Turn reference translations into a memory. You no longer need a separate tool.
Point it at your source files and their translations. It pairs them, aligns them a batch at a time, and writes a TMX you can import into any memory — one file per pair, or pooled into one.
Point it at a folder of source files and their translations, and it produces TMX — one file per pair, or pooled into one. Files whose names agree once the extension and the language code are taken off are paired for you; you pair the rest by hand, and see every pair before anything runs.
Have an AI model you choose align the text. Let the documents' own structure do it, for web pages that share a layout. Or send the pairs, as they are, to an alignment service you run yourself. The choice is per run, and the wizard remembers it.
It runs the batch with progress, a pair at a time, survives being cancelled, and picks up where it stopped. Close the program in the middle and nothing already aligned has to be paid for twice.
Alignment is not a project's work, so a run makes its own confidentiality choice. Whatever it sends to an AI model carries stand-ins in both languages, read back before it goes and refused if a protected name is still in it — and each half is restored in its own language before the TMX is writtenMeasured on two protected document pairs: none of the protected text reached the AI model in either request, and 27 of 27 aligned pairs were restored exactly, on both sides..
When the run is done, View Results opens what it wrote, pair by pair, without changing it.
There is no approve-each-pair step in the alignment run: what the aligner produces is what gets written. Read the result before you import it into a memory you rely on. Use the TM editor if any corrections need to be made.
