Private neural translation
Translate without uploading your data.
Text and documents are parsed and translated in a dedicated Web Worker. Model weights are fetched once, cached locally, and application network access is blocked during processing.
Server/API
None
Content upload
Blocked
Primary engine
NLLB-200
Execution
WebGPU → WASM
Model profile
Model not loaded
The first load downloads model weights and stores them in browser IndexedDB.
Text workspace
Instant translation
0 words · ~0 tokens
Translated output
Your translation will appear here.
Document workspace
Translate files locally
Excel preservationOnly textual cells are changed. Formula cells, numeric values, sheet structure and workbook column metadata are left untouched.
Word preservationDOCX is unzipped in memory and text inside <w:t> elements is translated before the original package is rebuilt.
PDF behaviorPDF text is extracted page-by-page and exported as a clean UTF-8 translated text document; the original PDF is never modified.