telnyx 7.25.0 → 7.27.0

package/src/resources/ai/assistants/assistants.ts

@@ -3,6 +3,7 @@
 import { APIResource } from '../../../core/resource';
 import * as AssistantsAPI from './assistants';
 import * as Shared from '../../shared';
+import * as AvatarsAPI from '../avatars';
 import * as ChatAPI from '../chat';
 import * as ToolsAPI from '../tools';
 import * as CanaryDeploysAPI from './canary-deploys';
@@ -209,7 +210,10 @@ export class Assistants extends APIResource {
   /**
    * Updates the specified AI assistant's attributes and returns the updated
    * assistant. The request can also control how the change is promoted across
-   * assistant versions.
+   * assistant versions. Every update saves a new version, and an assistant keeps at
+   * most 10 versions: once it has more, the oldest versions are permanently deleted,
+   * except the main version and any version referenced by a traffic distribution
+   * rule.
    *
    * @example
    * ```ts
@@ -647,6 +651,7 @@ export type AssistantTool =
   | AssistantTool.SendDtmf
   | AssistantTool.SendMessage
   | AssistantTool.SkipTurn
+  | AssistantTool.Vision
   | AssistantTool.Pay
   | AssistantTool.UpdateDynamicVariables;
 
@@ -829,6 +834,14 @@ export namespace AssistantTool {
        */
       targets: Array<Transfer.TargetsList> | string;
 
+      /**
+       * Plays a short beep once the transfer completes and the calls are bridged, so the
+       * transfer destination knows the caller is now on the line. The warm transfer
+       * message plays before the bridge; this marks the moment the caller actually
+       * joins. Works with and without `warm_transfer_acceptance`.
+       */
+      bridge_beep?: Transfer.BridgeBeep;
+
       /**
        * Custom headers to be added to the SIP INVITE for the transfer command.
        */
@@ -931,6 +944,25 @@ export namespace AssistantTool {
         sip_auth_username?: string;
       }
 
+      /**
+       * Plays a short beep once the transfer completes and the calls are bridged, so the
+       * transfer destination knows the caller is now on the line. The warm transfer
+       * message plays before the bridge; this marks the moment the caller actually
+       * joins. Works with and without `warm_transfer_acceptance`.
+       */
+      export interface BridgeBeep {
+        /**
+         * Whether to play the beep when the calls are bridged.
+         */
+        enabled?: boolean;
+
+        /**
+         * Who hears the beep. `self` plays it to the transfer destination only; `both`
+         * plays it to the destination and the original caller.
+         */
+        target_legs?: 'self' | 'both';
+      }
+
       export interface CustomHeader {
         name?: string;
 
@@ -1095,6 +1127,32 @@ export namespace AssistantTool {
          * `shared` the exchange stays in the conversation like any other messages.
          */
         end_user_target_context_mode?: 'private' | 'shared';
+
+        /**
+         * Fail closed when the acceptance consult cannot run. Off by default: if the
+         * consult cannot be set up for any reason — including where acceptance is
+         * unavailable at all (`gather_using_ai`, Conversation Relay or web calls, a
+         * conference, or additional invited participants) — the transfer degrades to a
+         * plain warm transfer (the warm transfer message is played to the destination as
+         * audio and the calls are bridged anyway). Set to true to treat those paths as a
+         * decline instead: the destination is disconnected without being asked, the
+         * assistant returns to the caller, and no bridge happens without an acceptance.
+         * Only consult-setup failures are affected; the destination must still explicitly
+         * accept on every consult that does run.
+         */
+        require_acceptance?: boolean;
+
+        /**
+         * How long, in seconds, the destination has to accept the transfer, counted from
+         * when the private consult starts (right after the destination answers, or after
+         * `warm_message_delay_ms`). The caller keeps hearing ringback the whole time. If
+         * no acceptance arrives in time, the destination is hung up and the assistant
+         * returns to the caller (the transfer fails as a timeout). Raise it for
+         * destinations that answer before a human is on the line, such as a call-center
+         * queue/ACD or an IVR playing hold audio; keep it as short as practical, since the
+         * caller waits on ringback until it expires.
+         */
+        timeout_secs?: number;
       }
     }
   }
@@ -1368,6 +1426,30 @@ export namespace AssistantTool {
     }
   }
 
+  /**
+   * (BETA) Lets the assistant look at recent image frames the client sent into the
+   * conversation (for example webcam or screen-share frames). Frames are sent with
+   * `conversation.item.create` and `item.buffer: true`; they never trigger a
+   * response on their own. The assistant decides when to look and which frames it
+   * needs. Requires a model that supports image input.
+   */
+  export interface Vision {
+    type: 'vision';
+
+    /**
+     * Whether this tool comes from the shared Tools Library. Responses merge shared
+     * tools into `tools` with `shared: true`; inline tools carry `shared: false`.
+     * Read-only: set by the server, not accepted in requests. When updating an
+     * assistant, omit `shared: true` tools from the request `tools` array and manage
+     * them through `tool_ids` instead — re-sending their definitions creates an inline
+     * duplicate (rejected with error code 10015 when the type allows only one instance
+     * per assistant).
+     */
+    shared?: boolean;
+
+    vision?: ToolsAPI.VisionToolParams;
+  }
+
   /**
    * (BETA) The pay tool allows the assistant to collect card payments from the
    * caller via DTMF during the conversation. Recording is automatically paused while
@@ -1440,6 +1522,35 @@ export interface AudioVisualizerConfig {
  */
 export type AuthenticationMethod = 'token' | 'certificate';
 
+/**
+ * **Beta:** Configuration for the Telnyx open-source avatar provider. Use a stock
+ * avatar or an organization-owned custom avatar returned by the Avatars API.
+ * Discover the available stock catalog with GET /v2/ai/avatars and
+ * filter[avatar_type]=stock; follow meta.total_pages to load every page.
+ */
+export interface AvatarSettingsRequest {
+  /**
+   * The Telnyx stock avatar or organization-owned custom avatar to render.
+   */
+  avatar_id: AvatarsAPI.AvatarID;
+
+  /**
+   * The background to render.
+   */
+  background_id: 'office' | 'studio' | 'reception' | 'illustrated_workspace';
+
+  /**
+   * Required avatar model identifier, separate from `avatar_id`. The supported value
+   * is `foxtrot-1`. Omission, null, and unsupported model identifiers are rejected.
+   */
+  model: 'foxtrot-1';
+
+  /**
+   * Selects the Telnyx open-source avatar provider.
+   */
+  provider: 'telnyx';
+}
+
 /**
  * Combine sub-expressions with a logical operator (`and` / `or` / `not`).
  *
@@ -2112,7 +2223,11 @@ export interface FlowNodeReq {
 
   /**
    * Per-node transcription override (model/language/region). Unset fields cascade
-   * from the assistant-level transcription.
+   * from the assistant-level transcription. A node that sets `model`,
+   * `fallback_models`, or `challenger` doesn't inherit the assistant's
+   * `fallback_models` or `challenger`; it uses only the ones it sets. Otherwise it
+   * inherits them, and they must fit the model and language the node runs; a change
+   * they no longer fit is rejected.
    */
   transcription?: TranscriptionSettings;
 
@@ -2189,6 +2304,14 @@ export interface InferenceEmbedding {
    */
   a2a_agents?: Array<AssistantA2AAgent>;
 
+  /**
+   * **Beta:** Configuration for the Telnyx open-source avatar provider. Use a stock
+   * avatar or an organization-owned custom avatar returned by the Avatars API.
+   * Discover the available stock catalog with GET /v2/ai/avatars and
+   * filter[avatar_type]=stock; follow meta.total_pages to load every page.
+   */
+  avatar_settings?: TelnyxAvatarSettings | null;
+
   /**
    * Conversation flow as returned by the API.
    */
@@ -2426,8 +2549,8 @@ export interface InferenceEmbeddingVoiceSettings {
    * preview it before saving.
    */
   background_audio?:
-    | InferenceEmbeddingVoiceSettings.UnionMember0
-    | InferenceEmbeddingVoiceSettings.UnionMember1
+    | InferenceEmbeddingVoiceSettings.BackgroundAudioUnionMember0
+    | InferenceEmbeddingVoiceSettings.BackgroundAudioUnionMember1
     | InferenceEmbeddingVoiceSettings.UnionMember2;
 
   /**
@@ -2512,6 +2635,17 @@ export interface InferenceEmbeddingVoiceSettings {
    */
   temperature?: number;
 
+  /**
+   * Optional sound played to the caller while the assistant runs a tool (for example
+   * a webhook or MCP tool). It replaces the background audio until the tool
+   * responds, then the background audio resumes. Defaults to null, which plays no
+   * sound; send `null` on update to clear it. Not played in conference calls.
+   */
+  tool_call_audio?:
+    | InferenceEmbeddingVoiceSettings.ToolCallAudioUnionMember0
+    | InferenceEmbeddingVoiceSettings.ToolCallAudioUnionMember1
+    | null;
+
   /**
    * Amplifies similarity to the original speaker voice. Increases computational load
    * and latency slightly. Only applicable when using ElevenLabs.
@@ -2528,7 +2662,7 @@ export interface InferenceEmbeddingVoiceSettings {
 }
 
 export namespace InferenceEmbeddingVoiceSettings {
-  export interface UnionMember0 {
+  export interface BackgroundAudioUnionMember0 {
     /**
      * Select from predefined media options.
      */
@@ -2546,7 +2680,7 @@ export namespace InferenceEmbeddingVoiceSettings {
     volume?: number;
   }
 
-  export interface UnionMember1 {
+  export interface BackgroundAudioUnionMember1 {
     /**
      * Provide a direct URL to an MP3 file. The audio will loop during the call.
      */
@@ -2571,6 +2705,36 @@ export namespace InferenceEmbeddingVoiceSettings {
      */
     value: string;
   }
+
+  export interface ToolCallAudioUnionMember0 {
+    /**
+     * Select from predefined media options.
+     */
+    type: 'predefined_media';
+
+    /**
+     * The predefined media to use. `typing` plays a keyboard typing sound.
+     */
+    value: 'typing';
+
+    /**
+     * Volume level for the predefined media. Supports values from 0.0 to 1.0 in 0.1
+     * increments. `0.0` plays no sound.
+     */
+    volume?: number;
+  }
+
+  export interface ToolCallAudioUnionMember1 {
+    /**
+     * Provide a direct URL to an MP3 file. The audio loops while the tool runs.
+     */
+    type: 'media_url';
+
+    /**
+     * HTTPS URL to an MP3 file.
+     */
+    value: string;
+  }
 }
 
 export interface InferenceEmbeddingWebhookToolParams {
@@ -2995,6 +3159,34 @@ export interface PrivacySettings {
    * border. Defaults to false.
    */
   in_transit_data_locality?: boolean;
+
+  /**
+   * Controls redaction of personally identifiable information (PII) from the
+   * conversation history that is stored for this assistant. `disabled` (default)
+   * stores content as-is. `redact` replaces detected PII with `<PII:type/>`
+   * placeholders (e.g. `<PII:name/>`, `<PII:phone/>`) as the stored copies are
+   * written: message content (user, assistant, and tool-result messages) and
+   * tool-call arguments are redacted when written to conversation history, so the
+   * stored transcript, its read-back via the history API, and any insights generated
+   * from the stored messages contain the placeholders instead of the PII; response
+   * bodies returned by webhook tools are redacted too, as stored in the webhook logs
+   * — a per-conversation record separate from the transcript. In the Mission Control
+   * Portal this field appears as a "PII Redaction" on/off toggle, where Enabled
+   * corresponds to `redact`. Redaction runs when the stored copy is written, after
+   * the LLM has generated its reply: the live conversation is not redacted, the
+   * assistant's tools keep receiving the full (unredacted) values during the call,
+   * and the LLM still sees full content in its context window. Call recordings
+   * (including any recording layout configured under telephony settings) and the
+   * payloads of messages/webhooks delivered during the call are never redacted by
+   * this setting. The set of detected PII types (names, street addresses, phone
+   * numbers, email addresses, government identifiers like SSN and driver's license
+   * numbers, credit card and bank account numbers, IP addresses, dates of birth, and
+   * other identifiers unique to a person) is fixed by Telnyx and cannot be
+   * configured or extended per assistant. Assistants with `redact` cannot be used
+   * for SMS/WhatsApp messaging (that combination is rejected with a 400 error);
+   * voice and web chat are unaffected.
+   */
+  pii_redaction?: 'disabled' | 'redact';
 }
 
 /**
@@ -3124,6 +3316,16 @@ export interface StartSpeakingPlan {
 }
 
 export interface TelephonySettings {
+  /**
+   * When enabled, the `messages` array of the `call.conversation.ended` webhook
+   * contains the full conversation history in order: user and assistant turns,
+   * assistant tool calls (`tool_calls`), tool results (`role: tool` with
+   * `tool_call_id`), and informational `system` and `developer` entries. The
+   * assistant's instructions are never included. When disabled, `messages` contains
+   * only user and assistant text turns.
+   */
+  conversation_ended_full_history?: boolean;
+
   /**
    * Default Texml App used for voice calls with your assistant. This will be created
    * automatically on assistant creation.
@@ -3189,9 +3391,16 @@ export interface TelephonySettings {
 
   /**
    * Maximum duration in seconds for the AI assistant to participate on the call.
-   * When this limit is reached the assistant will be stopped. This limit does not
-   * apply to portions of a call without an active assistant (for instance, a call
-   * transferred to a human representative).
+   * When this limit is reached the assistant will be stopped. Stopping the assistant
+   * does not itself hang up the call, and what happens next depends on how the
+   * assistant is running. On Call Control (ai_assistant_start) the call leg stays
+   * connected and continues to be billed until it is hung up (for example with the
+   * Hangup call command) or until the call ends for another reason. On TeXML, the
+   * assistant's `<Connect><AIAssistant>` instruction ends; in the default TeXML
+   * setup for assistants no instructions follow it, so the call is hung up. Your
+   * application receives a call.conversation.ended webhook when the assistant stops.
+   * This limit does not apply to portions of a call without an active assistant (for
+   * instance, a call transferred to a human representative).
    */
   time_limit_secs?: number;
 
@@ -3204,9 +3413,16 @@ export interface TelephonySettings {
 
   /**
    * Maximum duration in seconds of end user silence on the call. When this limit is
-   * reached the assistant will be stopped. This limit does not apply to portions of
-   * a call without an active assistant (for instance, a call transferred to a human
-   * representative).
+   * reached the assistant will be stopped. Stopping the assistant does not itself
+   * hang up the call, and what happens next depends on how the assistant is running.
+   * On Call Control (ai_assistant_start) the call leg stays connected and continues
+   * to be billed until it is hung up (for example with the Hangup call command) or
+   * until the call ends for another reason. On TeXML, the assistant's
+   * `<Connect><AIAssistant>` instruction ends; in the default TeXML setup for
+   * assistants no instructions follow it, so the call is hung up. Your application
+   * receives a call.conversation.ended webhook when the assistant stops. This limit
+   * does not apply to portions of a call without an active assistant (for instance,
+   * a call transferred to a human representative).
    */
   user_idle_timeout_secs?: number;
 
@@ -3349,6 +3565,35 @@ export namespace TelephonySettings {
   }
 }
 
+/**
+ * **Beta:** Configuration for the Telnyx open-source avatar provider. Use a stock
+ * avatar or an organization-owned custom avatar returned by the Avatars API.
+ * Discover the available stock catalog with GET /v2/ai/avatars and
+ * filter[avatar_type]=stock; follow meta.total_pages to load every page.
+ */
+export interface TelnyxAvatarSettings {
+  /**
+   * The Telnyx stock avatar or organization-owned custom avatar to render.
+   */
+  avatar_id: AvatarsAPI.AvatarID;
+
+  /**
+   * The background to render.
+   */
+  background_id: 'office' | 'studio' | 'reception' | 'illustrated_workspace';
+
+  /**
+   * Required avatar model identifier, separate from `avatar_id`. The supported value
+   * is `foxtrot-1`. Omission, null, and unsupported model identifiers are rejected.
+   */
+  model: 'foxtrot-1';
+
+  /**
+   * Selects the Telnyx open-source avatar provider.
+   */
+  provider: 'telnyx';
+}
+
 /**
  * A standalone tool step in a conversation flow, as returned by the API.
  */
@@ -3489,6 +3734,24 @@ export interface TranscriptionSettings {
    */
   api_key_ref?: string;
 
+  /**
+   * A second speech-to-text model that transcribes alongside `transcription.model`,
+   * and the rule that decides which transcript the assistant uses.
+   */
+  challenger?: TranscriptionSettings.Challenger | null;
+
+  /**
+   * Up to 3 streaming models that take over transcription, in this order, when the
+   * model in use fails, at the start of a call or mid-call. `model` must be a
+   * streaming model too, and must support `language` alongside other models. On
+   * update, a list replaces the stored one: omit the field to keep the stored list,
+   * or send `null` or `[]` to remove it. When an update changes `model` or
+   * `language`, stored fallbacks that no longer fit are removed without an error.
+   * Can't be combined with `challenger`, the language booster; to replace a stored
+   * language booster, send `challenger: null` in the same request.
+   */
+  fallback_models?: Array<TranscriptionSettings.FallbackModel> | null;
+
   /**
    * The language of the audio to be transcribed. If not set, or if set to `auto`,
    * supported models will automatically detect the language. For `deepgram/flux`,
@@ -3506,7 +3769,12 @@ export interface TranscriptionSettings {
    * `nl`, `en`, `fr`, `fy`, `de`, `it`, `pl`, `pt`, `es`, and `sv` to fix the
    * transcription language. For `cohere/ar-stt`, supported values are `ar` and `en`;
    * unlike other models, this model does not auto-detect and defaults to `ar` when
-   * `language` is omitted.
+   * `language` is omitted. For `speechmatics/linden-1`, there is no auto-detect; set
+   * a code from Speechmatics' Realtime language list (67 language codes, including
+   * `es-bilingual-en` for Spanish and English bilingual, and the spelling and script
+   * variants `en-US`, `en-GB`, `en-AU`, `cmn-Hans`, and `cmn-Hant`; `en` alone uses
+   * the default spelling); unset defaults to `en`, while an explicit `auto`,
+   * `auto_beta`, or `multi` is rejected with a 400.
    */
   language?: string;
 
@@ -3517,6 +3785,12 @@ export interface TranscriptionSettings {
    * - `deepgram/flux` is optimized for turn-taking with multilingual language hints.
    * - `deepgram/nova-3` is multilingual with automatic language detection.
    * - `deepgram/nova-2` is Deepgram's previous-generation multilingual model.
+   * - `speechmatics/standard` is a Speechmatics model self-hosted by Telnyx; it
+   *   supports a fixed set of language packs (including bilingual packs such as
+   *   `ar_en`) and defaults to `en`.
+   * - `speechmatics/linden-1` (preview) is a vendor-hosted streaming model with no
+   *   auto-detect; set `language` from Speechmatics' Realtime language list
+   *   (defaults to `en`).
    * - `azure/fast` is a multilingual Azure transcription model.
    * - `assemblyai/universal-3-5-pro` is a multilingual streaming model with
    *   configurable turn detection. The legacy alias `assemblyai/universal-streaming`
@@ -3534,11 +3808,14 @@ export interface TranscriptionSettings {
    * - `reson8/turns` is a turn-based streaming model covering 10 European languages
    *   with automatic language detection.
    * - `cohere/ar-stt` is a non-streaming Arabic and English transcription model.
+   * - `telnyx/basira` is a non-streaming Arabic transcription model.
    */
   model?:
     | 'deepgram/flux'
     | 'deepgram/nova-3'
     | 'deepgram/nova-2'
+    | 'speechmatics/standard'
+    | 'speechmatics/linden-1'
     | 'azure/fast'
     | 'assemblyai/universal-3-5-pro'
     | 'assemblyai/universal-streaming'
@@ -3550,6 +3827,7 @@ export interface TranscriptionSettings {
     | 'humain/realtime'
     | 'reson8/turns'
     | 'cohere/ar-stt'
+    | 'telnyx/basira'
     | 'distil-whisper/distil-large-v2'
     | 'openai/whisper-large-v3-turbo';
 
@@ -3562,6 +3840,124 @@ export interface TranscriptionSettings {
   settings?: TranscriptionSettingsConfig;
 }
 
+export namespace TranscriptionSettings {
+  /**
+   * A second speech-to-text model that transcribes alongside `transcription.model`,
+   * and the rule that decides which transcript the assistant uses.
+   */
+  export interface Challenger {
+    /**
+     * The language booster's model. It must be the same kind of model as
+     * `transcription.model`: both streaming (`deepgram/flux`, `deepgram/nova-3`,
+     * `deepgram/nova-2`, `assemblyai/universal-3-5-pro` or its legacy alias
+     * `assemblyai/universal-streaming`, `xai/grok-stt`, `soniox/stt-rt-v4`,
+     * `soniox/stt-rt-v5`, `humain/realtime`, `reson8/turns`) or both non-streaming
+     * (`azure/fast`, `nvidia/parakeet-v3`, `omi-health/omi-med-stt-v1`,
+     * `cohere/ar-stt`, `distil-whisper/distil-large-v2`,
+     * `openai/whisper-large-v3-turbo`, `telnyx/basira`). It can be the same model as
+     * `transcription.model` on a different `language`.
+     */
+    model:
+      | 'deepgram/flux'
+      | 'deepgram/nova-3'
+      | 'deepgram/nova-2'
+      | 'azure/fast'
+      | 'assemblyai/universal-3-5-pro'
+      | 'assemblyai/universal-streaming'
+      | 'xai/grok-stt'
+      | 'soniox/stt-rt-v4'
+      | 'soniox/stt-rt-v5'
+      | 'nvidia/parakeet-v3'
+      | 'omi-health/omi-med-stt-v1'
+      | 'humain/realtime'
+      | 'reson8/turns'
+      | 'cohere/ar-stt'
+      | 'telnyx/basira'
+      | 'distil-whisper/distil-large-v2'
+      | 'openai/whisper-large-v3-turbo';
+
+    /**
+     * The language this model transcribes. Omit it or set it to `null` to use the
+     * language of `transcription.model`. The request is rejected when this model
+     * doesn't support the language it would run. It is also rejected when it would run
+     * the same model on the same language as `transcription.model`.
+     */
+    language?: string | null;
+
+    /**
+     * How the assistant picks the transcript it uses. The models are compared on how
+     * complete and confident their transcripts are, not on language, so the rules work
+     * best when both models understand the callers' language.
+     *
+     * - `best_turn` (default): both models transcribe the whole call. Each turn uses
+     *   the language booster's transcript only when it scores higher than the
+     *   transcript of `transcription.model` (clearly higher with non-streaming
+     *   models). With streaming models, `transcription.model` also decides when each
+     *   turn ends. Available for every pair.
+     * - `best_engine`: both models transcribe the first turns, then the call continues
+     *   alone on the model whose transcripts scored higher. If neither clearly leads,
+     *   `transcription.model` continues. Streaming models only.
+     * - `merge_words`: both models transcribe each utterance and their words are
+     *   merged, keeping Arabic and English spoken in the same sentence. Available only
+     *   for `telnyx/basira` with `cohere/ar-stt`, in either order. The pair runs on
+     *   the language that applies to `telnyx/basira` (its own, or that of
+     *   `transcription.model`), which must be Arabic (`ar` or an `ar-` locale),
+     *   `multi`, or `auto`.
+     */
+    rule?: 'best_turn' | 'best_engine' | 'merge_words';
+
+    /**
+     * Settings for the language booster, with the same fields and limits as
+     * `transcription.settings`. Fields that don't apply to this model's provider are
+     * dropped, and the provider's defaults fill in the rest. Omit it or set it to
+     * `null` to use the settings of `transcription.model` where they apply to this
+     * model.
+     */
+    settings?: AssistantsAPI.TranscriptionSettingsConfig | null;
+  }
+
+  /**
+   * A streaming speech-to-text model that takes over transcription when the model in
+   * use fails.
+   */
+  export interface FallbackModel {
+    /**
+     * The fallback model. It must be a streaming model other than
+     * `transcription.model` and the other fallbacks: `deepgram/flux`,
+     * `deepgram/nova-3`, `deepgram/nova-2`, `assemblyai/universal-3-5-pro` (or its
+     * legacy alias `assemblyai/universal-streaming`), `xai/grok-stt`,
+     * `soniox/stt-rt-v4`, `soniox/stt-rt-v5`, `humain/realtime`, or `reson8/turns`.
+     */
+    model:
+      | 'deepgram/flux'
+      | 'deepgram/nova-3'
+      | 'deepgram/nova-2'
+      | 'assemblyai/universal-3-5-pro'
+      | 'assemblyai/universal-streaming'
+      | 'xai/grok-stt'
+      | 'soniox/stt-rt-v4'
+      | 'soniox/stt-rt-v5'
+      | 'humain/realtime'
+      | 'reson8/turns';
+
+    /**
+     * The language the fallback transcribes. Omit it or set it to `null` to use the
+     * language of `transcription.model`. The request is rejected when the fallback
+     * model doesn't support the language it would run.
+     */
+    language?: string | null;
+
+    /**
+     * Settings for the fallback, with the same fields and limits as
+     * `transcription.settings`. Fields that don't apply to this model's provider are
+     * dropped, and the provider's defaults fill in the rest. Omit it or set it to
+     * `null` to use the settings of `transcription.model` where they apply to this
+     * model.
+     */
+    settings?: AssistantsAPI.TranscriptionSettingsConfig | null;
+  }
+}
+
 export interface TranscriptionSettingsConfig {
   /**
    * Available only for soniox/stt-rt-v4 and soniox/stt-rt-v5. A comma-separated list
@@ -3589,9 +3985,9 @@ export interface TranscriptionSettingsConfig {
   enable_endpoint_detection?: boolean;
 
   /**
-   * Available only for assemblyai/universal-3-5-pro (and its legacy alias
-   * assemblyai/universal-streaming). Confidence level required to trigger an end of
-   * turn. Higher values require more certainty before ending a turn.
+   * @deprecated Deprecated and ignored. Accepted for backwards compatibility.
+   * AssemblyAI turn detection is configured with min_turn_silence and
+   * max_turn_silence.
    */
   end_of_turn_confidence_threshold?: number;
 
@@ -3614,10 +4010,13 @@ export interface TranscriptionSettingsConfig {
   interim_results?: boolean;
 
   /**
-   * Available only for deepgram/nova-3 and deepgram/flux. A comma-separated list of
-   * key terms to boost for recognition during transcription. Helps improve accuracy
-   * for domain-specific terminology, proper nouns, or uncommon words. This field may
-   * be templated with
+   * A comma-separated list of key terms to boost for recognition during
+   * transcription, used by Deepgram, Reson8, Speechmatics Linden 1 and AssemblyAI
+   * models. Helps improve accuracy for domain-specific terminology, proper nouns, or
+   * uncommon words. At most 1000 characters in total. For
+   * assemblyai/universal-3-5-pro (and its legacy alias
+   * assemblyai/universal-streaming), also at most 100 terms of up to 50 characters
+   * each. This field may be templated with
    * [dynamic variables](https://developers.telnyx.com/docs/inference/ai-assistants/dynamic-variables)
    * using mustache syntax (e.g. `Telnyx,{{customer_name}},VoIP`). Variables are
    * resolved at call time before the value is sent to the speech-to-text engine.
@@ -3654,6 +4053,24 @@ export interface TranscriptionSettingsConfig {
 
   numerals?: boolean;
 
+  /**
+   * Available only for assemblyai/universal-3-5-pro (and its legacy alias
+   * assemblyai/universal-streaming). Context about the conversation (domain, names,
+   * terms) to improve transcription; describe the content, do not give instructions.
+   * Leave empty unless needed. At most 1750 characters.
+   */
+  prompt?: string;
+
+  /**
+   * Available only for speechmatics/linden-1. When true, filler words such as "um"
+   * and "uh" are removed from the transcript. Disfluency removal is supported only
+   * for `ar`, `cmn`, `da`, `de`, `el`, `en`, `es`, `fr`, `he`, `hi`, `hu`, `it`,
+   * `ja`, `nl`, `pl`, `pt`, and `ru`, determined by the base language (so `en-US`,
+   * `en-GB`, `en-AU`, `cmn-Hans`, and `cmn-Hant` are supported); for other
+   * languages, filler words remain in the transcript.
+   */
+  remove_disfluencies?: boolean;
+
   smart_format?: boolean;
 }
 
@@ -4162,6 +4579,14 @@ export interface AssistantCreateParams {
    */
   a2a_agents?: Array<AssistantA2AAgent>;
 
+  /**
+   * Body param: **Beta:** Configuration for the Telnyx open-source avatar provider.
+   * Use a stock avatar or an organization-owned custom avatar returned by the
+   * Avatars API. Discover the available stock catalog with GET /v2/ai/avatars and
+   * filter[avatar_type]=stock; follow meta.total_pages to load every page.
+   */
+  avatar_settings?: AvatarSettingsRequest | null;
+
   /**
    * Body param: Conversation flow as supplied by API clients (create / update).
    *
@@ -4448,6 +4873,14 @@ export interface AssistantUpdateParams {
    */
   a2a_agents?: Array<AssistantA2AAgent>;
 
+  /**
+   * **Beta:** Configuration for the Telnyx open-source avatar provider. Use a stock
+   * avatar or an organization-owned custom avatar returned by the Avatars API.
+   * Discover the available stock catalog with GET /v2/ai/avatars and
+   * filter[avatar_type]=stock; follow meta.total_pages to load every page.
+   */
+  avatar_settings?: AvatarSettingsRequest | null;
+
   /**
    * Conversation flow as supplied by API clients (create / update).
    *
@@ -4779,6 +5212,7 @@ export declare namespace Assistants {
     type AssistantsList as AssistantsList,
     type AudioVisualizerConfig as AudioVisualizerConfig,
     type AuthenticationMethod as AuthenticationMethod,
+    type AvatarSettingsRequest as AvatarSettingsRequest,
     type BooleanOpExpression as BooleanOpExpression,
     type ComparisonExpression as ComparisonExpression,
     type ConversationFlow as ConversationFlow,
@@ -4815,6 +5249,7 @@ export declare namespace Assistants {
     type SpeakNodeReq as SpeakNodeReq,
     type StartSpeakingPlan as StartSpeakingPlan,
     type TelephonySettings as TelephonySettings,
+    type TelnyxAvatarSettings as TelnyxAvatarSettings,
     type ToolNode as ToolNode,
     type ToolNodeReq as ToolNodeReq,
     type TranscriptionEndpointingPlan as TranscriptionEndpointingPlan,