Import & exportJuly 31, 2026

String files, and a rebuilt glossary library

From the Transept team

The strings review grid open on an email campaign named spring-sale, English into German, reading “4 of 8 translated”. A three-column table lists the message keys (subject_line, preheader, hero_headline, cta_button, shipping_note, footer_unsub), the English source beside each one, and a Translation column where four rows carry German copy and the rest read “Untranslated”. Character counters sit under the translation cells — 46/90 on the preheader, 0/45 on the untranslated subject line — and the source column keeps its Liquid personalization tags, a Liquid conditional, and a %d placeholder intact. — The strings grid — one row per key, character limits where the file declared them, and the campaign’s tokens left untouched.

Until now Transept translated documents. This release adds the other shape localization work arrives in: string files — the keyed tables that software UIs and email platforms export. They import as normal Transept documents, so glossaries, style guides, translation memory, workflows and review all work on them unchanged.

The second half of the release rebuilds how glossaries and style guides are managed: dedicated library pages, one glossary covering several target languages, and a review step so nothing generated by a model lands on its own.

Added

String files import. Upload a CSV, a gettext `.po`, or an XLIFF 1.2 file and it becomes a Transept document with one entry per key, keeping context comments, character limits and plural slots. Translations already present in the file seed the document at no cost, and go into your translation memory. See Translate software strings and email templates.
A review grid for strings. String documents open on a key / source / translation table with a filter, character-limit meters on the rows that declare one, and inline editing — with a jump into the full editor on any row.
Two ways out. Export a keyed file for your developers, or export the original file with the translations written back into it. A file you have not translated exports byte-identical to what you uploaded.
Placeholder protection. Liquid and printf tokens ({{first_name}}, %s, {% if %}) are masked before a model sees the text and verified afterwards, so they come back intact and in place. Detected automatically on import.
Translation memory export. The memory panel gained an Export button next to Import — it downloads everything you can see as TMX. What you bring in, you can take out.
Multiple languages per glossary and style guide. A glossary now holds a column per target language, with an auto-build that can fill in a language you added later; style guides carry per-language rules over a shared voice. See Glossaries and Style guides.
Library pages for glossaries and style guides. Both now have their own pages with scope tabs, search, filters and a manage menu (duplicate, move to a team, delete), plus a Merge action that folds one resource into another with a preview of exactly what would change.
A Search tab in the editor. Search a document by word or by glossary term, scoped to source or translation, and jump straight to the block.

Changed

Nothing generated lands silently. A style guide or glossary built by AI now parks as a result you review first — editable, shown as a diff when it updates an existing resource, with a count of exactly what Apply will write. Targeted resources carry a Review pending badge until you deal with it.
Generation streams as it runs. Glossary and style-guide auto-builds now show the model working, wave by wave, and can be cancelled mid-run — a cancelled run costs you nothing beyond what it had already read.
Auto-build proposes constraints. Extraction can now suggest do not translate, spelling variants, and forbidden targets alongside a term, sparingly and removable with one click. It also stopped proposing ordinary word forms as variants.
One context field on glossary terms. The separate Notes field is retired; existing notes moved into the context hint, which is the field the translation prompt actually reads.
An all-languages run quotes itself first. Picking "All languages" now asks for an estimate before anything is dispatched, so a warning like "your glossary has no column for three of these languages" arrives while it is still actionable.

Fixed

Email platform exports import correctly. Klaviyo files were rejected outright over their key column, regional columns (es-MX, pt_BR, en-GB) were dropped without a word, and Braze {% translation %} defaults were translated in place — which left every locale slot empty and every recipient reading the wrong language. All four are fixed, and an unrecognised column is now reported instead of discarded.
Braze personalization tags survive a run. Braze writes tags as {{${first_name}}}, and our parser stopped a brace early — leaving a stray character in the translatable text that models moved, duplicated or dropped. Found while writing the email localization guide.
Style-guide generation works in production. Auto-generation from a page, a prompt, or a sample was failing on a misconfigured key; all three now run on the same engine as the rest, and the output no longer duplicates what belongs in a glossary.
Language versions export with their tables intact. A Markdown or HTML language version exported with tables and code blocks flattened into paragraphs while the original exported perfectly. Language versions now inherit the source file they were copied from.
The site header fits a laptop again. The marketing nav collapsed to a hamburger below 1280px, so 13" screens and non-maximised windows got mobile chrome; it now measures itself and collapses only when it genuinely would not fit. The editor header no longer overlaps its language chips at narrow widths.

Have a request or hit a snag? Ask Literess in the app, or write to [email protected].