Translate Markdown: AI Translation That Keeps Formatting
Translate Markdown into 25 languages while keeping every heading, list, link, code block, and table in place, and get a warning if one moved.
Markdown-aware translation
Most translators destroy Markdown: code blocks get “translated,” link syntax gets broken, headings lose their pound signs. Our pipeline sends the document to a Gemini model with a hardened 10-rule prompt that explicitly protects every Markdown construct (fenced code blocks, inline code, links, images, math, frontmatter, tables) and then runs a structural validation pass over the output to flag any divergence (lost URL, missing code fence, heading-count drift) as a visible warning.
It uses modern LLMs for the actual translation, which means context-aware results: idioms become idioms, technical terms stay technical, and the tone matches the original. Code, URLs, inline HTML and frontmatter values other than the title, description, summary and subtitle are meant to pass through unchanged, and the check after translation warns you when they did not.
What we do
25 supported languages including French, Spanish, German, Mandarin, Japanese, Arabic, Hindi, Russian, Korean
Code blocks, math (`$...$`, `$$...$$`), URLs, and inline HTML protected by the prompt, which forbids translating or changing them
Frontmatter YAML keys preserved; only `title`, `description`, `summary`, `subtitle` values get translated
Structural validation: every output is checked for code-fence count, URL set, and heading count mismatches against the source
Target-language preference remembered between sessions
Use it to localize documentation, translate AI-generated content, ship multi-language READMEs, or prepare technical content for non-English audiences.
How it works
Paste your Markdown into the editor, or drop a .md or .txt file
Pick a target language from the dropdown (your last choice is remembered)
We send it to Gemini Flash Lite with a 10-rule preservation prompt
A structural validation pass checks code-fence, URL, and heading counts
Warnings (if any) appear in a yellow banner above the output
Use cases
Localize technical documentation for a global audience
Translate blog posts while keeping their Markdown formatting
Localize a software product’s README into 10+ languages
Translate AI-generated text into the local language
Ship multi-language release notes from a single source
Frequently asked questions
Will my code blocks be translated?
No. The preservation prompt explicitly forbids translating content inside fenced code blocks or inline code, and a post-translation validation pass flags any code-fence count mismatch as a visible warning.
How is quality compared to DeepL or Google Translate?
We make no quality claim against them. What differs is the handling of Markdown: the model is instructed to leave code, links, math and frontmatter keys untouched, and every result is checked for lost code fences, URLs, headings and images, with a warning when something moved.
Are URLs translated?
No. Link targets are kept as-is, and we verify it: every URL in the source must also be present in the output, or a warning is raised.
How big a document can I translate?
It depends on your plan. Free: up to 10 000 characters per request. Starter: 25 000. Pro: 50 000. Business: 100 000. Larger documents are truncated with a visible warning telling you how many characters got dropped; split a longer document and translate it in parts.
Which languages are supported?
25 major languages: English, French, Spanish, German, Portuguese, Italian, Dutch, Polish, Czech, Romanian, Russian, Ukrainian, Mandarin, Japanese, Korean, Hindi, Arabic, Turkish, Vietnamese, Thai, Indonesian, plus the Nordic languages (Swedish, Danish, Norwegian, Finnish).
How is it priced?
1 credit per 10 000 characters of input, minimum 1 credit. A 5 000-char README costs 1 credit; a 35 000-char doc costs 4 credits. The same source text + same target language hits a content cache and is re-translated for free, which is useful when iterating on a doc you keep tweaking.
What if my document is longer than the plan cap?
Documents larger than the cap are truncated with a visible warning telling you how many characters were dropped. Within the cap, documents over ~8 000 characters are automatically chunked at safe boundaries (paragraph breaks, never inside a fenced code block or table) and translated piece-by-piece in parts of about 7 000 characters, then re-assembled, so a 50 000-character Pro request is handled as about eight Gemini calls under the hood.
Is there a glossary or bilingual mode?
No. There is no glossary and no side-by-side output. Brand names and well-known technical terms are kept in their original form when the target language usually borrows them.