Custom API Configuration
If you don't want to use the built-in translation service and instead wish to connect to third-party translation platforms using your own API keys, you can configure them here. Both traditional machine translation services and large language models are supported for custom integration.
Configuration Types
In the Translation Engine settings, the software provides three translation modes:
- Built-in Translation Service: Ready to use out of the box, no API key configuration required.
- Custom: Use your own API keys to access third-party platforms (general machine translation or LLM large models).
- Local Translation: Use the bundled Hy-MT2 local translation model for fully offline translation, with no internet connection or API keys required. Suitable for privacy-sensitive or unstable network scenarios; the local model automatically unloads when idle to free memory.
The translation engine used by voice translation also supports Custom and Local modes (but not the built-in mode).
After selecting "Custom" mode, choose according to the type of service you want to use:
- General Translation Service
- LLM Large Model Service
1. General Machine Translation Configuration
Suitable for traditional machine translation APIs like Tencent Translator, Baidu Translate, Volcano Engine, Microsoft Translator, Alibaba Cloud Translation, etc.
| Configuration Item | Description |
|---|---|
| API ID | Application ID or App ID assigned by the service provider (some providers don't require this field, you can fill in any number like 12345 to skip). |
| API Key / Secret | Key string generated in the service provider's console for verifying your identity. |
2. LLM / Large Model API Configuration
Suitable for conversational large language model APIs like OpenAI, DeepSeek, Doubao, Tongyi Qianwen, Kimi, MiniMax, Zhipu, etc., which obtain results by sending translation instructions to the model.
Supported LLM providers (refer to the in-software list for real-time availability):
| Provider | Description |
|---|---|
| OpenAI | GPT series models |
| DeepSeek | DeepSeek large models |
| Alibaba Bailian (Tongyi Qianwen) | Alibaba Cloud Tongyi series models |
| Volcano Engine (Doubao) | ByteDance Doubao large models |
| SiliconFlow | SiliconFlow model platform |
| Anthropic | Claude series models |
| Google Gemini | Gemini series models |
| Tencent Hunyuan | Tencent Hunyuan large models |
| Kimi (Moonshot AI) | Kimi series models |
| MiniMax | MiniMax large models |
| Baidu Qianfan | Baidu ERNIE series models |
| xAI | Grok series models |
| Zhipu | GLM series models |
| Configuration Item | Description |
|---|---|
| API Key | Interface key provided by the service provider, usually starting with sk-. |
| Model Name | The specific model name to use, such as gpt-4o, deepseek-chat, glm-4. In built-in mode, the available model list is automatically fetched for selection. |
| Custom Translation Prompt | System instruction to control the model's translation behavior, can be used to specify translation style, tone, or add special requirements. The default prompt for text translation is: "You are a translation expert, the text to be translated is the result of OCR recognition..."; the default prompt for voice translation is: "You are a voice translation expert, the input is continuous speech-recognized segments...". |
| Custom Request Address (API URL) | If you use locally deployed Ollama, LM Studio, or third-party proxy services, you can fill in the custom interface address here (e.g., http://192.168.1.100:11434/v1/chat/completions). Leave blank to use the default address of the selected service provider. |
| Auth Username / Auth Password | Some providers' built-in or custom authentication methods require the account and password assigned by the provider; fill in as needed. |
How to Get API Keys?
- Tencent/Baidu/Microsoft/Alibaba Cloud: Register and create an application in the corresponding cloud service console to get the key.
- DeepSeek: Recommended, excellent semantic understanding effect, go to Platform Console to register and get.
- OpenAI: Go to Official Website to register and get.
- For the other providers, please apply on the developer / API console of their respective official websites.
Local Translation Mode
Local translation mode requires no API keys and runs completely offline:
- Uses the bundled Hy-MT2 (about 1.8B) local translation model, based on CPU inference.
- The number of inference cores can be adjusted in settings (range 1 to the device core count, default 4) to balance speed and resource usage.
- The local model automatically unloads by default after being idle for 30 seconds to free memory, and is loaded on demand for the next translation.
- Supports the same language range as speech recognition (including auto-detection).
Local translation mode is mainly enabled in the "Translation Engine settings" by switching the text translation or voice translation engine to "Local", while ensuring the local translation model files are properly installed.