Translation

💡TL;DR

Notemd translates the active file or eligible files in a folder using your configured LLM. By default, it writes a separate file with a language suffix. Language coverage and translation quality depend on the model; the 21 UI locales and 34 website locales are not limits or guarantees for translation.

Overview

Choose the output language and translation provider before running the task. Use a copied test note containing representative terminology, links and code. This feature uses your selected LLM, not a dedicated Google Translate integration.

The registered translation actions operate on files and folders. To translate only a passage, copy it into a temporary note first. Selecting text in the editor does not narrow the Translate current file action to that selection.

How It Works

The Translate Command

1. Choose the target output language in Notemd's language control; it defaults to English and is separate from the UI language.
2. Open the source note and run Translate current file.
3. Notemd chunks long content, sends it to the translation provider/model, and joins the returned chunks.
4. Open the reported output path and review meaning, terminology, citations, Markdown links and code before using the result.

With default settings, Example.md translated to French becomes Example_fr.md beside the source. The original file is retained. A matching existing output is overwritten; there is no default append-to-source mode or guaranteed Translation heading.

Language Pairs

The model infers the source language. Target-language availability in a menu does not establish accuracy for a language pair. Evaluate terminology and mixed-language passages in a sample. Long notes are chunked and may lose cross-chunk context.

Per-Task Model Selection

Enable task-specific model settings to use translateProvider and translateModel. Otherwise translation uses the active provider. A local model requires a running server and downloaded model; a cloud provider receives the translation prompt and content.

Avoid choosing a model solely from a generic speed or quality ranking. Use a small domain-specific sample and verify its output against the source.

Batch Folder Translation

Run Batch translate folder, select the intended folder and inspect the scope. The default selection includes direct child files only; advanced selection profiles can change that behavior. Each output uses its source basename and the chosen language suffix, so two same-named source files directed into one custom output folder can collide. Choose destinations deliberately and review existing translated files before rerunning.

Configuration

Setting	Default	Effect
language	en	Selected target output language
translateProvider	DeepSeek	Translation provider when task-specific routing is enabled
translateModel	''	Use that provider's model when empty
useCustomTranslationSuffix	false	Use the language-code suffix by default
translationCustomSuffix	_translated	Suffix used when the custom suffix is enabled
useCustomTranslationSavePath	false	Save beside the source by default
translationSavePath	''	Custom Vault-relative output folder
batchConcurrency	1	Requested folder translation concurrency

If a requested custom folder cannot be created, the runtime can fall back to the source folder and reports the actual output path. Inspect that path instead of assuming the requested destination was used. An empty custom suffix can target the source filename; use a distinct suffix for a separate translation.

Example

Set the output language to English, copy a short Chinese note into a test folder, and run the file translation action. Inspect the resulting _en.md file side by side with the source. Check that technical terms, numbers, equations and links remain correct before translating the full folder.

Cancellation And Review

Cancel stops further task work where supported; it does not undo already written translations or guarantee cancellation of provider billing. Review the task report and output files after cancellation. Model instructions to preserve Markdown are a request, not a parser-level guarantee of semantic equivalence.

Tips

• Keep a source backup and check for output-name collisions.
• Configure a distinct output suffix and folder when experimenting.
• Verify code identifiers, mathematical notation, source URLs and proper nouns manually.
• Review diagnostic output for private content before sharing it.

Next Steps

• Batch processing: scope, concurrency and partial completion.
• Custom prompts: translation prompt substitutions.
• Provider setup: endpoints, model discovery and connection tests.