Source
0 characters · 0 segments
no pack loaded
Translation
—
Memory hits
Nothing reused yet.
Glossary applied
No terms locked.
On-device cost
No run yet.
Two people, one device. Each side types in their own language; both directions run from the packs you installed.
Batch translate
Drop a .txt, .md, .srt or .csv file. Each line and subtitle block is translated in order, then exported as one file.
No file loaded.
Preview
Nothing yet.
Quality by line
A batch is only useful if you know which lines it damaged. Every line the engine flagged as empty, untranslated, repetitive or truncated, and every line it had to re-decode, is named here with its number.
Translate a file to see which lines need a look.
Glossary
Locked terms are protected before the model sees the text and restored afterwards — names, brands, legal wording, technical vocabulary.
| Source | Preferred translation |
|---|
Translation memory
Every segment you approve is stored locally in this browser. Identical sentences are reused instantly, with zero compute.
Memory is empty.
Install a language pack
A pack is a folder of model weights that lives in this browser (Cache Storage) or next to the app in models/. Import a .zip or a folder — nothing is uploaded anywhere.
Offline is the default. Press Get on a pair below to download it once; after that the model lives on this device and translation makes no network calls. If you only want to try it, switch on Online mode in the toolbar — the model streams while you work and nothing is installed.
Measuring…
Suggested packs
One click downloads the model once, from the publisher's own pinned revision, and keeps it on this device. After that the pack works offline. Sizes are the real quantised (q8) weights.
Installed packs
No packs installed yet.
This device
WASM threads—
EngineTransformers.js · ONNX Runtime Web
Network calls for translation0
Accounts requirednone
Storage used—
Offline ready—