Skip to content

Frontend Configuration

The voice frontend is the realtime speech model the Gateway connects to. All settings on this page live in the user configuration file (~/.config/qwaudio/config.env, see Configuration); apply changes with qwenaudio gateway restart.

Credentials and endpoint

The default provider is DashScope (QWEN_AUDIO_REALTIME_PROVIDER=dashscope):

dotenv
DASHSCOPE_API_KEY=your-key
SettingDefaultDescription
DASHSCOPE_API_KEYModel Studio API key, shared by the realtime frontend and other gateway features
QWEN_AUDIO_REALTIME_API_KEYEmptyHigher-priority alias of DASHSCOPE_API_KEY for the realtime frontend only
QWEN_AUDIO_REALTIME_BASE_URL / QWEN_AUDIO_REALTIME_URLEmptyOverride the DashScope Realtime endpoint (private deployment or proxy)
DASHSCOPE_WORKSPACE_IDEmptySwitch to a dedicated Model Studio workspace endpoint

A fully local frontend is available via QWEN_AUDIO_REALTIME_PROVIDER=speech-to-speech; see Speech-to-Speech. A custom provider implements the provider contract; see Custom Provider.

Frontend tools are configured separately: Web Search (QWEN_AUDIO_WEB_SEARCH_PROVIDER, see Configuration), and general chatbot tools through the Frontend MCP client, the Frontend OpenAPI adapter, or a Frontend Profile.

Realtime model selection

One Gateway owns one active Realtime model. The Desktop settings page can configure the model for a locally owned Gateway, and the CLI provides the equivalent commands:

bash
qwenaudio config show
qwenaudio config set --realtime-model qwen3.5-omni-flash-realtime
qwenaudio gateway restart

The exact supported IDs are:

ModelModel inputModel outputCurrent client transport
qwen3.5-omni-flash-realtimetext, audio, imagetext, audiotext, audio
qwen3.5-omni-plus-realtimetext, audio, imagetext, audiotext, audio
qwen-audio-3.0-realtime-plus (default)text, audiotext, audiotext, audio
qwen-audio-3.0-realtime-flashtext, audiotext, audiotext, audio

All four profiles support Function Calling. Model capability is not the same as an implemented client transport: JPEG observation frames and native video are both disabled in this release. WebUI and TUI read the authoritative profile from Gateway health and only display it. Separate clients cannot select conflicting models on one Gateway. A Desktop attached to a borrowed Gateway, or a later CLI runtime using a conflicting configured model, refuses the mismatch instead of silently changing the running service. To roll back, set the legacy ID above and restart the Gateway.