Engines and site settings

Engines

Engine	engine value	Notes
DeepL	deepl	Formality and glossaries where DeepL supports them. HTML is translated with tag handling.
LLM	llm	A large language model from the site’s LLM provider (for example Mistral AI, France). Receives glossary terms, brand voice and fuzzy matches from the translation memory as instructions.

Enabling the LLM means choosing its LLM provider in the site settings: content goes to that provider only. The hosted service offers only providers that process in the EU under a data processing agreement.

Content is only sent to the engines you enable for the site. With no engine enabled, no content leaves Ponyglot, and translation jobs only use the translation memory.

Engine per target language

Each target language of a site has an engine setting:

Setting	Behaviour
Automatic (auto)	DeepL where it supports the language pair, otherwise the LLM, among the enabled engines.
DeepL (deepl)	Always DeepL. The engine must be enabled for the site.
LLM (llm)	Always the LLM. The engine must be enabled for the site.

If no enabled engine can translate a language, the job reports it in errors and those segments stay missing. Translation memory hits still arrive.

Language codes: en is translated as British English and pt as European Portuguese by DeepL. Use en-us or pt-br as target languages for the other variants.

Site settings

Setting	Meaning
Name, base URL	For the dashboard.
Source language	The language of the content the connector pushes. It can’t also be a target language.
Adapters	Which kinds of content the site uses; for information.
Engines	Engines content may be sent to.
Target languages	Each with formality (Default, Formal, Informal), engine and Writable.
API keys	One or more per site; shown once, stored hashed, revocable.

A target language is writable if Writable is checked and the connector reported the language in its handshake. Jobs only translate writable languages. Before the first handshake, the setting alone decides.