| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| LingoGeekSetup.exe | 2026-09-02 | 38.2 MB | |
| LingoGeek 1.0.1 source code.tar.gz | 2026-09-02 | 192.2 kB | |
| LingoGeek 1.0.1 source code.zip | 2026-09-02 | 200.7 kB | |
| README.md | 2026-09-02 | 968 Bytes | |
| Totals: 4 Items | 38.6 MB | 11 | |
Two fixes found by running the app against real documents.
Markdown structure is now protected from the translator. Heading hashes, list bullets, blockquote markers, horizontal rules, fenced code blocks and link targets are held back and put on again afterwards. Previously a heading could come back as a plain paragraph, because the model translated the hashes away. Keeping the syntax out of the way also improves the translation itself, since the model is no longer reading markup as part of the sentence.
Soft wrapped paragraphs are also joined before translating, rather than a line at a time.
The chosen languages no longer reset. The language list is reloaded when the queue empties, to pick up a pack that has just downloaded, and that was putting the source and target back to their defaults. Whatever you picked now survives.
Also: a job that has started but not yet counted its blocks says "translating" rather than "translating 0 of 0".
