Engine choice

Ponyglot translates with DeepL and with a large language model (LLM). Each has strengths: DeepL is fast, consistent and supports glossaries and formality for many languages; an LLM covers more languages and follows instructions such as tone and style.

Only engines you enable

Content is only sent to the translation engines you enable. Engines are opt-in per site. If a site has only DeepL enabled, no content of that site ever reaches an LLM, and the other way round. Enabling the LLM includes choosing its provider, for example Mistral AI (France): content goes to that provider and no other, and the hosted service only offers providers that process in the EU under a data processing agreement. With no engine enabled, Ponyglot only uses the translation memory.

DeepL where supported, otherwise the LLM

Each target language of a site has an engine setting. The default, Automatic, uses DeepL where it supports the language pair and the LLM otherwise, always among the enabled engines. You can also fix a language to one engine, for example to use the LLM for German because it follows your brand voice’s tone instructions.

What each engine receives

	DeepL	LLM
Source text	Yes	Yes
Page title as context	Yes	Yes
Glossary	As a DeepL glossary, where supported	Terms that occur in the text
Formality	Yes, where supported	Yes
Tone instructions, forbidden words	No	Yes
Fuzzy matches from the translation memory	No	Yes, as references

Text that the translation memory covers isn’t sent to any engine.

When no engine can translate

If no enabled engine supports a language pair, the job reports it in errors and those segments stay missing. Enable another engine or change the language’s engine setting, then request a new job.