A local LLM translator that handles PDF tables and formulas

What it does

DocuTranslate is a Python tool that shovels documents through LLMs for translation. It supports PDF, Word, Excel, EPUB, JSON, subtitle files, and more, then attempts to preserve the original structure instead of giving you back a wall of plain text. It runs locally with a built-in web UI and REST API, and can be shared across a LAN.

The interesting bit

Instead of treating every file as plain text, it uses the MinerU parser to pull out tables, formulas, and code blocks from academic PDFs before translation, and it can auto-generate a glossary to keep terms consistent across chunks. It also doubles as an MCP server, so your AI assistant can invoke it directly.

Key highlights

• Translates PDFs, DOCX, XLSX, EPUB, JSON, SRT, ASS, Markdown, and plain text
• Preserves Word and Excel formatting; supports JSONPath to target specific values
• Uses MinerU (online or self-hosted) for PDF layout analysis including tables and formulas
• Offers async concurrency controls, custom prompts, and rate-limit handling
• Ships as a sub-40MB portable app for Windows and Mac, or via Docker/PyPI

Caveats

• PDFs are converted to Markdown first, so original page layout is lost
• Legacy .doc and .xls formats are not supported, only modern .docx and .xlsx

Verdict

Grab it if you regularly translate long documents or academic papers and need more structure than a raw text paste. Skip it if you require pixel-perfect PDF layout reproduction or only work in legacy Office formats.