AgoraIO · seymourtang · Sep 14, 2026 · Sep 14, 2026 · Sep 14, 2026 · Sep 14, 2026
diff --git a/.fern/metadata.json b/.fern/metadata.json
@@ -14,5 +14,5 @@
     },
     "exclude_types_from_init_exports": true
   },
-  "originGitCommit": "3fdb45ab45802742325afca544fdc22f5ae58db8"
+  "originGitCommit": "26314a9d3199f5e135302c3c3645b6bbb8f556d4"
 }
diff --git a/.fern/replay.lock b/.fern/replay.lock
diff --git a/changelog.md b/changelog.md
@@ -4,6 +4,12 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/).
 
+## [v2.10.0] — 2026-09-16
+
+### Changed
+
+- **Gemini Live production routing and compatibility** — Gemini 3.8 sessions now use the normal production gateway without a preview feature header. Existing imports from `agora_agent.agentkit.preview` remain compatible and resolve to the production `GeminiLive` implementation and model constants.
+
 ## [v2.9.0] — 2026-09-15
 
 ### Added
@@ -19,6 +25,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/).
 
 ### Changed
 
+- **OpenAI GPT Live production routing** — moved `OpenAIGPTLive` to the production MLLM vendors and removed its preview registration. Existing imports from `agora_agent.agentkit.preview` remain compatible and now route to Production automatically.
 - **OpenAI GPT Live defaults** — `OpenAIGPTLive` now defaults to the shortened `gpt-live-1` model name and omits the alpha selector. Explicit `model` and `alpha_selector` options remain available for future preview contracts.
 
 ## [v2.8.0] — 2026-09-10

diff --git a/compat/agora-agent-server-sdk/pyproject.toml b/compat/agora-agent-server-sdk/pyproject.toml
@@ -3,7 +3,7 @@ name = "agora-agent-server-sdk"
 
 [tool.poetry]
 name = "agora-agent-server-sdk"
-version = "v2.9.0"
+version = "v2.10.0"
 description = "Compatibility shim for the renamed agora-agents package."
 readme = "README.md"
 authors = []
@@ -35,7 +35,7 @@ Repository = 'https://github.com/AgoraIO/agora-agents-python'
 
 [tool.poetry.dependencies]
 python = "^3.8"
-agora-agents = ">=2.9.0,<3.0.0"
+agora-agents = ">=2.10.0,<3.0.0"
 
 [build-system]
 requires = ["poetry-core"]

diff --git a/docs/concepts/vendors.md b/docs/concepts/vendors.md
@@ -158,9 +158,9 @@ Used with `agent.with_mllm()` for the [MLLM flow](../guides/mllm-flow.md). These
 | Class | Provider | Area | Required Parameters |
 |---|---|---|---|
 | `OpenAIRealtime` | OpenAI Realtime | Global | `api_key`; optional `turn_detection` |
-| `OpenAIGPTLive` (preview) | OpenAI GPT Live | Global | `api_key`; optional `greeting` |
+| `OpenAIGPTLive` | OpenAI GPT Live | Global | `api_key`; optional `greeting` |
 | `AzureOpenAIRealtime` | Azure OpenAI Realtime | Global | `api_key`, `url`, `turn_detection`; optional `max_history` |
-| `GeminiLive` | Google Gemini Live API | Global | `api_key`; `model` defaults to `models/gemini-3.8-live`. The two 3.8 IDs use preview routing; older IDs use production. |
+| `GeminiLive` | Google Gemini Live API | Global | `api_key`; `model` defaults to `models/gemini-3.8-live`. All model IDs use production routing. |
 | `VertexAI` | Vertex AI (Gemini Live) | Global | `model`, `project_id`, `location`, `adc_credentials_string`; optional `turn_detection` |
 | `XaiGrok` | xAI Grok (`mllm.vendor`: `xai`) | Global | `api_key`; optional `voice`, `language`, `sample_rate`, `turn_detection` |
 | `QwenOmni` | Alibaba Cloud Qwen Omni Realtime | CN | `api_key`, `url`; optional `turn_detection` |

diff --git a/docs/guides/mllm-flow.md b/docs/guides/mllm-flow.md
@@ -95,7 +95,7 @@ asyncio.run(main())
 
 ## Gemini Live
 
-Use `GeminiLive` for the existing Gemini Live models and both Gemini 3.8 models. The 3.8 IDs select the preview route automatically; Extended Thinking also accepts `thinking_level`. See the [Preview Endpoint guide](./preview-endpoint.md).
+Use `GeminiLive` for the existing Gemini Live models and both Gemini 3.8 models. All use the production route; Extended Thinking also accepts `thinking_level`.
 
 Gemini Live uses a Google AI API key:
 

diff --git a/docs/guides/openai-gpt-live-v3.md b/docs/guides/openai-gpt-live-v3.md
@@ -1,6 +1,6 @@
-# GPT Live v3 preview
+# GPT Live v3
 
-This preview targets `gpt-live-1` on `/v1/live/sessions`. Use an alpha-enabled OpenAI key. Do not use this alpha for production traffic.
+GPT Live targets `gpt-live-1` on `/v1/live/sessions` through the production gateway. Existing imports from `agora_agent.agentkit.preview` remain supported as compatibility aliases.
 
 <!-- snippet: fragment -->
 ```python
@@ -51,15 +51,15 @@ GPT Live places MCP at `properties.mllm.mcp_servers`, the tool gate at `properti
 }
 ```
 
-This fragment omits the normal name, channel, token and UID fields populated by the SDK session. The preview route and `agora-feature: live-models` gate are selected automatically from the vendor.
+This fragment omits the normal name, channel, token and UID fields populated by the SDK session. The production route is selected automatically from the vendor.
 
 The SDK omits `alpha_selector` by default. Set it only when a future preview contract requires an `OpenAI-Alpha` selector.
 
 ## Silence and backend rollout
 
-Keep silence settings in the existing agent parameters builder, never in vendor params. The public API spelling is `silence_config`, with `{timeout_ms, action, content}`. The supplied extension contract describes internal `parameters.main.silence` and supports `action: "think"`; the public documentation currently says `silence_config` does not apply to MLLM. Serialization is covered by tests, but the public documentation does not establish that the preview allocator maps it to GPT Live's internal MAIN setting. Confirm that backend mapping before relying on silence nudges. The SDK does not invent a new public `main` field.
+Keep silence settings in the existing agent parameters builder, never in vendor params. The public API spelling is `silence_config`, with `{timeout_ms, action, content}`. The supplied extension contract describes internal `parameters.main.silence` and supports `action: "think"`; the public documentation currently says `silence_config` does not apply to MLLM. Serialization is covered by tests, but the public documentation does not establish that the allocator maps it to GPT Live's internal MAIN setting. Confirm that backend mapping before relying on silence nudges. The SDK does not invent a new public `main` field.
 
-## Preview-only options
+## Compatibility options
 
 The supplied backend contract marks custom voice objects, `responses_params`, and first-class context management as pending PR #1522. The SDK does not expose typed options for those fields. Use raw params only once your target backend supports that PR. Until then, context management is reachable via `session_params.context_management`.
 

diff --git a/docs/guides/preview-endpoint.md b/docs/guides/preview-endpoint.md
@@ -10,10 +10,8 @@ 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 uses the `live-models` feature. Gemini 3.8 MLLMs use `gemini-live`.
-Gemini STT has graduated to
-production and uses the normal regional endpoint. Existing imports of `GeminiSTT` and `GeminiSTTModels` from
-`agora_agent.agentkit.preview` remain supported as compatibility aliases.
+OpenAI GPT Live, Gemini STT, and Gemini Live 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
 from agora_agent import Agent, OpenAIGPTLive
@@ -26,16 +24,13 @@ session = (
 agent_id = session.start()
 ```
 
-This session uses the preview base URL and sends `agora-feature: live-models`. A session using `GeminiSTT` uses the
-client's normal GA regional endpoint without that header.
+Preview providers use the preview base URL and `agora-feature` gate. Sessions using `GeminiSTT`, `GeminiLive`, or GPT
+Live use the client's normal production regional endpoint without that header.
 
-Use the single `GeminiLive(api_key=..., model=...)` class with `with_mllm`.
-The model IDs are `models/gemini-3.8-live` and
-`models/gemini-3.8-live-extended-thinking`; the low-latency ID is the default.
-Set `thinking_level="medium"` for extended thinking. `GeminiLive` sends it
-only for the extended-thinking ID. The Gemini
+Use the single production `GeminiLive(api_key=..., model=...)` class with `with_mllm`. The model IDs are
+`models/gemini-3.8-live` and `models/gemini-3.8-live-extended-thinking`; the low-latency ID is the default. Set
+`thinking_level="medium"` for extended thinking. `GeminiLive` sends it only for the extended-thinking ID. The Gemini
 credential is sent once as `mllm.api_key`, never as `mllm.params.api_key`.
-Gemini sessions send `agora-feature: gemini-live`, while GPT Live retains `live-models`.
 
 ## Session-scoped routing
 
@@ -61,7 +56,7 @@ The registry is keyed first by request category and then by the serialized vendo
 ```python
 _PREVIEW_FEATURES_BY_CATEGORY = {
     "asr": {"new_vendor": "new-vendor-feature"},
-    "mllm": {"openai_gpt_live": "live-models"},
+    "mllm": {},
 }
 ```
 

diff --git a/docs/reference/agent.md b/docs/reference/agent.md
@@ -228,7 +228,7 @@ filler_words = FillerWordsConfig(
 agent = agent.with_filler_words(filler_words)
 ```
 
-`generated_config` is optional. Its `llm_provider`, `prompt`, and `fallback_strategy` fields are optional as well; when omitted, the service uses its defaults.
+`generated_config` is optional. Its `llm_provider`, `prompt`, and `fallback_strategy` fields are optional as well; when omitted, the service uses its defaults. Set `context_message_limit` to cap the number of recent messages and `history_character_limit` to cap the number of conversation-history characters passed to the filler-word model.
 
 ## `create_session()`