The main speech engine is English. Due to the large amount of open data available for English model training, even if you have a strong accent, you'll probably get better command accuracy by tweaking Englishy commands rather than trying to use a lower accuracy engine for your native language. (By Englishy I do mean you can say a non-existent or non-English word, ask the engine what it heard, and map that semi-phonetic result to a command.)
It's extremely customizable, there's plumbing in the open source config repo for programming language specific commands and adding support for new text editors and such. There is a dictation mode. The current public English engine isn't the best for dictation yet, but it supports switching engine between command/dictation mode.
There's an online WebSpeech dictation-only engine in the beta which has extremely good dictation accuracy for most languages, as well as vosk in the beta for offline dictation (vosk has e.g. a decent German model).
The Stenomask is one option for using voice control in an office environment where you don't want to make noise.