Aliyun plugin for LiveKit Agents

Realtime speech-to-text, text-to-speech, and LLM support for LiveKit Agents using Aliyun DashScope.

The plugin exposes:

• livekit.plugins.aliyun.STT
• livekit.plugins.aliyun.TTS
• livekit.plugins.aliyun.LLM

Installation

pip install livekit-plugins-alicloud

Pre-requisites

You'll need an API key from Aliyun. It can be set as an environment variable: DASHSCOPE_API_KEY

Quick Start

from dotenv import load_dotenv

from livekit.agents import Agent, AgentSession
from livekit.plugins import aliyun

load_dotenv()

session = AgentSession(
    llm=aliyun.LLM(),
    stt=aliyun.STT(),
    tts=aliyun.TTS(),
)

agent = Agent(instructions="Reply briefly and clearly.")

LLM

Common Options

from livekit.plugins import aliyun

llm = aliyun.LLM(
    model="qwen-plus",
    enable_thinking=False,
)

• api_key: DashScope API key. Overrides the environment variable.
• model: DashScope OpenAI-compatible chat model name. Defaults to qwen-plus.
• base_url: Override the OpenAI-compatible endpoint. Defaults to https://dashscope.aliyuncs.com/compatible-mode/v1.
• enable_thinking: Whether to enable Qwen deep-thinking mode. Defaults to False.
• thinking_budget: Optional token budget for thinking mode. Requires enable_thinking=True.
• max_tokens: Optional maximum output token count.
• seed: Optional deterministic sampling seed.
• tool_choice: Supports auto, none, and forcing a specific function tool. The required option is not supported.
• preserve_thinking: Not supported in this version. Reasoning content is not persisted or returned as ordinary text.

STT

Common Options

from livekit.plugins import aliyun

stt = aliyun.STT(
    model="qwen3-asr-flash-realtime",
    language="zh",
    sample_rate=16000,
    interim_results=True,
)

• api_key: DashScope API key. Overrides the environment variable.
• model: Realtime ASR model name.
• language: Optional language code such as zh, yue, or en.
• sample_rate: Input sample rate. Typical value is 16000.
• interim_results: Whether to emit interim transcripts.
• turn_detection: Server-side VAD settings. Set it to None to use manual mode.

Manual Mode

To disable server-side turn detection and use manual mode:

from livekit.plugins import aliyun

stt = aliyun.STT(turn_detection=None)

TTS

Common Options

from livekit.plugins import aliyun

tts = aliyun.TTS(
    model="qwen3-tts-flash-realtime",
    voice="Cherry",
    language_type="Chinese",
)

• api_key: DashScope API key. Overrides the environment variable.
• model: Realtime TTS model name.
• voice: Realtime TTS voice name.
• language_type: One of Auto, Chinese, English, German, Italian, Portuguese, Spanish, Japanese, Korean, French, or Russian.
• base_url: Override the websocket endpoint. For the international region, use wss://dashscope-intl.aliyuncs.com/api-ws/v1/realtime.