Your documents don't need to become JPEGs to talk to an LLM

What it does MDFlux is a desktop wrapper around Microsoft’s MarkItDown that turns PDFs, Office documents, EPUBs, images, and even audio into structured Markdown. It runs locally, bundles its own Python runtime in the Full edition, and handles OCR for scanned pages that plain extractors treat as blank. It accepts single files or entire folders, offers an optional rule-based or local-AI cleanup pass, and returns diff-able text with headings and tables intact.

The interesting bit The README is admirably upfront: the core engine is Microsoft’s library; MDFlux’s value is the packaging—OCR, batching, a Changes tab, and a Tauri/Svelte desktop shell that spares you the command line. The token math is the real pitch: ordinary documents run about four times lighter than sending pages as images to a vision model, and the authors clock a scanned PDF at 1,893 tokens versus 10,731 for the vision route—roughly a 5.7× saving.

Key highlights

• Converts a wide format list—PDF, DOCX, PPTX, XLSX, EPUB, HTML, CSV, JSON, XML, TXT, Markdown, images, and audio—into clean Markdown.
• Built-in OCR recovers text from image-only PDFs without touching a cloud API.
• The Full edition is fully offline-ready and bundles an immutable runtime; no account or API key is required for core conversion.
• Optional cleanup modes: off, rule-based, or AI (local or remote), with a Changes tab that diffs what was altered.
• Batch processing with per-file diagnostics and cancellation.

Caveats

• Windows and Linux x64 only; macOS is not currently supported.
• The Lite edition is smaller but provisions dependencies on first launch, so it needs an internet connection for that initial run.
• AI cleanup can route to remote providers (OpenAI, Groq, DeepSeek, etc.), which sends text off-device if you choose that mode.

Verdict Worth a look if you’re building RAG pipelines, archiving research, or just tired of paying vision-model rates to read your own files. If you already live happily in the terminal with MarkItDown and don’t need OCR or batching, the wrapper adds little.