Import and export knowledge sources
Open Workspace → Settings → Knowledge sources, expand a resource and choose Import / export resource. Owners, Admins and assigned resource editors can use this workflow. Ordinary members with a project link can read suggestions, but cannot import or export the whole source. Resource editing does not grant translation access.
Glossary CSV
Section titled “Glossary CSV”- Select a UTF-8 CSV file with a unique header row.
- Choose Raw source values for ordinary CSV, or Localisync spreadsheet-safe export for a file exported by this workflow.
- Map the concept column and optional description column by their exact header names.
- Add a language mapping for each language to import. Specify a language code and
preferred-term column. Optional columns hold a JSON array of alternatives and
literal
true/falsecase-sensitive and do-not-translate flags. - Choose Preview import. Unmapped columns are disclosed as a warning.
The export header is concept,description, followed by each language code and its
.alternatives, .caseSensitive, .doNotTranslate columns. Empty language terms
are omitted only when the related metadata is also empty. A nonempty concept and
at least one nonempty term are required. Different language variants are not silently
collapsed into a more general language.
CSV exports quote formula-sensitive cells with a reversible leading apostrophe, including existing leading apostrophes. Select the spreadsheet-safe profile when importing them again. Raw input is never silently unescaped. CSV is data; formulas are not executed or evaluated by Localisync.
Translation memory TMX
Section titled “Translation memory TMX”Select a TMX 1.4b file and explicit source and target language codes. The selected pair is used for every unit; units without both languages are counted in the preview warning. Language variants must be explicit and unique within each unit.
Segments must contain plain text. XML document types, external entities and inline
codes are rejected; markup is never flattened into a different translation. Unknown
notes/properties are disclosed as non-imported metadata. A unit can carry an explicit
ICU contract in one prop named x-localisync-icu-contract, containing JSON with
format: "icu" and a parameters array of { name, type }, where type is text or
number. Without this property the segment is Plain text. The same ICU validation
used for managed entries checks both texts and the declared contract.
TMX export uses version="1.4", as specified by the
TMX 1.4b specification, and writes
the explicit contract for ICU entries. Exported review-derived pairs remain data;
importing them creates managed entries, never human review evidence.
Review and apply
Section titled “Review and apply”The preview freezes the resource revision, incoming rows and matching current entries. It is paginated; choices on earlier pages remain selected. Every row starts as Skip. Inspect its text and existing matches, then choose Create entry, Create a separate entry, or a specific Replace managed match. Review-derived evidence is immutable and cannot be a replacement target.
Repeated glossary concepts or memory source/language/contract identities form duplicate groups. Choose one row per group and skip the others. The preview shows the total duplicate count and up to 20 matching row numbers. Explicitly confirm the mappings, warnings and selection, then Apply selected rows. Unselected rows are skipped. All creates and replacements commit together with an audit result, or none do.
A concurrent entry, resource or supporting-review change causes a conflict. The preview and choices remain visible; create a fresh preview after comparing the current source. A replay of the same Apply request returns its recorded result without adding entries. No translation, approval, release or publication is created by this operation.
Limits and recovery
Section titled “Limits and recovery”- Input: UTF-8, at most 1 MiB and 10,000 entries.
- Whole-resource file operations: at most 10,000 stored entries and 20 MiB of content.
- Preview: at most 20 MiB, retained for 24 hours, with at most 100 existing matches per identity.
- Parsing: one isolated process per API instance at a time, a 128 MiB JavaScript heap, a monitored 256 MiB resident-memory budget and a 30-second deadline. Busy processing returns an explicit retryable conflict.
Split oversized files or resolve excessive existing duplicates. Invalid XML/CSV, language mapping or ICU contracts stop preview creation. No partially parsed rows are saved. Expiry and current resource permissions are checked before Apply; permission loss cannot be bypassed by keeping an old preview open. Keep the content worker running to expire preview records. Restoring a backup does not automatically apply any preview.
Choose whether we remember appearance preferences and use Google Analytics on Localisync and its documentation. Optional storage stays off until you accept.
Read our cookie policy