AI Translation Overview
Supervertaler Workbench can use large language models (LLMs) to translate, review, and rephrase segments. It connects to cloud providers or local models – you choose what works for you.
Supported providers
| Provider | Type | Cost |
|---|---|---|
| OpenAI (GPT-4o, GPT-4o mini) | Cloud | Pay-per-use |
| Anthropic (Claude) | Cloud | Pay-per-use |
| DeepL | Cloud | Free tier available |
| Ollama (Llama, Mistral, etc.) | Local | Free |
How it works
- Select a segment (or multiple segments) in the translation grid
- Click Translate or press
Ctrl+T - Supervertaler Workbench sends the source text, along with context (surrounding segments, TM matches, terminology), to the AI provider
- The translation appears in the target column
Prompts
AI translation uses configurable prompts that tell the model how to translate. Supervertaler Workbench includes built-in prompts for common tasks, and you can write your own.
Prompts use the .svprompt format – a simple text file with YAML metadata and a Markdown body. This format is shared with Supervertaler for Trados, so prompts work in both tools.
Privacy
- API keys are stored locally on your machine
- Source text is sent to the provider’s API only when you trigger a translation
- When using Ollama, all processing happens on your machine – nothing is sent to the internet