or a client-generated id for web sessions).
+     * Id of the voice session.
      * @public
      */
     sessionId: string;
@@ -2039,14 +1800,22 @@ export interface SendStopInput {
 export interface SendStopOutput {
 }
 /**
- * Transfer the call to a PBX destination and terminate the session. Honoured on PBX clients.
+ * Transfer the call to a PBX destination and end the session. Only a PBX client does the transfer.
  * @public
  */
 export interface SendTransferInput {
+    /**
+     * Dialplan context of the destination.
+     * @public
+     */
     context: string;
+    /**
+     * Extension of the destination.
+     * @public
+     */
     extension: string;
     /**
-     * Caller-facing voice session identifier, unique per connection. Supplied by the client at connect (the PBX call/session id, or a client-generated id for web sessions).
+     * Id of the voice session.
      * @public
      */
     sessionId: string;