Translator
Translate with the model you pick, and compare models on the same text.
Translator works with your own engine: a local model, the configured cloud model, or the app's routing. With a local model, the text stays on your computer.
The cartridge id is @mnemosyne-plugins/translator. Its code lives at
Mnemosyne-OS/mnemosyne_OS-translator.
This page describes Translator 2.0, merged after Mnemosyne OS 2.2.0, which shipped with Translator 1.3.2. Five parts of 2.0 need a newer Mnemosyne OS: the model loading dialog, Free memory, Web page, Discuss in chat and the far view. On an older host those buttons say This version of Mnemosyne does not have this feature yet. Update the app. Translating text and documents works as before.
The tabs
Translate, Documents, Web page, Compare, Vocabulary and History, under one bar with your languages, the engine and Free memory.
Your languages
In Settings, list My languages and set My usual pair, for example English and Spanish. The target follows what you type: English goes into Spanish, Spanish into English. Any language can still be picked as a target.
Pick the engine
Which engine translates offers Auto (the app's routing), Cloud and each installed local model. A local model loads through the same dialog as the chat, with its memory check. Picking a model here never changes the chat's model.
Free memory unloads the idle engines (speech recognition, voice, memory embedder) to make room. Free memory before loading a local model is a setting.
Translate
Type or paste a text, then Translate, or press Ctrl+Enter. Open a text
file reads a .md or .txt. A long text is translated piece by piece, with
a progress line. Stop keeps what was translated so far.
The text goes to the model with an instruction to translate it, never to answer it. Your memory is not searched for these calls, so nothing from your vaults can slip into a translation.
Documents
Add a file (PDF, DOCX, EPUB, TXT or MD), then Translate all. Tick OCR to read scanned pages. Each file shows its state: Waiting, Reading, Translating, Done.
Web page
Type an https:// address and press Read. The page is read on your
computer. Translate the page, then:
- Report lays it out with its address and the original text.
- Discuss in chat hands it to the chat.
- Keep saves it in the Translator's vault.
- Export .md saves a file.
Compare
Give one text and 2 to 6 engines, then Run. The engines run one at a time: free memory, load, translate, unload, next. Leave the chat idle during a run.
The result shows load time, translation time, characters per second, length, Agreement between translations and word-level Differences. Agreement is shared words, not quality: two engines can agree on a mistake. Save the run as a scenario and run it again later. Report .md exports it.
Vocabulary (beta)
After a translation, press Words to collect the words worth learning. It is a second model pass, so it only runs on a click. Review shows each card with Knew it or Didn't know. Spaced repetition brings a card back before you forget it.
Where your data goes
- The history and the vocabulary live in the window's storage. They are also copied, when possible, into the Translator's own vault.
- That vault is walled off from the rest of your memory. You make it permanent from the vault manager.
- Delete all in History clears the history.
- A cloud model sends the text to its provider. A local model keeps it on your computer.