Mark your proper nouns, products, SKUs, and acronyms as «do-not-translate»: the app masks them before translation and then programmatically verifies they are still there, identical.
Asking a translator to respect your brand is not an instruction; it’s a verification. Here, the term is masked beforehand and counted afterwards.
The brand glossary is a term list that must appear in the translation exactly as in the original: your company name, your products, your technologies, your catalog references, and your acronyms.
You fill it out once, and it applies to all translations in that run, across all target languages.
What distinguishes it from writing «do not translate my brand» in a prompt is the method. Before sending anything to the translator, each occurrence of the term is replaced in the text by an untouchable internal marker, so the model never sees the word it might translate. Additionally, the term is declared in the instructions as an untouchable name. When the translation returns, a verifier counts the markers: if there are fewer than what went out, the text is not accepted.
When something is missing, the app doesn't shrug its shoulders. First, it tries to deterministically replace the missing piece, without spending another call; if the problem persists, it launches a targeted retry that tells the translator exactly what has been left out, up to two times. And when the masking is undone, the term returns written exactly as you registered it.
The quality report for each language includes a line for every term from your glossary that appeared in the original, with its actual status in that translation: «Intact (2 times)» or «We expected 2 and found 1». What could not be guaranteed is declared in the interventions section, for you to review before sending. By default, matching is case-sensitive, so registering «Apple» or «Pulse» doesn't block common words spelled the same way; you can disable it term by term, and activate inflected forms matching when you want to cover plurals and declensions.
Use it free without an account. If you want to save your result, you can sign up afterwards.
Save / Sign upMasking before translating, not pleading afterwards
the term is masked in the text before going to the model. What the translator doesn't see, they can't translate.
Code-based verification, not trust
upon return, we count how many times the term appeared originally and how many times it appears now. If the numbers don't match, there's deterministic repair and up to two targeted retries.
Case-sensitive by default
«Apple», «Pulse», or «Air» are protected as brands without blocking common words spelled the same. You can deactivate this feature term by term when needed.
Optional plurals and declensions
activate inflected form matching, and the term is also covered when the target language declines or pluralizes it.
A report that tells you the truth term by term
each translation includes the real status of every term in your glossary —intact or lost, with the count— and declares what could not be guaranteed.
Three modes in the same list
«do-not-translate», «always translate as» with the equivalent you define per language, and «usage note» to explain in your own words how a term should be used.
A glossary that goes beyond a single translation run
with a free account, you save up to 20 glossaries of 400 terms, import and export them in semicolon-separated CSV, with preview and duplicate warnings before adding anything.
Try it without an account
in guest mode, you work with this session's glossary, which applies equally to the translation; creating an account only ensures it isn't lost when you close.
Mark your terms, translate, and view the report count. The first three translations don't require an account.
In the features that use AI, responses are generated by artificial intelligence and are not a substitute for professional advice.