Connecting Your Own Translation Stack
Connecting Your Own Translation Stack
Existing translation capacity can be plugged into Transbox through three independent paths, which can be combined.
Path 1: Provider API Keys
For Baidu / Microsoft / Volcengine / NiuTrans / Alibaba accounts you already own.
- Location: Settings → Translation engines → Machine translation APIs
- Each provider has its own switch; fill the credentials and save
- Requests go from this device straight to the provider and are billed to your account
Path 2: Custom LLM
For self-hosted or purchased models that speak the OpenAI protocol.
- Location: Settings → Translation engines → Large language models → Add model
- Fill in model ID, Base URL and API key; optionally a display name and custom headers
- Once enabled it appears under Custom models in the engine dropdown, and is available in Chatbox and the in-editor AI assistant
- LLM translation automatically carries over terms matched from your local termbases, keeping terminology consistent
- This is a plan entitlement
Path 3: Self-deployed Server
For private deployments where accounts, billing and language resources are managed by your own server.
- Location: Settings → Network & advanced → Online service → Service address
- Empty means the built-in service; a full HTTP/HTTPS address switches the client to your server
- Restart the client after changing it so every service connection points to the new address
This path replaces the server; it does not add an engine. See Private deployment.
Choosing a Path
| Situation | Path |
|---|---|
| You just want to stop spending platform credits | Path 1 |
| You want a stronger model and will pay token costs | Path 2 |
| Data must stay inside the intranet | Path 3 + Path 2 (internal model) |
FAQ
Q: Which settings does a custom model take?
A: Model ID, Base URL, API key, display name and optional custom headers.
Q: Where can a custom model be used?
A: Text translation, the mini window, project pre-translation and segment MT, Agent translation, Chatbox and the in-editor AI assistant.
Q: Do platform models still work after adding a custom one?
A: Yes. Both coexist; pick whichever you need.
Q: After self-deploying a server, do I still need MT keys?
A: Depends on whether your server has translation capacity built in. If you prefer direct-to-provider calls, configure Path 1 as well.