feat(agentkit): add Engine v2.14 support to the Python SDK

.fern/metadata.json

@@ -14,5 +14,5 @@
     },
     "exclude_types_from_init_exports": true
   },
-  "originGitCommit": "26314a9d3199f5e135302c3c3645b6bbb8f556d4"
+  "originGitCommit": "0073fa0ae670f77e4987407b8cca1cf179d5f90b"
 }

changelog.md

@@ -4,6 +4,21 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/).
 
+## [Unreleased]
+
+### Added
+
+- **RTZR global STT** — Added the global-only `RtzrSTT` AgentKit vendor with typed credentials, recognition options, vendor catalog registration, and generated request serialization.
+- **Speak parameters** — Exposed the generated `SpeakConfig`/`ParametersSpeak` types through `SessionParams` so `parameters.speak.batch` can be configured with `Agent.with_parameters()`.
+
+### Changed
+
+- **Sarvam TTS v2.14 parameters** — Added `speech_sample_rate`, `enable_preprocessing`, `model`, and flattened `additional_params` support. The legacy `sample_rate` input remains supported as a deprecated alias.
+ - **Generated SDK alignment** — Updated AgentKit exports, global vendor routing, request validation, avatar sample-rate handling, documentation, and feature-focused regression tests for the latest `engine-v2.14` generated API.
+
+- **Gemini TTS production routing** — Synchronous and asynchronous `GeminiTTS` sessions, including raw configurations, now use the configured regional production endpoint without the `gemini-live` preview gate. The implementation is registered in the global TTS catalog, while package-root and historical preview imports remain aliases with the same v2.11.0 options, defaults, validation, optional `style`, and `tts.params` wire fields. Compatibility with the generated request schema no longer depends on preview routing.
+- **Gemini TTS generated schema support** — The production vendor now uses the generated Gemini TTS shape and accepts `additional_params` and `skip_patterns` while keeping preview-era calls compatible.
+
 ## [v2.12.0] — 2026-10-07
 
 ### Added

docs/concepts/vendors.md

@@ -60,6 +60,7 @@ Used with `agent.with_tts()`. Each TTS vendor produces audio at a specific sampl
 | `OpenAITTS` | OpenAI | `voice` for Agora-managed global `tts-1`; `api_key`, `model`, `base_url`, `voice` for BYOK | 24000 Hz (fixed) |
 | `CartesiaTTS` | Cartesia | `api_key`, `voice_id`, `model_id` | 8000–48000 Hz |
 | `GoogleTTS` | Google Cloud | `key`, `voice_name` | — |
+| `GeminiTTS` | Google Gemini TTS | `api_key`; model and voice have defaults | — |
 | `AmazonTTS` | Amazon Polly | `access_key`, `secret_key`, `region`, `voice_id`, `engine` | — |
 | `HumeAITTS` | Hume AI | `key`, `voice_id`, `provider` | — |
 | `RimeTTS` | Rime | `model_id`; BYOK also requires `key` and `speaker`, managed requires `base_url` | — |
@@ -125,6 +126,7 @@ top-level `asr.keywords`. Both vendors also accept `additional_params`, serializ
 | `AssemblyAISTT` | AssemblyAI | `api_key`, `language` |
 | `AresSTT` | Ares | — (all optional) |
 | `SarvamSTT` | Sarvam | `api_key`, `language` |
+| `RtzrSTT` | RTZR | `client_id`, `client_secret` |
 | `XaiSTT` | xAI | `api_key` |
 
 ### CN STT Vendors

docs/guides/avatars.md

@@ -86,6 +86,25 @@ session.say('Hello! I am your visual assistant.')
 session.stop()
 ```
 
+## Anam Avatar
+
+Anam supports portrait models such as Cara 4 through `avatar_model`. Set
+`video_width` and `video_height` together for a custom resolution, or omit both
+to use the model default:
+
+```python
+from agora_agent import AnamAvatar
+
+avatar = AnamAvatar(
+    api_key="your-anam-key",
+    avatar_id="your-avatar-id",
+    avatar_model="cara_mk4",
+    video_width=720,
+    video_height=1280,
+)
+agent = agent.with_avatar(avatar)
+```
+
 ## Generic Avatar
 
 `GenericAvatar` supports custom avatar providers. `agora_appid`, `agora_channel`, and `agora_token` are optional when using `AgentSession.start()`.

docs/guides/preview-endpoint.md

@@ -10,7 +10,7 @@ Some providers may be released through a preview gateway before their production
 and `AsyncAgentSession` detect registered preview providers from the resolved start request and route the entire
 session automatically.
 
-OpenAI GPT Live, Gemini STT, and Gemini Live have graduated to the production gateway and use the normal regional
+OpenAI GPT Live, Gemini STT, Gemini Live, and Gemini TTS have graduated to the production gateway and use the normal regional
 endpoint. Existing imports from `agora_agent.agentkit.preview` remain supported as compatibility aliases.
 
 ```python
@@ -24,7 +24,7 @@ session = (
 agent_id = session.start()
 ```
 
-Preview providers use the preview base URL and `agora-feature` gate. Sessions using `GeminiSTT`, `GeminiLive`, or GPT
+Preview providers use the preview base URL and `agora-feature` gate. Sessions using `GeminiSTT`, `GeminiLive`, `GeminiTTS`, or GPT
 Live use the client's normal production regional endpoint without that header.
 
 Use the single production `GeminiLive(api_key=..., model=...)` class with `with_mllm`. The model IDs are
@@ -69,23 +69,26 @@ Add routing tests for both synchronous and asynchronous sessions when registerin
 the preview base URL, exact feature header, all lifecycle requests, caller-header precedence, and that the original
 client remains configured for production.
 
-## Gemini 3.8 Flash TTS preview
+## Gemini 3.8 Flash TTS production migration
 
 `GeminiTTS` emits `tts.vendor = "gemini"` with `api_key`, `model`, `voice`,
 and optional `style` inside `tts.params`. It defaults to `gemini-3.8-flash-tts`
-and `Puck`. Model names are sent unchanged; there is no automatic fallback
-or model rewriting.
-Model strings remain open for preview rollout changes. Blank keys are rejected.
-
-AgentSession detects the TTS vendor from the resolved request body, including
-handwritten configs, and uses the existing preview host with
-`agora-feature: gemini-live` throughout the session lifecycle. Use the retained
-session for stop/say/interrupt; the shared client remains on its normal route.
-Gemini ASR alone still uses the production route. No sample-rate or avatar
-compatibility is assumed by this preview provider.
+and `Puck`. Model names are sent unchanged; there is no automatic fallback or
+model rewriting.
+Model strings remain open for future model IDs. Blank keys are rejected.
+
+Existing v2.11.0 calls now use the client's configured regional production
+endpoint throughout both synchronous and asynchronous session lifecycles.
+Handwritten Gemini TTS configs follow the same route, and no preview feature
+header is added. `GeminiTTS` and `GeminiTTSModels` now live in `agentkit.vendors.tts`
+and are registered in the global vendor catalog. Imports from the package root,
+`agora_agent.agentkit.preview`, and `agora_agent.agentkit.preview.gemini_tts`
+remain aliases to the same production objects. `skip_patterns` maps to the
+generated top-level TTS field. `additional_params` is merged into `tts.params`,
+with named options taking precedence. The provider does not expose a configurable sample rate.
 
 ```python
-from agora_agent.agentkit.preview import GeminiTTS, GeminiTTSModels
+from agora_agent import GeminiTTS, GeminiTTSModels
 
 agent.with_tts(GeminiTTS(
     api_key=google_api_key,

docs/guides/regional-routing.md

@@ -40,7 +40,7 @@ If you omit `with_stt()`, AgentKit uses `FengmingSTT` by default for `Area.CN` c
 
 | Client area | STT classes | LLM classes | MLLM classes | TTS classes | Avatar classes |
 |---|---|---|---|---|---|
-| `Area.US`, `Area.EU`, `Area.AP` | `DeepgramSTT`, `SpeechmaticsSTT`, `MicrosoftSTT`, `OpenAISTT`, `GeminiSTT`, `GoogleSTT`, `AmazonSTT`, `AssemblyAISTT`, `AresSTT`, `SarvamSTT`, `XaiSTT` | `OpenAI`, `AzureOpenAI`, `Anthropic`, `Gemini`, `Groq`, `VertexAILLM`, `AmazonBedrock`, `Dify`, `CustomLLM` | `OpenAIRealtime`, `AzureOpenAIRealtime`, `GeminiLive`, `VertexAI`, `XaiGrok` | `ElevenLabsTTS`, `MicrosoftTTS`, `OpenAITTS`, `CartesiaTTS`, `GoogleTTS`, `AmazonTTS`, `DeepgramTTS`, `GradiumTTS`, `MistralTTS`, `TypecastTTS`, `HumeAITTS`, `RimeTTS`, `FishAudioTTS`, `MiniMaxTTS`, `MurfTTS`, `SarvamTTS`, `GenericTTS`, `XaiTTS` | `LiveAvatarAvatar`, `HeyGenAvatar`, `AkoolAvatar`, `AnamAvatar`, `GenericAvatar` |
+| `Area.US`, `Area.EU`, `Area.AP` | `DeepgramSTT`, `SpeechmaticsSTT`, `MicrosoftSTT`, `OpenAISTT`, `GeminiSTT`, `GoogleSTT`, `AmazonSTT`, `AssemblyAISTT`, `AresSTT`, `SarvamSTT`, `XaiSTT` | `OpenAI`, `AzureOpenAI`, `Anthropic`, `Gemini`, `Groq`, `VertexAILLM`, `AmazonBedrock`, `Dify`, `CustomLLM` | `OpenAIRealtime`, `AzureOpenAIRealtime`, `GeminiLive`, `VertexAI`, `XaiGrok` | `ElevenLabsTTS`, `MicrosoftTTS`, `OpenAITTS`, `CartesiaTTS`, `GeminiTTS`, `GoogleTTS`, `AmazonTTS`, `DeepgramTTS`, `GradiumTTS`, `MistralTTS`, `TypecastTTS`, `HumeAITTS`, `RimeTTS`, `FishAudioTTS`, `MiniMaxTTS`, `MurfTTS`, `SarvamTTS`, `GenericTTS`, `XaiTTS` | `LiveAvatarAvatar`, `HeyGenAvatar`, `AkoolAvatar`, `AnamAvatar`, `GenericAvatar` |
 | `Area.CN` | `FengmingSTT`, `TencentSTT`, `MicrosoftCNSTT`, `XfyunSTT`, `XfyunBigModelSTT`, `XfyunDialectSTT` | `AliyunLLM`, `BytedanceLLM`, `DeepSeekLLM`, `TencentLLM` | `QwenOmni` | `MiniMaxCNTTS`, `TencentTTS`, `BytedanceTTS`, `MicrosoftCNTTS`, `CosyVoiceTTS`, `BytedanceDuplexTTS`, `StepFunTTS`, `GenericTTS` | `SenseTimeAvatar`, `SpatiusAvatar` |
 
 Global client example:

docs/reference/agent.md

@@ -178,6 +178,7 @@ Enable or disable MCP and inline REST tool invocation by setting `advanced_featu
 ### `with_parameters(parameters: SessionParams) -> Agent`
 
 Set session parameters (silence config, farewell config, data channel, audio scenario, etc.).
+Use `SpeakConfig(batch=False)` in the `speak` field to control sentence segmentation for speak requests.
 
 ### `with_audio_scenario(audio_scenario: ParametersAudioScenario) -> Agent`
 

docs/reference/vendors.md

@@ -20,7 +20,7 @@ Construct vendors directly from `agora_agent`, then bind a client with `Agent(cl
 
 | Area | STT classes | LLM classes | MLLM classes | TTS classes | Avatar classes |
 |---|---|---|---|---|---|
-| `Area.US`, `Area.EU`, `Area.AP` | `DeepgramSTT`, `SpeechmaticsSTT`, `MicrosoftSTT`, `OpenAISTT`, `GeminiSTT`, `GoogleSTT`, `AmazonSTT`, `AssemblyAISTT`, `AresSTT`, `SarvamSTT`, `XaiSTT` | `OpenAI`, `AzureOpenAI`, `Anthropic`, `Gemini`, `Groq`, `VertexAILLM`, `AmazonBedrock`, `Dify`, `CustomLLM` | `OpenAIRealtime`, `AzureOpenAIRealtime`, `GeminiLive`, `VertexAI`, `XaiGrok` | `ElevenLabsTTS`, `MicrosoftTTS`, `OpenAITTS`, `CartesiaTTS`, `GoogleTTS`, `AmazonTTS`, `DeepgramTTS`, `GradiumTTS`, `MistralTTS`, `TypecastTTS`, `HumeAITTS`, `RimeTTS`, `FishAudioTTS`, `MiniMaxTTS`, `MurfTTS`, `SarvamTTS`, `GenericTTS`, `XaiTTS` | `LiveAvatarAvatar`, `HeyGenAvatar`, `AkoolAvatar`, `AnamAvatar`, `GenericAvatar` |
+| `Area.US`, `Area.EU`, `Area.AP` | `DeepgramSTT`, `SpeechmaticsSTT`, `MicrosoftSTT`, `OpenAISTT`, `GeminiSTT`, `GoogleSTT`, `AmazonSTT`, `AssemblyAISTT`, `AresSTT`, `SarvamSTT`, `XaiSTT` | `OpenAI`, `AzureOpenAI`, `Anthropic`, `Gemini`, `Groq`, `VertexAILLM`, `AmazonBedrock`, `Dify`, `CustomLLM` | `OpenAIRealtime`, `AzureOpenAIRealtime`, `GeminiLive`, `VertexAI`, `XaiGrok` | `ElevenLabsTTS`, `MicrosoftTTS`, `OpenAITTS`, `CartesiaTTS`, `GeminiTTS`, `GoogleTTS`, `AmazonTTS`, `DeepgramTTS`, `GradiumTTS`, `MistralTTS`, `TypecastTTS`, `HumeAITTS`, `RimeTTS`, `FishAudioTTS`, `MiniMaxTTS`, `MurfTTS`, `SarvamTTS`, `GenericTTS`, `XaiTTS` | `LiveAvatarAvatar`, `HeyGenAvatar`, `AkoolAvatar`, `AnamAvatar`, `GenericAvatar` |
 | `Area.CN` | `FengmingSTT`, `TencentSTT`, `MicrosoftCNSTT`, `XfyunSTT`, `XfyunBigModelSTT`, `XfyunDialectSTT` | `AliyunLLM`, `BytedanceLLM`, `DeepSeekLLM`, `TencentLLM` | `QwenOmni` | `MiniMaxCNTTS`, `TencentTTS`, `BytedanceTTS`, `MicrosoftCNTTS`, `CosyVoiceTTS`, `BytedanceDuplexTTS`, `StepFunTTS`, `GenericTTS` | `SenseTimeAvatar`, `SpatiusAvatar` |
 
 Global example:
@@ -272,6 +272,25 @@ The SDK also includes named helpers for the remaining Agora-supported LLM provid
 | `sample_rate` | `int` | No | `None` | Sample rate: 8000–48000 Hz |
 | `skip_patterns` | `List[int]` | No | `None` | Skip patterns |
 
+### `GeminiTTS`
+
+Uses the configured regional production endpoint for synchronous and asynchronous
+sessions. Package-root and historical preview imports alias the production class.
+Credentials and options serialize inside `tts.params`, with `tts.vendor = "gemini"`.
+
+| Parameter | Type | Required | Default | Description |
+|---|---|---|---|---|
+| `api_key` | `str` | Yes | — | Google Gemini API key |
+| `model` | `str` | No | `GeminiTTSModels.FLASH_38` | Model identifier, sent verbatim |
+| `voice` | `str` | No | `Puck` | Voice name |
+| `style` | `Optional[str]` | No | `None` | Natural-language speaking instruction |
+| `additional_params` | `Optional[Dict[str, Any]]` | No | `None` | Provider-specific parameters merged into `tts.params`; named options take precedence |
+| `skip_patterns` | `Optional[List[int]]` | No | `None` | Patterns to skip in TTS output |
+
+Blank credentials, model, or voice are rejected. Existing v2.11.0 calls and raw
+configs remain compatible without a preview feature header. No sample rate
+option is exposed.
+
 ### `GoogleTTS`
 
 | Parameter | Type | Required | Default | Description |
@@ -446,7 +465,11 @@ AgentKit serializes `credential_mode` at the top level of the Rime TTS configura
 | `pitch` | `float` | No | `None` | Pitch adjustment |
 | `pace` | `float` | No | `None` | Speed of speech |
 | `loudness` | `float` | No | `None` | Volume level |
-| `sample_rate` | `int` | No | `None` | Audio sample rate |
+| `speech_sample_rate` | `int` | No | `None` | Output speech sample rate in Hz |
+| `enable_preprocessing` | `bool` | No | `None` | Normalize English words and numeric entities before synthesis |
+| `model` | `str` | No | `None` | Sarvam TTS model; defaults to `bulbul:v3` |
+| `sample_rate` | `int` | No | `None` | Deprecated alias for `speech_sample_rate` |
+| `additional_params` | `Dict[str, Any]` | No | `None` | Additional Sarvam TTS parameters, flattened into `params` |
 | `skip_patterns` | `List[int]` | No | `None` | Skip patterns |
 
 ### `SmallestAITTS`
@@ -636,6 +659,26 @@ stt = GeminiSTT(
 | `language` | `str` | No | `None` | Language code for speech recognition |
 | `additional_params` | `Dict[str, Any]` | No | `None` | Additional xAI STT parameters |
 
+### `RtzrSTT`
+
+Global-only RTZR speech-to-text provider (`asr.vendor: "rtzr"`).
+
+| Parameter | Type | Required | Default | Description |
+|---|---|---|---|---|
+| `client_id` | `str` | Yes | — | RTZR client ID |
+| `client_secret` | `str` | Yes | — | RTZR client secret |
+| `api_base` | `str` | No | `None` | RTZR API base URL |
+| `model_name` | `str` | No | `None` | RTZR recognition model name |
+| `language` | `str` | No | `None` | RTZR recognition language code |
+| `sample_rate` | `int` | No | `None` | Input audio sample rate in Hz |
+| `encoding` | `str` | No | `None` | Input audio encoding |
+| `use_itn` | `bool` | No | `None` | Enable inverse text normalization |
+| `use_disfluency_filter` | `bool` | No | `None` | Filter disfluencies |
+| `use_profanity_filter` | `bool` | No | `None` | Filter profanity |
+| `use_punctuation` | `bool` | No | `None` | Add punctuation to recognized text |
+| `keywords` | `List[str]` | No | `None` | Keywords that improve recognition accuracy |
+| `additional_params` | `Dict[str, Any]` | No | `None` | Additional RTZR parameters |
+
 ### `SmallestAISTT`
 
 Global-only Smallest AI speech-to-text provider (`asr.vendor: "smallestai"`).
@@ -1111,6 +1154,9 @@ Same options as `HeyGenAvatar`, but serializes `vendor: "liveavatar"`. `agora_to
 |---|---|---|---|---|
 | `api_key` | `str` | Yes | — | Anam API key |
 | `avatar_id` | `str` | Yes | — | Anam avatar ID |
+| `avatar_model` | `str` | No | `None` | Anam model, such as `cara_mk4` for Cara 4 portrait mode |
+| `video_width` | `int` | No | `None` | Output width in pixels; set together with `video_height` |
+| `video_height` | `int` | No | `None` | Output height in pixels; set together with `video_width` |
 | `enable` | `bool` | No | `True` | Enable or disable the avatar |
 
 ### `GenericAvatar`

src/agora_agent/__init__.py

@@ -72,6 +72,9 @@
         GeminiTTSModels,
         OpenAIGPTLive,
         OpenAISTT,
+        RtzrSTT,
+        SpeakConfig,
+        ParametersSpeak,
         OpenAITTS,
         RimeTTS,
         SarvamSTT,

src/agora_agent/agentkit/__init__.py

@@ -55,6 +55,8 @@
     FarewellConfig,
     ParametersDataChannel,
     ParametersAudioScenario,
+    SpeakConfig,
+    ParametersSpeak,
     InterruptionConfig,
     InterruptionMode,
     MllmTurnDetectionConfig,
@@ -155,7 +157,6 @@
     OpenAITtsPresetModels,
     normalize_preset_input,
 )
-from .preview import GeminiTTS, GeminiTTSModels
 from .vendors import (
     GEMINI_MLLM_DEFAULT_MODEL,
     AkoolAvatar,
@@ -185,6 +186,8 @@
     Gemini,
     GeminiSTT,
     GeminiSTTModels,
+    GeminiTTS,
+    GeminiTTSModels,
     GeminiLive,
     GeminiLiveModels,
     GenericAvatar,
@@ -211,6 +214,7 @@
     OpenAIGPTLive,
     OpenAISampleRate,
     OpenAISTT,
+    RtzrSTT,
     OpenAITTS,
     GenericTTS,
     RimeTTS,
@@ -309,6 +313,8 @@
     "FarewellConfig",
     "ParametersDataChannel",
     "ParametersAudioScenario",
+    "SpeakConfig",
+    "ParametersSpeak",
     "InterruptionConfig",
     "InterruptionMode",
     "MllmTurnDetectionConfig",
@@ -455,6 +461,7 @@
     "MicrosoftSTT",
     "MicrosoftCNSTT",
     "OpenAISTT",
+    "RtzrSTT",
     "GeminiSTT",
     "GeminiSTTModels",
     "GoogleSTT",