# 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](../configuration.md));
[apply changes for your run mode](../operations/gateway.md#applying-configuration-changes): restart foreground runs, use `gateway restart` for an installed service, or click Apply in Desktop.

## Credentials and endpoint

The default provider is DashScope (`QWEN_AUDIO_REALTIME_PROVIDER=dashscope`):

```dotenv
DASHSCOPE_API_KEY=your-key
```

| Setting | Default | Description |
| --- | --- | --- |
| `DASHSCOPE_API_KEY` | — | Model Studio API key, shared by the realtime frontend and other gateway features |
| `QWEN_AUDIO_REALTIME_API_KEY` | Empty | Higher-priority alias of `DASHSCOPE_API_KEY` for the realtime frontend only |
| `QWEN_AUDIO_REALTIME_BASE_URL` / `QWEN_AUDIO_REALTIME_URL` | Empty | Override the DashScope Realtime endpoint (private deployment or proxy) |
| `DASHSCOPE_WORKSPACE_ID` | Empty | Switch to a dedicated Model Studio workspace endpoint |

Other frontends include [Speech-to-Speech](../voice-frontends/speech-to-speech.md) and
[ModelBest](../voice-frontends/minicpm-o.md), whose MiniCPM-o 4.5 endpoint may be local or hosted.
A custom provider implements
the provider contract; see [Custom Provider](../voice-frontends/custom-provider.md).

MiniCPM-o's public audio Realtime transport currently supports continuous
audio input and text/audio output, but not conversation items, structured
Function Calling, or input transcription. It is therefore a realtime voice-chat
frontend rather than a frontend for backend-Agent orchestration.

Frontend tools are configured separately: Web Search (`QWEN_AUDIO_WEB_SEARCH_PROVIDER`,
see [Configuration](../configuration.md)), and general chatbot tools through the
[Frontend MCP client](../reference/frontend-mcp.md), the
[Frontend OpenAPI adapter](../reference/frontend-openapi.md), or a
[Frontend Profile](../reference/frontend-profile.md).

## 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
# For an installed user background service only:
qwenaudio gateway restart
```

The exact supported IDs are:

| Model | Model input | Model output | Realtime transport |
| --- | --- | --- | --- |
| `qwen3.5-omni-flash-realtime` | text, audio, image/video frames | text, audio | text, audio, live JPEG frames |
| `qwen3.5-omni-plus-realtime` | text, audio, image/video frames | text, audio | text, audio, live JPEG frames |
| `qwen-audio-3.0-realtime-plus` (default) | text, audio | text, audio | text, audio |
| `qwen-audio-3.0-realtime-flash` | text, audio | text, audio | text, audio |

All four profiles support Function Calling. Model capability remains distinct from transport:
Omni accepts the WebUI's capability-negotiated live JPEG stream, while ordinary uploaded images
continue through the attachment path. Desktop and TUI do not capture live frames. Clients read
the authoritative profile from Gateway health. 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.