diff --git a/.fern/metadata.json b/.fern/metadata.json
index 62289a3..a2b4d2c 100644
--- a/.fern/metadata.json
+++ b/.fern/metadata.json
@@ -14,5 +14,5 @@
     },
     "exclude_types_from_init_exports": true
   },
-  "originGitCommit": "26314a9d3199f5e135302c3c3645b6bbb8f556d4"
+  "originGitCommit": "0073fa0ae670f77e4987407b8cca1cf179d5f90b"
 }
\ No newline at end of file
diff --git a/.fern/replay.lock b/.fern/replay.lock
index 699a092..82f3c76 100644
--- a/.fern/replay.lock
+++ b/.fern/replay.lock
@@ -72,14 +72,28 @@ generations:
     cli_version: unknown
     generator_versions:
       fernapi/fern-python-sdk: 4.37.0
-current_generation: f7e9b5d993665a16b43eb1d7d442b118fbeffa4b
+  - commit_sha: 88a3914f38e882be10f7a0a6cba18e6f5ec98259
+    tree_hash: 0315c486fe5d65313c32c2127af67aaa39f8acf7
+    timestamp: 2026-09-24T03:09:07.591Z
+    cli_version: unknown
+    generator_versions:
+      fernapi/fern-python-sdk: 4.37.0
+  - commit_sha: b1f9cb3bf3fd8da2262807fd603dee3741187918
+    tree_hash: e378b36a4bb43fd42f28703a58814ae759fe9434
+    timestamp: 2026-10-09T14:05:05.278Z
+    cli_version: unknown
+    generator_versions:
+      fernapi/fern-python-sdk: 4.37.0
+current_generation: b1f9cb3bf3fd8da2262807fd603dee3741187918
+forgotten_hashes:
+  - sha256:19124603cb0683406e750ffda0725786727208c0f7ddb481c256a0990b2c5585
 patches:
   - id: patch-13d8e068
     content_hash: sha256:97d879ab169016b9abaf1866427a8ebc45af608f69e063a07e5126c9613c329b
     original_commit: 13d8e0683ffa08bcfc9a381eb9604adf5abdbe0b
     original_message: "chore: ignore local venv and planning docs"
     original_author: plutoless <qianze.zhang@hotmail.com>
-    base_generation: f7e9b5d993665a16b43eb1d7d442b118fbeffa4b
+    base_generation: b1f9cb3bf3fd8da2262807fd603dee3741187918
     files:
       - .gitignore
     patch_content: |
@@ -108,7 +122,7 @@ patches:
     original_commit: a065088b4c45e2ca7e1bb91b1a899b8444800121
     original_message: "fix(vendors): send Speechmatics STT key instead of api_key"
     original_author: digitallysavvy <hermes.frangoudis@gmail.com>
-    base_generation: f7e9b5d993665a16b43eb1d7d442b118fbeffa4b
+    base_generation: b1f9cb3bf3fd8da2262807fd603dee3741187918
     files:
       - src/agora_agent/types/speechmatics_asr_params.py
     patch_content: |
@@ -251,7 +265,7 @@ patches:
     original_commit: 33a436b3bdb6f2a6fe4d8987b6536c2f0ac2862e
     original_message: "fix: address v2.8.0 release review findings"
     original_author: digitallysavvy <hermes.frangoudis@gmail.com>
-    base_generation: f7e9b5d993665a16b43eb1d7d442b118fbeffa4b
+    base_generation: b1f9cb3bf3fd8da2262807fd603dee3741187918
     files:
       - src/agora_agent/agents/types/start_agents_request_properties_advanced_features.py
     patch_content: |
@@ -312,3 +326,4 @@ patches:
                     frozen = True
                     smart_union = True
                     extra = pydantic.Extra.allow
+    user_owned: true
diff --git a/changelog.md b/changelog.md
index 7542307..b7c42fc 100644
--- a/changelog.md
+++ b/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
diff --git a/docs/concepts/vendors.md b/docs/concepts/vendors.md
index 13f6cd5..e0c4918 100644
--- a/docs/concepts/vendors.md
+++ b/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
diff --git a/docs/guides/avatars.md b/docs/guides/avatars.md
index 9e085ff..3f3c5bc 100644
--- a/docs/guides/avatars.md
+++ b/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()`.
diff --git a/docs/guides/preview-endpoint.md b/docs/guides/preview-endpoint.md
index dcb2a59..44fe0d4 100644
--- a/docs/guides/preview-endpoint.md
+++ b/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,
diff --git a/docs/guides/regional-routing.md b/docs/guides/regional-routing.md
index ff6e4cc..f97e6c5 100644
--- a/docs/guides/regional-routing.md
+++ b/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:
diff --git a/docs/reference/agent.md b/docs/reference/agent.md
index 68e46f1..f4e8aed 100644
--- a/docs/reference/agent.md
+++ b/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`
 
diff --git a/docs/reference/vendors.md b/docs/reference/vendors.md
index 959690b..57a8058 100644
--- a/docs/reference/vendors.md
+++ b/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`
diff --git a/src/agora_agent/__init__.py b/src/agora_agent/__init__.py
index f731ff6..1a29ed4 100644
--- a/src/agora_agent/__init__.py
+++ b/src/agora_agent/__init__.py
@@ -72,6 +72,9 @@
         GeminiTTSModels,
         OpenAIGPTLive,
         OpenAISTT,
+        RtzrSTT,
+        SpeakConfig,
+        ParametersSpeak,
         OpenAITTS,
         RimeTTS,
         SarvamSTT,
diff --git a/src/agora_agent/agentkit/__init__.py b/src/agora_agent/agentkit/__init__.py
index 0844a5c..0e658e8 100644
--- a/src/agora_agent/agentkit/__init__.py
+++ b/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",
diff --git a/src/agora_agent/agentkit/agent.py b/src/agora_agent/agentkit/agent.py
index 5eb5659..34be68c 100644
--- a/src/agora_agent/agentkit/agent.py
+++ b/src/agora_agent/agentkit/agent.py
@@ -47,6 +47,7 @@
 from ..agents.types.start_agents_request_properties_parameters_farewell_config import StartAgentsRequestPropertiesParametersFarewellConfig
 from ..agents.types.start_agents_request_properties_parameters_data_channel import StartAgentsRequestPropertiesParametersDataChannel
 from ..agents.types.start_agents_request_properties_parameters_audio_scenario import StartAgentsRequestPropertiesParametersAudioScenario
+from ..agents.types.start_agents_request_properties_parameters_speak import StartAgentsRequestPropertiesParametersSpeak
 from ..agents.types.start_agents_request_properties_interruption import StartAgentsRequestPropertiesInterruption
 from ..agents.types.start_agents_request_properties_interruption_mode import StartAgentsRequestPropertiesInterruptionMode
 from ..agents.types.start_agents_request_properties_geofence import StartAgentsRequestPropertiesGeofence
@@ -140,6 +141,8 @@
 FarewellConfig = StartAgentsRequestPropertiesParametersFarewellConfig
 ParametersDataChannel = StartAgentsRequestPropertiesParametersDataChannel
 ParametersAudioScenario = StartAgentsRequestPropertiesParametersAudioScenario
+SpeakConfig = StartAgentsRequestPropertiesParametersSpeak
+ParametersSpeak = SpeakConfig
 InterruptionConfig = StartAgentsRequestPropertiesInterruption
 InterruptionMode = StartAgentsRequestPropertiesInterruptionMode
 MllmTurnDetectionConfig = MllmTurnDetection
@@ -160,6 +163,7 @@
 
 
 class SessionParamsInput(typing_extensions.TypedDict, total=False):
+    speak: StartAgentsRequestPropertiesParametersSpeak
     silence_config: StartAgentsRequestPropertiesParametersSilenceConfig
     farewell_config: StartAgentsRequestPropertiesParametersFarewellConfig
     data_channel: StartAgentsRequestPropertiesParametersDataChannel
diff --git a/src/agora_agent/agentkit/agent_session.py b/src/agora_agent/agentkit/agent_session.py
index 18a2bac..b1171db 100644
--- a/src/agora_agent/agentkit/agent_session.py
+++ b/src/agora_agent/agentkit/agent_session.py
@@ -260,6 +260,7 @@ def _validate_avatar_config(self) -> None:
             sample_rate = (
                 tts_params.get("sample_rate")
                 or tts_params.get("sample_rate_hertz")
+                or tts_params.get("speech_sample_rate")
                 or tts_params.get("samplingRate")
             )
         if isinstance(sample_rate, int):
diff --git a/src/agora_agent/agentkit/avatar_types.py b/src/agora_agent/agentkit/avatar_types.py
index 577803f..ad52941 100644
--- a/src/agora_agent/agentkit/avatar_types.py
+++ b/src/agora_agent/agentkit/avatar_types.py
@@ -102,6 +102,10 @@ def validate_avatar_config(
             raise ValueError("Anam avatar requires api_key")
         if not params.get("avatar_id"):
             raise ValueError("Anam avatar requires avatar_id")
+        has_video_width = params.get("video_width") is not None
+        has_video_height = params.get("video_height") is not None
+        if has_video_width != has_video_height:
+            raise ValueError("Anam avatar requires video_width and video_height together")
     elif is_generic_avatar(config):
         params = config.get("params", {})
         if not params.get("api_key"):
diff --git a/src/agora_agent/agentkit/presets.py b/src/agora_agent/agentkit/presets.py
index f160cee..ce76a2c 100644
--- a/src/agora_agent/agentkit/presets.py
+++ b/src/agora_agent/agentkit/presets.py
@@ -108,7 +108,7 @@ def infer_asr_preset(asr: typing.Optional[typing.Dict[str, typing.Any]]) -> typi
     if not asr or asr.get("vendor") != "deepgram":
         return None
     params = asr.get("params") or {}
-    if params.get("key"):
+    if params.get("api_key"):
         return None
     return _DEEPGRAM_MODEL_TO_PRESET.get(_normalize_model_name(params.get("model")) or "")
 
diff --git a/src/agora_agent/agentkit/preview/__init__.py b/src/agora_agent/agentkit/preview/__init__.py
index 55b1675..62f089d 100644
--- a/src/agora_agent/agentkit/preview/__init__.py
+++ b/src/agora_agent/agentkit/preview/__init__.py
@@ -1,8 +1,4 @@
-"""Preview endpoint support.
-
-Temporary package: delete it when these providers ship on the production
-gateway. See ``client.py`` for the routing and gate header.
-"""
+"""Preview routing helpers and legacy aliases for production vendors."""
 
 from .client import (
     PREVIEW_API_BASE_URL,
@@ -13,7 +9,7 @@
     create_preview_session_clients,
     required_preview_features,
 )
-from .gemini_tts import GeminiTTS, GeminiTTSModels
+from ..vendors.tts import GeminiTTS, GeminiTTSModels
 from .vendors import (
     GEMINI_MLLM_DEFAULT_MODEL,
     GEMINI_PREVIEW_MLLM_URL,
diff --git a/src/agora_agent/agentkit/preview/client.py b/src/agora_agent/agentkit/preview/client.py
index e917b1c..1cd30d6 100644
--- a/src/agora_agent/agentkit/preview/client.py
+++ b/src/agora_agent/agentkit/preview/client.py
@@ -5,9 +5,9 @@
 for preview traffic while the caller's ``Agora`` / ``AsyncAgora`` client stays
 on its production endpoint.
 
-Everything under ``agentkit/preview/`` is temporary. When a provider ships on
-the production gateway, remove its preview registration and move its class into
-the corresponding production vendor module.
+Preview registrations are temporary. When a provider ships on the production
+gateway, remove its registration and move its class into the corresponding
+production vendor module, retaining historical imports as aliases.
 """
 
 from __future__ import annotations
@@ -37,7 +37,7 @@ class PreviewFeatures:
     vendors on the preview endpoint.
     """
 
-    #: Gemini TTS preview gate. Gemini ASR and Live use production.
+    #: Legacy Gemini gate retained for compatibility. ASR, Live, and TTS use production.
     GEMINI_LIVE = "gemini-live"
     LIVE_MODELS = "live-models"
 
@@ -131,10 +131,10 @@ def create_preview_session_clients(
     raise TypeError("Unsupported Agora client wrapper")
 
 
-#: ASR vendors served only by the preview endpoint.
+#: Vendors served only by the preview endpoint, keyed by category.
 _PREVIEW_FEATURES_BY_CATEGORY: typing.Dict[str, typing.Dict[str, PreviewFeature]] = {
     "asr": {},
-    "tts": {"gemini": PreviewFeatures.GEMINI_LIVE},
+    "tts": {},
     "mllm": {},
 }
 
diff --git a/src/agora_agent/agentkit/preview/gemini_tts.py b/src/agora_agent/agentkit/preview/gemini_tts.py
index fbf0282..9cc3b2e 100644
--- a/src/agora_agent/agentkit/preview/gemini_tts.py
+++ b/src/agora_agent/agentkit/preview/gemini_tts.py
@@ -1,35 +1,5 @@
-"""Gemini TTS preview; reuses the gemini-live session gate."""
+"""Legacy preview import aliases for the production Gemini TTS implementation."""
 
-from typing import Any, Dict, Optional
+from ..vendors.tts import GeminiTTS, GeminiTTSModels
 
-from ..vendors.base import BaseTTS
-from pydantic import ConfigDict, Field, field_validator
-
-
-class GeminiTTSModels:
-    """Gemini 3.8 Flash TTS preview model."""
-
-    FLASH_38 = "gemini-3.8-flash-tts"
-
-
-class GeminiTTS(BaseTTS):
-    """Preview-only TTS; model names are sent verbatim for rollout flexibility."""
-
-    model_config = ConfigDict(extra="forbid")
-    api_key: str = Field(..., repr=False)
-    model: str = GeminiTTSModels.FLASH_38
-    voice: str = "Puck"
-    style: Optional[str] = None
-
-    @field_validator("api_key", "model", "voice")
-    @classmethod
-    def require_nonblank(cls, value: str) -> str:
-        if not value.strip():
-            raise ValueError("GeminiTTS requires a nonblank value")
-        return value
-
-    def to_config(self) -> Dict[str, Any]:
-        params: Dict[str, Any] = {"api_key": self.api_key, "model": self.model, "voice": self.voice}
-        if self.style is not None:
-            params["style"] = self.style
-        return {"vendor": "gemini", "params": params}
+__all__ = ["GeminiTTS", "GeminiTTSModels"]
diff --git a/src/agora_agent/agentkit/regional_agent.py b/src/agora_agent/agentkit/regional_agent.py
index 8658154..e5ba3fc 100644
--- a/src/agora_agent/agentkit/regional_agent.py
+++ b/src/agora_agent/agentkit/regional_agent.py
@@ -34,6 +34,7 @@
     GoogleSTT,
     MicrosoftSTT,
     OpenAISTT,
+    RtzrSTT,
     SarvamSTT,
     SmallestAISTT,
     SpeechmaticsSTT,
@@ -57,6 +58,7 @@
     DeepgramTTS,
     ElevenLabsTTS,
     FishAudioTTS,
+    GeminiTTS,
     GoogleTTS,
     GradiumTTS,
     HumeAITTS,
@@ -91,10 +93,12 @@
     AssemblyAISTT,
     SpeechmaticsSTT,
     SarvamSTT,
+    RtzrSTT,
     SmallestAISTT,
     XaiSTT,
 ]
 GlobalTTS = typing.Union[
+    GeminiTTS,
     MicrosoftTTS,
     ElevenLabsTTS,
     MiniMaxTTS,
diff --git a/src/agora_agent/agentkit/vendors/__init__.py b/src/agora_agent/agentkit/vendors/__init__.py
index 6cd22b6..89bf147 100644
--- a/src/agora_agent/agentkit/vendors/__init__.py
+++ b/src/agora_agent/agentkit/vendors/__init__.py
@@ -58,6 +58,7 @@
     GoogleSTT,
     MicrosoftSTT,
     OpenAISTT,
+    RtzrSTT,
     SarvamSTT,
     SmallestAISTT,
     SpeechmaticsSTT,
@@ -69,6 +70,8 @@
     DeepgramTTS,
     ElevenLabsTTS,
     FishAudioTTS,
+    GeminiTTS,
+    GeminiTTSModels,
     GoogleTTS,
     GradiumTTS,
     HumeAITTS,
@@ -112,6 +115,8 @@
     "CartesiaTTS",
     "DeepgramTTS",
     "GoogleTTS",
+    "GeminiTTS",
+    "GeminiTTSModels",
     "GradiumTTS",
     "AmazonTTS",
     "HumeAITTS",
@@ -137,6 +142,7 @@
     "MicrosoftSTT",
     "MicrosoftCNSTT",
     "OpenAISTT",
+    "RtzrSTT",
     "GeminiSTT",
     "GeminiSTTModels",
     "GoogleSTT",
diff --git a/src/agora_agent/agentkit/vendors/avatar.py b/src/agora_agent/agentkit/vendors/avatar.py
index 83952e3..01d86f9 100644
--- a/src/agora_agent/agentkit/vendors/avatar.py
+++ b/src/agora_agent/agentkit/vendors/avatar.py
@@ -2,7 +2,7 @@
 from typing import Any, Dict, Optional
 
 from .base import BaseAvatar
-from pydantic import BaseModel, ConfigDict, Field, field_validator
+from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator
 
 LIVEAVATAR_SAMPLE_RATE = 24000
 HEYGEN_SAMPLE_RATE = LIVEAVATAR_SAMPLE_RATE
@@ -199,9 +199,30 @@ class AnamAvatarOptions(BaseModel):
 
     api_key: str = Field(..., description="Anam API key")
     avatar_id: str = Field(..., description="Anam avatar ID")
+    avatar_model: Optional[str] = Field(
+        default=None,
+        description="Anam avatar model, such as cara_mk4 for Cara 4 portrait mode",
+    )
+    video_width: Optional[int] = Field(
+        default=None,
+        description="Output video width in pixels; set together with video_height",
+    )
+    video_height: Optional[int] = Field(
+        default=None,
+        description="Output video height in pixels; set together with video_width",
+    )
     enable: Optional[bool] = Field(default=None, description="Enable avatar (default: true)")
     additional_params: Optional[Dict[str, Any]] = Field(default=None, description="Additional vendor-specific parameters")
 
+    @model_validator(mode="after")
+    def _validate_video_dimensions(self) -> "AnamAvatarOptions":
+        additional_params = self.additional_params or {}
+        has_video_width = self.video_width is not None or additional_params.get("video_width") is not None
+        has_video_height = self.video_height is not None or additional_params.get("video_height") is not None
+        if has_video_width != has_video_height:
+            raise ValueError("Anam avatar requires video_width and video_height together")
+        return self
+
 
 class AnamAvatar(AnamAvatarOptions, BaseAvatar):
     @property
@@ -209,13 +230,24 @@ def required_sample_rate(self) -> int:
         return 0
 
     def to_config(self) -> Dict[str, Any]:
-        params: Dict[str, Any] = {
-            "api_key": self.api_key,
-            "avatar_id": self.avatar_id,
-        }
-
-        if self.additional_params is not None:
-            params = {**self.additional_params, **params}
+        params: Dict[str, Any] = dict(self.additional_params or {})
+        params.update(
+            {
+                "api_key": self.api_key,
+                "avatar_id": self.avatar_id,
+            }
+        )
+        if self.avatar_model is not None:
+            params["avatar_model"] = self.avatar_model
+        if self.video_width is not None:
+            params["video_width"] = self.video_width
+        if self.video_height is not None:
+            params["video_height"] = self.video_height
+
+        has_video_width = params.get("video_width") is not None
+        has_video_height = params.get("video_height") is not None
+        if has_video_width != has_video_height:
+            raise ValueError("Anam avatar requires video_width and video_height together")
 
         enable = self.enable if self.enable is not None else True
         return {"enable": enable, "vendor": "anam", "params": params}
diff --git a/src/agora_agent/agentkit/vendors/catalog.py b/src/agora_agent/agentkit/vendors/catalog.py
index ebfb918..7ee986b 100644
--- a/src/agora_agent/agentkit/vendors/catalog.py
+++ b/src/agora_agent/agentkit/vendors/catalog.py
@@ -26,6 +26,7 @@
     GoogleSTT,
     MicrosoftSTT,
     OpenAISTT,
+    RtzrSTT,
     SarvamSTT,
     SmallestAISTT,
     SpeechmaticsSTT,
@@ -37,6 +38,7 @@
     DeepgramTTS,
     ElevenLabsTTS,
     FishAudioTTS,
+    GeminiTTS,
     GoogleTTS,
     GradiumTTS,
     HumeAITTS,
@@ -83,6 +85,7 @@ def __init__(
         "assemblyai": AssemblyAISTT,
         "speechmatics": SpeechmaticsSTT,
         "sarvam": SarvamSTT,
+        "rtzr": RtzrSTT,
         "smallestai": SmallestAISTT,
         "xai": XaiSTT,
     },
@@ -106,6 +109,7 @@ def __init__(
         "xai": XaiGrok,
     },
     tts={
+        "gemini": GeminiTTS,
         "microsoft": MicrosoftTTS,
         "elevenlabs": ElevenLabsTTS,
         "minimax": MiniMaxTTS,
diff --git a/src/agora_agent/agentkit/vendors/namespaces.py b/src/agora_agent/agentkit/vendors/namespaces.py
index 782c5aa..b3d982b 100644
--- a/src/agora_agent/agentkit/vendors/namespaces.py
+++ b/src/agora_agent/agentkit/vendors/namespaces.py
@@ -23,6 +23,7 @@
     GoogleSTT,
     MicrosoftSTT,
     OpenAISTT,
+    RtzrSTT,
     SarvamSTT,
     SmallestAISTT,
     SpeechmaticsSTT,
@@ -34,6 +35,7 @@
     DeepgramTTS,
     ElevenLabsTTS,
     FishAudioTTS,
+    GeminiTTS,
     GoogleTTS,
     GradiumTTS,
     HumeAITTS,
@@ -62,6 +64,7 @@ class GlobalSTTVendors:
     assemblyai = AssemblyAISTT
     speechmatics = SpeechmaticsSTT
     sarvam = SarvamSTT
+    rtzr = RtzrSTT
     smallestai = SmallestAISTT
     xai = XaiSTT
 
@@ -108,6 +111,7 @@ class CNMLLMVendors:
 
 
 class GlobalTTSVendors:
+    gemini = GeminiTTS
     microsoft = MicrosoftTTS
     elevenlabs = ElevenLabsTTS
     minimax = MiniMaxTTS
diff --git a/src/agora_agent/agentkit/vendors/region.py b/src/agora_agent/agentkit/vendors/region.py
index 495c8bd..d4944fc 100644
--- a/src/agora_agent/agentkit/vendors/region.py
+++ b/src/agora_agent/agentkit/vendors/region.py
@@ -28,6 +28,7 @@
     "assemblyai",
     "speechmatics",
     "sarvam",
+    "rtzr",
     "smallestai",
     "xai",
 )
@@ -43,6 +44,7 @@
     "generic",
 )
 GLOBAL_TTS_VENDORS: typing.Tuple[str, ...] = (
+    "gemini",
     "microsoft",
     "elevenlabs",
     "minimax",
diff --git a/src/agora_agent/agentkit/vendors/stt.py b/src/agora_agent/agentkit/vendors/stt.py
index f05fe79..ffeb9d1 100644
--- a/src/agora_agent/agentkit/vendors/stt.py
+++ b/src/agora_agent/agentkit/vendors/stt.py
@@ -97,7 +97,7 @@ def to_config(self) -> Dict[str, Any]:
         params: Dict[str, Any] = dict(self.additional_params or {})
 
         if self.api_key is not None:
-            params["key"] = self.api_key
+            params["api_key"] = self.api_key
         if self.model is not None:
             params["model"] = self.model
         if self.language is not None:
@@ -115,6 +115,50 @@ def to_config(self) -> Dict[str, Any]:
         return config
 
 
+class RtzrSTTOptions(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    client_id: str = Field(..., min_length=1, description="RTZR client ID")
+    client_secret: str = Field(..., min_length=1, description="RTZR client secret")
+    api_base: Optional[str] = Field(default=None, description="RTZR API base URL")
+    model_name: Optional[str] = Field(default=None, description="RTZR recognition model name")
+    language: Optional[str] = Field(default=None, description="RTZR recognition language code")
+    sample_rate: Optional[int] = Field(default=None, gt=0, description="Input audio sample rate in Hz")
+    encoding: Optional[str] = Field(default=None, description="Input audio encoding")
+    use_itn: Optional[bool] = Field(default=None, description="Enable inverse text normalization")
+    use_disfluency_filter: Optional[bool] = Field(default=None, description="Filter disfluencies such as stuttering")
+    use_profanity_filter: Optional[bool] = Field(default=None, description="Filter profanity")
+    use_punctuation: Optional[bool] = Field(default=None, description="Add punctuation to recognized text")
+    keywords: Optional[List[str]] = Field(default=None, description="Keywords that improve recognition accuracy")
+    additional_params: Optional[Dict[str, Any]] = Field(default=None)
+
+
+class RtzrSTT(RtzrSTTOptions, BaseSTT):
+    """RTZR streaming speech-to-text provider."""
+
+    def to_config(self) -> Dict[str, Any]:
+        params: Dict[str, Any] = dict(self.additional_params or {})
+        params.update({"client_id": self.client_id, "client_secret": self.client_secret})
+
+        for name in (
+            "api_base",
+            "model_name",
+            "language",
+            "sample_rate",
+            "encoding",
+            "use_itn",
+            "use_disfluency_filter",
+            "use_profanity_filter",
+            "use_punctuation",
+            "keywords",
+        ):
+            value = getattr(self, name)
+            if value is not None:
+                params[name] = value
+
+        return {"vendor": "rtzr", "params": params}
+
+
 class MicrosoftSTTOptions(BaseModel):
     model_config = ConfigDict(extra="forbid")
 
diff --git a/src/agora_agent/agentkit/vendors/tts.py b/src/agora_agent/agentkit/vendors/tts.py
index 0d8d3ff..ba9f22f 100644
--- a/src/agora_agent/agentkit/vendors/tts.py
+++ b/src/agora_agent/agentkit/vendors/tts.py
@@ -1,3 +1,4 @@
+import warnings
 from typing import Any, Dict, List, Literal, Optional
 from urllib.parse import urlsplit
 
@@ -7,6 +8,45 @@
 from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator
 
 
+class GeminiTTSModels:
+    """Gemini 3.8 Flash TTS model."""
+
+    FLASH_38 = "gemini-3.8-flash-tts"
+
+
+class GeminiTTS(BaseTTS):
+    """Production Gemini TTS; preview-era options and wire fields remain compatible."""
+
+    model_config = ConfigDict(extra="forbid")
+    api_key: str = Field(..., repr=False)
+    model: str = GeminiTTSModels.FLASH_38
+    voice: str = "Puck"
+    style: Optional[str] = None
+    additional_params: Optional[Dict[str, Any]] = None
+    skip_patterns: Optional[List[int]] = None
+
+    @field_validator("api_key", "model", "voice")
+    @classmethod
+    def require_nonblank(cls, value: str) -> str:
+        if not value.strip():
+            raise ValueError("GeminiTTS requires a nonblank value")
+        return value
+
+    def to_config(self) -> Dict[str, Any]:
+        params: Dict[str, Any] = dict(self.additional_params or {})
+        params.update({
+            "api_key": self.api_key,
+            "model": self.model,
+            "voice": self.voice,
+        })
+        if self.style is not None:
+            params["style"] = self.style
+        result: Dict[str, Any] = {"vendor": "gemini", "params": params}
+        if self.skip_patterns is not None:
+            result["skip_patterns"] = self.skip_patterns
+        return result
+
+
 class ElevenLabsTTSOptions(BaseModel):
     model_config = ConfigDict(extra="forbid")
 
@@ -568,9 +608,38 @@ class SarvamTTSOptions(BaseModel):
     pitch: Optional[float] = Field(default=None, description="Pitch adjustment")
     pace: Optional[float] = Field(default=None, description="Speed of speech")
     loudness: Optional[float] = Field(default=None, description="Volume level")
-    sample_rate: Optional[int] = Field(default=None, description="Audio sample rate in Hz")
+    speech_sample_rate: Optional[int] = Field(default=None, gt=0, description="Output speech sample rate in Hz")
+    enable_preprocessing: Optional[bool] = Field(
+        default=None,
+        description="Normalize English words and numeric entities before synthesis",
+    )
+    model: Optional[str] = Field(default=None, description="Sarvam TTS model")
+    sample_rate: Optional[int] = Field(
+        default=None,
+        gt=0,
+        description="Deprecated alias for speech_sample_rate",
+    )
+    additional_params: Optional[Dict[str, Any]] = Field(default=None, description="Additional Sarvam TTS parameters")
     skip_patterns: Optional[List[int]] = Field(default=None)
 
+    @model_validator(mode="before")
+    @classmethod
+    def _warn_deprecated_sample_rate(cls, values: Any) -> Any:
+        if isinstance(values, dict) and "sample_rate" in values:
+            warnings.warn(
+                "SarvamTTS.sample_rate is deprecated; use speech_sample_rate instead.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+        return values
+
+    @model_validator(mode="after")
+    def _validate_sample_rate_alias(self) -> "SarvamTTSOptions":
+        legacy_sample_rate = self.__dict__.get("sample_rate")
+        if self.speech_sample_rate is not None and legacy_sample_rate is not None:
+            raise ValueError("SarvamTTS accepts only one of speech_sample_rate or sample_rate")
+        return self
+
 
 class SarvamTTS(SarvamTTSOptions, BaseTTS):
     @property
@@ -578,19 +647,26 @@ def resolved_sample_rate(self) -> Optional[int]:
         return None
 
     def to_config(self) -> Dict[str, Any]:
-        params: Dict[str, Any] = {
+        params: Dict[str, Any] = dict(self.additional_params or {})
+        params.update({
             "api_subscription_key": self.key,
             "speaker": self.speaker,
             "target_language_code": self.target_language_code,
-        }
+        })
         if self.pitch is not None:
             params["pitch"] = self.pitch
         if self.pace is not None:
             params["pace"] = self.pace
         if self.loudness is not None:
             params["loudness"] = self.loudness
-        if self.sample_rate is not None:
-            params["sample_rate"] = self.sample_rate
+        legacy_sample_rate = self.__dict__.get("sample_rate")
+        speech_sample_rate = self.speech_sample_rate if self.speech_sample_rate is not None else legacy_sample_rate
+        if speech_sample_rate is not None:
+            params["speech_sample_rate"] = speech_sample_rate
+        if self.enable_preprocessing is not None:
+            params["enable_preprocessing"] = self.enable_preprocessing
+        if self.model is not None:
+            params["model"] = self.model
 
         result: Dict[str, Any] = {"vendor": "sarvam", "params": params}
         if self.skip_patterns is not None:
diff --git a/src/agora_agent/agents/types/start_agents_request_properties_parameters.py b/src/agora_agent/agents/types/start_agents_request_properties_parameters.py
index 33bb237..d3153e9 100644
--- a/src/agora_agent/agents/types/start_agents_request_properties_parameters.py
+++ b/src/agora_agent/agents/types/start_agents_request_properties_parameters.py
@@ -15,6 +15,7 @@
 from .start_agents_request_properties_parameters_silence_config import (
     StartAgentsRequestPropertiesParametersSilenceConfig,
 )
+from .start_agents_request_properties_parameters_speak import StartAgentsRequestPropertiesParametersSpeak
 
 
 class StartAgentsRequestPropertiesParameters(UncheckedBaseModel):
@@ -22,6 +23,11 @@ class StartAgentsRequestPropertiesParameters(UncheckedBaseModel):
     Agent configuration parameters.
     """
 
+    speak: typing.Optional[StartAgentsRequestPropertiesParametersSpeak] = pydantic.Field(default=None)
+    """
+    Settings for the agent's speak behavior.
+    """
+
     silence_config: typing.Optional[StartAgentsRequestPropertiesParametersSilenceConfig] = pydantic.Field(default=None)
     """
     Settings related to agent silence behavior. Does not apply when you integrate a `mllm`.
diff --git a/src/agora_agent/agents/types/start_agents_request_properties_parameters_speak.py b/src/agora_agent/agents/types/start_agents_request_properties_parameters_speak.py
new file mode 100644
index 0000000..95003bc
--- /dev/null
+++ b/src/agora_agent/agents/types/start_agents_request_properties_parameters_speak.py
@@ -0,0 +1,29 @@
+# This file was auto-generated by Fern from our API Definition.
+
+import typing
+
+import pydantic
+from ...core.pydantic_utilities import IS_PYDANTIC_V2
+from ...core.unchecked_base_model import UncheckedBaseModel
+
+
+class StartAgentsRequestPropertiesParametersSpeak(UncheckedBaseModel):
+    """
+    Settings for the agent's speak behavior.
+    """
+
+    batch: typing.Optional[bool] = pydantic.Field(default=None)
+    """
+    Whether to skip sentence segmentation for speak requests:
+    - `false`: Skip sentence segmentation.
+    - Omitted or `true`: Preserve sentence segmentation.
+    """
+
+    if IS_PYDANTIC_V2:
+        model_config: typing.ClassVar[pydantic.ConfigDict] = pydantic.ConfigDict(extra="allow", frozen=True)  # type: ignore # Pydantic v2
+    else:
+
+        class Config:
+            frozen = True
+            smart_union = True
+            extra = pydantic.Extra.allow
diff --git a/src/agora_agent/core/client_wrapper.py b/src/agora_agent/core/client_wrapper.py
index ad3aa18..4f480c9 100644
--- a/src/agora_agent/core/client_wrapper.py
+++ b/src/agora_agent/core/client_wrapper.py
@@ -26,10 +26,10 @@ def __init__(
 
     def get_headers(self) -> typing.Dict[str, str]:
         headers: typing.Dict[str, str] = {
-            "User-Agent": "agora-agents/v2.12.0",
+            "User-Agent": "agora-agents/v2.13.0",
             "X-Fern-Language": "Python",
             "X-Fern-SDK-Name": "agora-agents",
-            "X-Fern-SDK-Version": "v2.12.0",
+            "X-Fern-SDK-Version": "v2.13.0",
             **(self.get_custom_headers() or {}),
         }
         headers["Authorization"] = httpx.BasicAuth(self._get_username(), self._get_password())._auth_header
diff --git a/src/agora_agent/types/asr.py b/src/agora_agent/types/asr.py
index 838ea54..dc5461d 100644
--- a/src/agora_agent/types/asr.py
+++ b/src/agora_agent/types/asr.py
@@ -18,6 +18,7 @@
 from .google_asr_params import GoogleAsrParams
 from .microsoft_asr_params import MicrosoftAsrParams
 from .open_ai_asr_params import OpenAiAsrParams
+from .rtzr_asr_params import RtzrAsrParams
 from .sarvam_asr_params import SarvamAsrParams
 from .smallest_ai_asr_params import SmallestAiAsrParams
 from .speechmatics_asr_params import SpeechmaticsAsrParams
@@ -210,6 +211,21 @@ class Config:
             extra = pydantic.Extra.allow
 
 
+class Asr_Rtzr(UncheckedBaseModel):
+    vendor: typing.Literal["rtzr"] = "rtzr"
+    language: typing.Optional[AsrLanguage] = None
+    params: RtzrAsrParams
+
+    if IS_PYDANTIC_V2:
+        model_config: typing.ClassVar[pydantic.ConfigDict] = pydantic.ConfigDict(extra="allow", frozen=True)  # type: ignore # Pydantic v2
+    else:
+
+        class Config:
+            frozen = True
+            smart_union = True
+            extra = pydantic.Extra.allow
+
+
 class Asr_Xai(UncheckedBaseModel):
     vendor: typing.Literal["xai"] = "xai"
     language: typing.Optional[AsrLanguage] = None
@@ -299,6 +315,7 @@ class Config:
         Asr_Assemblyai,
         Asr_Speechmatics,
         Asr_Sarvam,
+        Asr_Rtzr,
         Asr_Xai,
         Asr_Xfyun,
         Asr_XfyunBigmodel,
diff --git a/src/agora_agent/types/deepgram_asr_params.py b/src/agora_agent/types/deepgram_asr_params.py
index 6688333..e1f3656 100644
--- a/src/agora_agent/types/deepgram_asr_params.py
+++ b/src/agora_agent/types/deepgram_asr_params.py
@@ -17,7 +17,7 @@ class DeepgramAsrParams(UncheckedBaseModel):
     WebSocket URL for Deepgram's streaming API
     """
 
-    key: str = pydantic.Field()
+    api_key: str = pydantic.Field()
     """
     Deepgram API key
     """
diff --git a/src/agora_agent/types/gemini_tts.py b/src/agora_agent/types/gemini_tts.py
new file mode 100644
index 0000000..c58a273
--- /dev/null
+++ b/src/agora_agent/types/gemini_tts.py
@@ -0,0 +1,29 @@
+# This file was auto-generated by Fern from our API Definition.
+
+import typing
+
+import pydantic
+from ..core.pydantic_utilities import IS_PYDANTIC_V2
+from ..core.unchecked_base_model import UncheckedBaseModel
+from .gemini_tts_params import GeminiTtsParams
+
+
+class GeminiTts(UncheckedBaseModel):
+    """
+    Gemini Text-to-Speech configuration.
+    """
+
+    params: GeminiTtsParams
+    skip_patterns: typing.Optional[typing.List[int]] = pydantic.Field(default=None)
+    """
+    Controls whether the TTS module skips bracketed content when reading LLM response text.
+    """
+
+    if IS_PYDANTIC_V2:
+        model_config: typing.ClassVar[pydantic.ConfigDict] = pydantic.ConfigDict(extra="allow", frozen=True)  # type: ignore # Pydantic v2
+    else:
+
+        class Config:
+            frozen = True
+            smart_union = True
+            extra = pydantic.Extra.allow
diff --git a/src/agora_agent/types/gemini_tts_params.py b/src/agora_agent/types/gemini_tts_params.py
new file mode 100644
index 0000000..70983fb
--- /dev/null
+++ b/src/agora_agent/types/gemini_tts_params.py
@@ -0,0 +1,42 @@
+# This file was auto-generated by Fern from our API Definition.
+
+import typing
+
+import pydantic
+from ..core.pydantic_utilities import IS_PYDANTIC_V2
+from ..core.unchecked_base_model import UncheckedBaseModel
+
+
+class GeminiTtsParams(UncheckedBaseModel):
+    """
+    Gemini TTS configuration parameters.
+    """
+
+    api_key: str = pydantic.Field()
+    """
+    Gemini API key.
+    """
+
+    model: str = pydantic.Field()
+    """
+    Gemini TTS model name.
+    """
+
+    voice: str = pydantic.Field()
+    """
+    Gemini voice name.
+    """
+
+    style: typing.Optional[str] = pydantic.Field(default=None)
+    """
+    Style instruction for the generated speech.
+    """
+
+    if IS_PYDANTIC_V2:
+        model_config: typing.ClassVar[pydantic.ConfigDict] = pydantic.ConfigDict(extra="allow", frozen=True)  # type: ignore # Pydantic v2
+    else:
+
+        class Config:
+            frozen = True
+            smart_union = True
+            extra = pydantic.Extra.allow
diff --git a/src/agora_agent/types/rtzr_asr.py b/src/agora_agent/types/rtzr_asr.py
new file mode 100644
index 0000000..e49023d
--- /dev/null
+++ b/src/agora_agent/types/rtzr_asr.py
@@ -0,0 +1,27 @@
+# This file was auto-generated by Fern from our API Definition.
+
+import typing
+
+import pydantic
+from ..core.pydantic_utilities import IS_PYDANTIC_V2
+from ..core.unchecked_base_model import UncheckedBaseModel
+from .asr_language import AsrLanguage
+from .rtzr_asr_params import RtzrAsrParams
+
+
+class RtzrAsr(UncheckedBaseModel):
+    """
+    RTZR ASR configuration.
+    """
+
+    language: typing.Optional[AsrLanguage] = None
+    params: RtzrAsrParams
+
+    if IS_PYDANTIC_V2:
+        model_config: typing.ClassVar[pydantic.ConfigDict] = pydantic.ConfigDict(extra="allow", frozen=True)  # type: ignore # Pydantic v2
+    else:
+
+        class Config:
+            frozen = True
+            smart_union = True
+            extra = pydantic.Extra.allow
diff --git a/src/agora_agent/types/rtzr_asr_params.py b/src/agora_agent/types/rtzr_asr_params.py
new file mode 100644
index 0000000..40f2baa
--- /dev/null
+++ b/src/agora_agent/types/rtzr_asr_params.py
@@ -0,0 +1,82 @@
+# This file was auto-generated by Fern from our API Definition.
+
+import typing
+
+import pydantic
+from ..core.pydantic_utilities import IS_PYDANTIC_V2
+from ..core.unchecked_base_model import UncheckedBaseModel
+
+
+class RtzrAsrParams(UncheckedBaseModel):
+    """
+    RTZR ASR configuration parameters.
+    """
+
+    client_id: str = pydantic.Field()
+    """
+    RTZR client ID.
+    """
+
+    client_secret: str = pydantic.Field()
+    """
+    RTZR client secret.
+    """
+
+    api_base: typing.Optional[str] = pydantic.Field(default=None)
+    """
+    RTZR API base URL.
+    """
+
+    model_name: typing.Optional[str] = pydantic.Field(default=None)
+    """
+    RTZR recognition model name.
+    """
+
+    language: typing.Optional[str] = pydantic.Field(default=None)
+    """
+    RTZR recognition language code. Defaults to Korean (`ko`).
+    """
+
+    sample_rate: typing.Optional[int] = pydantic.Field(default=None)
+    """
+    Input audio sample rate in Hz.
+    """
+
+    encoding: typing.Optional[str] = pydantic.Field(default=None)
+    """
+    Input audio encoding.
+    """
+
+    use_itn: typing.Optional[bool] = pydantic.Field(default=None)
+    """
+    Whether to enable inverse text normalization.
+    """
+
+    use_disfluency_filter: typing.Optional[bool] = pydantic.Field(default=None)
+    """
+    Whether to filter disfluencies such as stuttering.
+    """
+
+    use_profanity_filter: typing.Optional[bool] = pydantic.Field(default=None)
+    """
+    Whether to filter profanity.
+    """
+
+    use_punctuation: typing.Optional[bool] = pydantic.Field(default=None)
+    """
+    Whether to add punctuation to the recognized text.
+    """
+
+    keywords: typing.Optional[typing.List[str]] = pydantic.Field(default=None)
+    """
+    Keywords to improve recognition accuracy.
+    """
+
+    if IS_PYDANTIC_V2:
+        model_config: typing.ClassVar[pydantic.ConfigDict] = pydantic.ConfigDict(extra="allow", frozen=True)  # type: ignore # Pydantic v2
+    else:
+
+        class Config:
+            frozen = True
+            smart_union = True
+            extra = pydantic.Extra.allow
diff --git a/src/agora_agent/types/sarvam_tts_params.py b/src/agora_agent/types/sarvam_tts_params.py
index 855299f..d544dea 100644
--- a/src/agora_agent/types/sarvam_tts_params.py
+++ b/src/agora_agent/types/sarvam_tts_params.py
@@ -20,32 +20,42 @@ class SarvamTtsParams(UncheckedBaseModel):
 
     speaker: str = pydantic.Field()
     """
-    Voice ID (e.g., anushka, abhilash, karun, hitesh, manisha, vidya, arya)
+    Speaker voice to use.
     """
 
     target_language_code: SarvamTtsParamsTargetLanguageCode = pydantic.Field()
     """
-    Target language code (e.g., en-IN)
+    Target language code in BCP-47 format (e.g., `hi-IN`, `bn-IN`, `en-IN`).
     """
 
     pitch: typing.Optional[float] = pydantic.Field(default=None)
     """
-    Pitch adjustment for the voice
+    Pitch control for the `bulbul:v2` model.
     """
 
     pace: typing.Optional[float] = pydantic.Field(default=None)
     """
-    Speed of speech
+    Speech speed. Defaults to `1.0`.
     """
 
     loudness: typing.Optional[float] = pydantic.Field(default=None)
     """
-    Volume level of the speech
+    Audio loudness control for the `bulbul:v2` model.
     """
 
-    sample_rate: typing.Optional[float] = pydantic.Field(default=None)
+    speech_sample_rate: typing.Optional[int] = pydantic.Field(default=None)
     """
-    Audio sample rate in Hz
+    Output speech sample rate in Hz. Defaults to `24000`.
+    """
+
+    enable_preprocessing: typing.Optional[bool] = pydantic.Field(default=None)
+    """
+    Whether to normalize English words and numeric entities. Defaults to `false`.
+    """
+
+    model: typing.Optional[str] = pydantic.Field(default=None)
+    """
+    TTS model to use. Defaults to `bulbul:v3`.
     """
 
     if IS_PYDANTIC_V2:
diff --git a/src/agora_agent/types/tts.py b/src/agora_agent/types/tts.py
index ce737ad..e83d139 100644
--- a/src/agora_agent/types/tts.py
+++ b/src/agora_agent/types/tts.py
@@ -16,6 +16,7 @@
 from .deepgram_tts_params import DeepgramTtsParams
 from .eleven_labs_tts_params import ElevenLabsTtsParams
 from .fish_audio_tts_params import FishAudioTtsParams
+from .gemini_tts_params import GeminiTtsParams
 from .generic_http_tts_params import GenericHttpTtsParams
 from .google_tts_params import GoogleTtsParams
 from .gradium_tts_params import GradiumTtsParams
@@ -216,6 +217,21 @@ class Config:
             extra = pydantic.Extra.allow
 
 
+class Tts_Gemini(UncheckedBaseModel):
+    vendor: typing.Literal["gemini"] = "gemini"
+    params: GeminiTtsParams
+    skip_patterns: typing.Optional[typing.List[int]] = None
+
+    if IS_PYDANTIC_V2:
+        model_config: typing.ClassVar[pydantic.ConfigDict] = pydantic.ConfigDict(extra="allow", frozen=True)  # type: ignore # Pydantic v2
+    else:
+
+        class Config:
+            frozen = True
+            smart_union = True
+            extra = pydantic.Extra.allow
+
+
 class Tts_Amazon(UncheckedBaseModel):
     vendor: typing.Literal["amazon"] = "amazon"
     params: AmazonTtsParams
@@ -412,6 +428,7 @@ class Config:
         Tts_Rime,
         Tts_Fishaudio,
         Tts_Google,
+        Tts_Gemini,
         Tts_Amazon,
         Tts_Sarvam,
         Tts_GenericHttp,
diff --git a/tests/custom/test_agentkit_agent.py b/tests/custom/test_agentkit_agent.py
index d7cf47e..6d9cd5d 100644
--- a/tests/custom/test_agentkit_agent.py
+++ b/tests/custom/test_agentkit_agent.py
@@ -6,6 +6,7 @@
     LlmStyle,
     MllmConfig,
     MllmVendor,
+    SpeakConfig,
     SttConfig,
     SttVendor,
     TtsConfig,
@@ -85,6 +86,21 @@ def test_with_audio_scenario_preserves_existing_parameters():
     assert _parameter(agent.config, "audio_scenario") == "chorus"
 
 
+def test_with_parameters_serializes_speak_settings():
+    agent = Agent(test_client()).with_parameters({"speak": SpeakConfig(batch=False)})
+
+    properties = agent.to_properties(
+        channel="room",
+        agent_uid="1",
+        remote_uids=["100"],
+        token="token",
+        skip_vendor_validation=True,
+    )
+
+    assert properties.parameters.speak is not None
+    assert properties.parameters.speak.batch is False
+
+
 def test_enable_rtm_defaults_data_channel_to_rtm():
     properties = Agent(test_client(), advanced_features={"enable_rtm": True}).to_properties(
         channel="room",
diff --git a/tests/custom/test_anam_avatar.py b/tests/custom/test_anam_avatar.py
index a4fa395..4cb12ef 100644
--- a/tests/custom/test_anam_avatar.py
+++ b/tests/custom/test_anam_avatar.py
@@ -22,11 +22,48 @@ def test_anam_avatar_to_config_shape() -> None:
     assert is_anam_avatar(config)
 
 
+def test_anam_avatar_serializes_portrait_options_and_overrides_additional_params() -> None:
+    config = AnamAvatar(
+        api_key="anam-key",
+        avatar_id="anam-avatar",
+        avatar_model="cara_mk4",
+        video_width=720,
+        video_height=1280,
+        additional_params={
+            "avatar_model": "overridden-model",
+            "video_width": 1,
+            "video_height": 2,
+        },
+    ).to_config()
+
+    assert config["params"] == {
+        "api_key": "anam-key",
+        "avatar_id": "anam-avatar",
+        "avatar_model": "cara_mk4",
+        "video_width": 720,
+        "video_height": 1280,
+    }
+
+
+@pytest.mark.parametrize("kwargs", [{"video_width": 720}, {"video_height": 1280}])
+def test_anam_avatar_requires_video_dimensions_as_pair(kwargs: dict) -> None:
+    with pytest.raises(ValueError, match="Anam avatar requires video_width and video_height together"):
+        AnamAvatar(api_key="anam-key", avatar_id="anam-avatar", **kwargs)
+
+
 @pytest.mark.parametrize(
     ("params", "message"),
     [
         ({}, "Anam avatar requires api_key"),
         ({"api_key": "key"}, "Anam avatar requires avatar_id"),
+        (
+            {"api_key": "key", "avatar_id": "avatar", "video_width": 720},
+            "Anam avatar requires video_width and video_height together",
+        ),
+        (
+            {"api_key": "key", "avatar_id": "avatar", "video_height": 1280},
+            "Anam avatar requires video_width and video_height together",
+        ),
     ],
 )
 def test_validate_avatar_config_rejects_incomplete_anam(
diff --git a/tests/custom/test_gemini_tts.py b/tests/custom/test_gemini_tts.py
index c3a491a..64f8437 100644
--- a/tests/custom/test_gemini_tts.py
+++ b/tests/custom/test_gemini_tts.py
@@ -1,20 +1,31 @@
 import json
+from typing import get_args
 
 import httpx
 import pytest
 
 from agora_agent import Agent, Agora, Area, AsyncAgora, Gemini, GeminiSTT, GeminiTTS, GeminiTTSModels
 from agora_agent.agentkit.preview import PREVIEW_API_BASE_URL, required_preview_features
+from agora_agent.agentkit.preview import GeminiTTS as PreviewGeminiTTS
+from agora_agent.agentkit.preview.gemini_tts import GeminiTTS as LegacyGeminiTTS
+from agora_agent.agentkit.regional_agent import GlobalTTS
+from agora_agent.agentkit.vendors import GeminiTTS as VendorGeminiTTS
+from agora_agent.agentkit.vendors.base import BaseSTT, BaseTTS
+from agora_agent.agentkit.vendors.catalog import GLOBAL_VENDOR_NAMESPACE
+from agora_agent.agentkit.vendors.namespaces import GlobalTTSVendors
+from agora_agent.agentkit.vendors.region import CN_TTS_VENDORS, GLOBAL_TTS_VENDORS
+from agora_agent.agentkit.vendors.tts import GeminiTTS as ProductionGeminiTTS
+from agora_agent.types.tts import Tts_Gemini
 
-MODELS = [GeminiTTSModels.FLASH_38]
+MODELS = [GeminiTTSModels.FLASH_38, "future-tts-model"]
 
 
-def make_agent(client, model):
+def make_agent(client, model, tts_type=GeminiTTS):
     return (
         Agent(client=client)
         .with_stt(GeminiSTT(api_key="test-key"))
         .with_llm(Gemini(api_key="test-key", model="gemini-3.6-flash"))
-        .with_tts(GeminiTTS(api_key="test-key", model=model, voice="Puck", style="warm and reassuring"))
+        .with_tts(tts_type(api_key="test-key", model=model, voice="Puck", style="warm and reassuring"))
     )
 
 
@@ -26,58 +37,85 @@ def handle(request):
     return httpx.MockTransport(handle)
 
 
-def assert_requests(requests, model):
-    assert len(requests) == 4
+def assert_requests(requests, model, production_url):
+    assert len(requests) == 9
     assert json.loads(requests[0].content)["properties"]["tts"] == {
         "vendor": "gemini",
         "params": {"api_key": "test-key", "model": model, "voice": "Puck", "style": "warm and reassuring"},
     }
     for request in requests:
-        assert str(request.url).startswith(PREVIEW_API_BASE_URL)
-        assert request.headers["agora-feature"] == "gemini-live"
+        assert str(request.url).startswith(production_url + "/")
+        assert "agora-feature" not in request.headers
         assert request.headers["x-custom"] == "kept"
 
 
 @pytest.mark.parametrize("model", MODELS)
-def test_sync_lifecycle(model):
+@pytest.mark.parametrize("area", [Area.US, Area.EU, Area.AP])
+@pytest.mark.parametrize(
+    "tts_type", [GeminiTTS, PreviewGeminiTTS, LegacyGeminiTTS], ids=["root", "preview", "legacy-module"]
+)
+def test_sync_lifecycle(model, area, tts_type):
     requests = []
     with httpx.Client(transport=transport(requests)) as http_client:
         client = Agora(
-            area=Area.US,
+            area=area,
             app_id="0" * 32,
             app_certificate="1" * 32,
             httpx_client=http_client,
-            headers={"agora-feature": "", "x-custom": "kept"},
+            headers={"x-custom": "kept"},
         )
         production_url = client.get_current_url()
-        session = make_agent(client, model).create_session(channel="test", agent_uid="1", remote_uids=["100"])
+        session = make_agent(client, model, tts_type).create_session(channel="test", agent_uid="1", remote_uids=["100"])
         session.start()
         session.say("hello")
         session.interrupt()
+        session.think("think")
+        session.update({})
+        session.get_history()
+        session.get_info()
+        session.get_turns()
         session.stop()
-        assert_requests(requests, model)
+        assert_requests(requests, model, production_url)
         assert client.get_current_url() == production_url
-        assert client._client_wrapper.get_custom_headers()["agora-feature"] == ""
+        client.agents.get("0" * 32, "agent-1")
+        assert str(requests[-1].url).startswith(production_url + "/")
+        assert "agora-feature" not in requests[-1].headers
 
 
 @pytest.mark.asyncio
 @pytest.mark.parametrize("model", MODELS)
-async def test_async_lifecycle(model):
+@pytest.mark.parametrize("area", [Area.US, Area.EU, Area.AP])
+@pytest.mark.parametrize(
+    "tts_type", [GeminiTTS, PreviewGeminiTTS, LegacyGeminiTTS], ids=["root", "preview", "legacy-module"]
+)
+async def test_async_lifecycle(model, area, tts_type):
     requests = []
     async with httpx.AsyncClient(transport=transport(requests)) as http_client:
         client = AsyncAgora(
-            area=Area.US,
+            area=area,
             app_id="0" * 32,
             app_certificate="1" * 32,
             httpx_client=http_client,
-            headers={"agora-feature": "", "x-custom": "kept"},
+            headers={"x-custom": "kept"},
+        )
+        production_url = client.get_current_url()
+        session = make_agent(client, model, tts_type).create_async_session(
+            channel="test", agent_uid="1", remote_uids=["100"]
         )
-        session = make_agent(client, model).create_async_session(channel="test", agent_uid="1", remote_uids=["100"])
         await session.start()
         await session.say("hello")
         await session.interrupt()
+        await session.think("think")
+        await session.update({})
+        await session.get_history()
+        await session.get_info()
+        await session.get_turns()
         await session.stop()
-        assert_requests(requests, model)
+        assert_requests(requests, model, production_url)
+        assert client.get_current_url() == production_url
+        await client.agents.get("0" * 32, "agent-1")
+        assert str(requests[-1].url).startswith(production_url + "/")
+        assert "agora-feature" not in requests[-1].headers
 
 
 def test_defaults_and_raw_detection():
@@ -85,23 +123,37 @@ def test_defaults_and_raw_detection():
         "vendor": "gemini",
         "params": {"api_key": "test-key", "model": "gemini-3.8-flash-tts", "voice": "Puck"},
     }
-    assert required_preview_features({"tts": {"vendor": "gemini", "params": {"model": "future-model"}}}) == [
-        "gemini-live"
-    ]
-    assert set(required_preview_features({"tts": {"vendor": "gemini"}, "mllm": {"vendor": "openai_gpt_live"}})) == {
-        "gemini-live",
-    }
+    assert required_preview_features({"tts": {"vendor": "gemini", "params": {"model": "future-model"}}}) == []
+    assert required_preview_features({"tts": {"vendor": "gemini"}, "mllm": {"vendor": "openai_gpt_live"}}) == []
     assert required_preview_features({"asr": {"vendor": "gemini"}, "tts": {"vendor": "google"}}) == []
 
 
+def test_additional_params_are_merged_before_named_options():
+    additional_params = {"temperature": 0.7, "style": "overridden", "voice": "overridden"}
+    config = GeminiTTS(
+        api_key="test-key", style="warm", additional_params=additional_params, skip_patterns=[1]
+    ).to_config()
+    assert config["params"] == {
+        "temperature": 0.7,
+        "api_key": "test-key",
+        "model": "gemini-3.8-flash-tts",
+        "voice": "Puck",
+        "style": "warm",
+    }
+    assert config["skip_patterns"] == [1]
+    assert additional_params == {"temperature": 0.7, "style": "overridden", "voice": "overridden"}
+    generated = Tts_Gemini.model_validate(config)
+    assert generated.params.style == "warm"
+    assert generated.params.temperature == 0.7
+
+
 @pytest.mark.parametrize("field", ["api_key", "model", "voice"])
 def test_rejects_blank_fields(field):
     with pytest.raises(ValueError):
         GeminiTTS(**{**{"api_key": "test-key"}, field: "  "})
 
 
-@pytest.mark.parametrize("header", ["agora-feature", "Agora-Feature", "AGORA-FEATURE"])
-def test_preview_gate_survives_raw_per_call_headers(header):
+def test_production_raw_calls_preserve_per_call_headers():
     requests = []
     with httpx.Client(transport=transport(requests)) as http_client:
         client = Agora(
@@ -109,24 +161,24 @@ def test_preview_gate_survives_raw_per_call_headers(header):
             app_id="0" * 32,
             app_certificate="1" * 32,
             httpx_client=http_client,
-            headers={header: "caller-value"},
+            headers={"x-custom": "caller-value"},
         )
         session = make_agent(client, MODELS[0]).create_session(channel="test", agent_uid="1", remote_uids=["100"])
         session.start()
-        headers = {header: "", "x-custom": "preserved"}
+        headers = {"x-request": "preserved"}
         session.raw.get("0" * 32, "agent-1", request_options={"additional_headers": headers})
         for req in requests:
-            assert req.headers.get_list("agora-feature") == ["gemini-live"]
-        assert requests[-1].headers["x-custom"] == "preserved"
-        assert headers[header] == ""
-        assert client._client_wrapper.get_custom_headers()[header] == "caller-value"
+            assert "agora-feature" not in req.headers
+            assert str(req.url).startswith(client.get_current_url() + "/")
+        assert requests[-1].headers["x-request"] == "preserved"
+        assert headers == {"x-request": "preserved"}
         client.agents.get("0" * 32, "agent-1")
         assert not str(requests[-1].url).startswith(PREVIEW_API_BASE_URL)
-        assert requests[-1].headers["agora-feature"] == "caller-value"
+        assert requests[-1].headers["x-custom"] == "caller-value"
 
 
 @pytest.mark.asyncio
-async def test_async_preview_gate_survives_raw_per_call_headers():
+async def test_async_production_raw_calls_preserve_per_call_headers():
     requests = []
     async with httpx.AsyncClient(transport=transport(requests)) as http_client:
         client = AsyncAgora(
@@ -134,13 +186,95 @@ async def test_async_preview_gate_survives_raw_per_call_headers():
             app_id="0" * 32,
             app_certificate="1" * 32,
             httpx_client=http_client,
-            headers={"Agora-Feature": "caller-value"},
+            headers={"x-custom": "caller-value"},
         )
         session = make_agent(client, MODELS[0]).create_async_session(channel="test", agent_uid="1", remote_uids=["100"])
         await session.start()
-        await session.raw.get("0" * 32, "agent-1", request_options={"additional_headers": {"AGORA-FEATURE": ""}})
+        await session.raw.get("0" * 32, "agent-1", request_options={"additional_headers": {"x-request": "preserved"}})
         for req in requests:
-            assert req.headers.get_list("agora-feature") == ["gemini-live"]
+            assert "agora-feature" not in req.headers
+            assert str(req.url).startswith(client.get_current_url() + "/")
+        assert requests[-1].headers["x-request"] == "preserved"
         await client.agents.get("0" * 32, "agent-1")
         assert not str(requests[-1].url).startswith(PREVIEW_API_BASE_URL)
-        assert requests[-1].headers["agora-feature"] == "caller-value"
+        assert requests[-1].headers["x-custom"] == "caller-value"
+
+
+def test_preview_imports_alias_production_and_register_in_global_catalog():
+    from agora_agent.agentkit.preview import GeminiTTSModels as PreviewModels
+    from agora_agent.agentkit.preview.gemini_tts import GeminiTTSModels as LegacyModels
+
+    assert GeminiTTS is VendorGeminiTTS is ProductionGeminiTTS is PreviewGeminiTTS is LegacyGeminiTTS
+    assert GeminiTTSModels is PreviewModels is LegacyModels
+    assert GLOBAL_VENDOR_NAMESPACE.tts["gemini"] is GeminiTTS
+    assert GlobalTTSVendors.gemini is GeminiTTS
+    assert GeminiTTS in get_args(GlobalTTS)
+    assert "gemini" in GLOBAL_TTS_VENDORS
+    assert "gemini" not in CN_TTS_VENDORS
+
+
+def test_raw_config_starts_on_production_and_drops_none():
+    config = {
+        "vendor": "gemini",
+        "params": {"api_key": "test-key", "model": "future-tts-model", "voice": "Puck", "style": None},
+    }
+
+    class RawGeminiTTS(BaseTTS):
+        def to_config(self):
+            return config
+
+    requests = []
+    with httpx.Client(transport=transport(requests)) as http_client:
+        client = Agora(area=Area.EU, app_id="0" * 32, app_certificate="1" * 32, httpx_client=http_client)
+        session = make_agent(client, MODELS[0]).with_tts(RawGeminiTTS()).create_session(
+            channel="raw", agent_uid="1", remote_uids=["100"]
+        )
+        session.start()
+        session.stop()
+        assert json.loads(requests[0].content)["properties"]["tts"] == {
+            "vendor": "gemini",
+            "params": {"api_key": "test-key", "model": "future-tts-model", "voice": "Puck"},
+        }
+        assert config["params"]["style"] is None
+        for request in requests:
+            assert str(request.url).startswith(client.get_current_url() + "/")
+            assert "agora-feature" not in request.headers
+
+
+def test_gemini_tts_schema_compatibility_still_validates_other_fields():
+    class InvalidASR(BaseSTT):
+        def to_config(self):
+            return {"vendor": "gemini", "params": {"api_key": "test-key"}}
+
+    requests = []
+    with httpx.Client(transport=transport(requests)) as http_client:
+        client = Agora(area=Area.US, app_id="0" * 32, app_certificate="1" * 32, httpx_client=http_client)
+        session = make_agent(client, MODELS[0]).with_stt(InvalidASR()).create_session(
+            channel="invalid", agent_uid="1", remote_uids=["100"]
+        )
+        with pytest.raises(ValueError):
+            session.start()
+        assert requests == []
+
+
+def test_gemini_tts_can_share_a_session_with_a_registered_preview_provider(monkeypatch):
+    from agora_agent.agentkit.preview import client as preview_client
+
+    monkeypatch.setitem(preview_client._PREVIEW_FEATURES_BY_CATEGORY["asr"], "future_asr", "future-asr")
+
+    class PreviewASR(BaseSTT):
+        def to_config(self):
+            return {"vendor": "future_asr", "params": {"model": "future-model"}}
+
+    requests = []
+    with httpx.Client(transport=transport(requests)) as http_client:
+        client = Agora(area=Area.US, app_id="0" * 32, app_certificate="1" * 32, httpx_client=http_client)
+        session = make_agent(client, MODELS[0]).with_stt(PreviewASR()).create_session(
+            channel="mixed", agent_uid="1", remote_uids=["100"]
+        )
+        session.start()
+        session.stop()
+        assert json.loads(requests[0].content)["properties"]["tts"]["vendor"] == "gemini"
+        for request in requests:
+            assert str(request.url).startswith(PREVIEW_API_BASE_URL)
+            assert request.headers["agora-feature"] == "future-asr"
diff --git a/tests/custom/test_regional_vendors.py b/tests/custom/test_regional_vendors.py
index 92ff328..10516a4 100644
--- a/tests/custom/test_regional_vendors.py
+++ b/tests/custom/test_regional_vendors.py
@@ -13,6 +13,7 @@
     MiniMaxTTS,
     MistralTTS,
     OpenAI,
+    RtzrSTT,
     SmallestAISTT,
     SmallestAITTS,
     SpatiusAvatar,
@@ -229,6 +230,14 @@ def test_xai_grok_remains_mllm_vendor() -> None:
     assert agent.mllm is not None and agent.mllm["vendor"] == "xai"
 
 
+def test_rtzr_is_registered_as_a_global_stt_vendor() -> None:
+    assert "rtzr" in GLOBAL_ASR_VENDORS
+    assert "rtzr" not in CN_ASR_VENDORS
+    assert GLOBAL_VENDOR_NAMESPACE.asr["rtzr"] is RtzrSTT
+    assert GlobalSTTVendors.rtzr is RtzrSTT
+    assert RtzrSTT in get_args(GlobalSTT)
+
+
 def test_smallest_ai_is_registered_as_global_only() -> None:
     assert "smallestai" in GLOBAL_ASR_VENDORS
     assert "smallestai" in GLOBAL_TTS_VENDORS
diff --git a/tests/custom/test_request_body.py b/tests/custom/test_request_body.py
index f5bab70..3106870 100644
--- a/tests/custom/test_request_body.py
+++ b/tests/custom/test_request_body.py
@@ -323,7 +323,7 @@ def test_byok_pipeline_full_properties_shape() -> None:
     # ASR
     asr = props["asr"]
     assert asr["vendor"] == "deepgram"
-    assert asr["params"]["key"] == "dg-key"
+    assert asr["params"]["api_key"] == "dg-key"
     assert asr["params"]["model"] == "nova-2"
     assert asr["params"]["language"] == "en"
 
@@ -630,7 +630,7 @@ def test_6b_tts_preset_with_byok_llm_and_asr() -> None:
 
     properties = dump(call["properties"])
     # BYOK ASR: key and model both retained (nothing stripped for BYOK path)
-    assert properties["asr"]["params"]["key"] == "dg-key"
+    assert properties["asr"]["params"]["api_key"] == "dg-key"
     assert properties["asr"]["params"]["model"] == "nova-2"
     # BYOK LLM key retained
     assert properties["llm"]["api_key"] == "openai-key"
@@ -779,7 +779,7 @@ def test_byok_deepgram_stt_params() -> None:
     )
     props = build_properties(agent, allow_missing={"llm", "tts"})
     assert props["asr"]["vendor"] == "deepgram"
-    assert props["asr"]["params"]["key"] == "dg-key"
+    assert props["asr"]["params"]["api_key"] == "dg-key"
     assert props["asr"]["params"]["model"] == "nova-2"
     assert props["asr"]["params"]["language"] == "en"
 
@@ -1163,7 +1163,7 @@ def test_byok_sarvam_tts_params() -> None:
             key="sarvam-key",
             speaker="anushka",
             target_language_code="en-IN",
-            sample_rate=24000,
+            speech_sample_rate=24000,
         )
     )
     props = build_properties(agent, allow_missing={"asr", "llm"})
diff --git a/tests/custom/test_stt_language.py b/tests/custom/test_stt_language.py
index 27ce63f..9e2b8be 100644
--- a/tests/custom/test_stt_language.py
+++ b/tests/custom/test_stt_language.py
@@ -11,6 +11,7 @@
     MicrosoftSTT,
     OpenAI,
     OpenAISTT,
+    RtzrSTT,
     SarvamSTT,
     SpeechmaticsSTT,
     TurnDetectionConfig,
@@ -110,13 +111,13 @@ def test_stt_vendor_params_match_documented_shapes() -> None:
         DeepgramSTT(model="enhanced")
 
     assert DeepgramSTT(api_key="dg-key", language="en").to_config()["params"] == {
-        "key": "dg-key",
+        "api_key": "dg-key",
         "language": "en",
     }
 
-    # api_key → wire key "key"; keyterm passes through unchanged
+    # api_key uses the generated Deepgram ASR field name; keyterm passes through unchanged
     assert DeepgramSTT(api_key="dg-key", model="nova-3", language="en", keyterm="term").to_config()["params"] == {
-        "key": "dg-key",
+        "api_key": "dg-key",
         "model": "nova-3",
         "language": "en",
         "keyterm": "term",
@@ -188,6 +189,55 @@ def test_stt_vendor_params_match_documented_shapes() -> None:
     }
 
 
+def test_rtzr_stt_serializes_generated_params() -> None:
+    assert RtzrSTT(
+        client_id="client-id",
+        client_secret="client-secret",
+        api_base="https://rtzr.example.com",
+        model_name="general",
+        language="ko",
+        sample_rate=16000,
+        encoding="pcm_s16le",
+        use_itn=True,
+        use_disfluency_filter=False,
+        use_profanity_filter=True,
+        use_punctuation=True,
+        keywords=["Agora"],
+    ).to_config() == {
+        "vendor": "rtzr",
+        "params": {
+            "client_id": "client-id",
+            "client_secret": "client-secret",
+            "api_base": "https://rtzr.example.com",
+            "model_name": "general",
+            "language": "ko",
+            "sample_rate": 16000,
+            "encoding": "pcm_s16le",
+            "use_itn": True,
+            "use_disfluency_filter": False,
+            "use_profanity_filter": True,
+            "use_punctuation": True,
+            "keywords": ["Agora"],
+        },
+    }
+
+
+def test_rtzr_stt_reaches_generated_request_properties() -> None:
+    props = properties(
+        base_agent().with_stt(RtzrSTT(client_id="client-id", client_secret="client-secret", language="ko"))
+    )
+
+    assert props["asr"] == {
+        "vendor": "rtzr",
+        "language": "en-US",
+        "params": {
+            "client_id": "client-id",
+            "client_secret": "client-secret",
+            "language": "ko",
+        },
+    }
+
+
 def test_assemblyai_params_stay_nested_and_asr_language_comes_from_turn_detection() -> None:
     props = properties(
         Agent(test_client(), turn_detection=TurnDetectionConfig(language="fr-FR"))
diff --git a/tests/custom/test_tts_vendors.py b/tests/custom/test_tts_vendors.py
index 43388bb..6abf094 100644
--- a/tests/custom/test_tts_vendors.py
+++ b/tests/custom/test_tts_vendors.py
@@ -198,11 +198,22 @@ def test_tts_vendor_params_match_generated_core_shapes() -> None:
         "language_boost": "auto",
     }
 
-    assert SarvamTTS(key="sarvam-key", speaker="anushka", target_language_code="en-IN", sample_rate=24000).to_config()["params"] == {
+    assert SarvamTTS(
+        key="sarvam-key",
+        speaker="anushka",
+        target_language_code="en-IN",
+        speech_sample_rate=24000,
+        enable_preprocessing=True,
+        model="bulbul:v3",
+        additional_params={"custom_option": "value", "model": "overridden"},
+    ).to_config()["params"] == {
         "api_subscription_key": "sarvam-key",
         "speaker": "anushka",
         "target_language_code": "en-IN",
-        "sample_rate": 24000,
+        "speech_sample_rate": 24000,
+        "enable_preprocessing": True,
+        "model": "bulbul:v3",
+        "custom_option": "value",
     }
 
     assert MurfTTS(