MDFlux - a desktop GUI I built on top of MarkItDown

[ibrahimqureshae]

Hi folks. Sharing a project I built on top of MarkItDown, in case it's useful.

The conversion engine is MarkItDown under the hood, so on clean files the output is basically identical. What I added is everything around it that the library didn't give me.

The main pain points I hit using MarkItDown directly:

• No GUI. It's a CLI and a library, which locks out anyone who won't pip install and write a script. MDFlux is a portable desktop app, unzip and run, no terminal. First launch sets up its own private Python env, everything after runs offline.
• Scanned PDFs return ~0 characters. No text layer to extract, so MarkItDown hands you nothing. MDFlux runs OCR locally (RapidOCR + pypdfium2) and recovers the text as Markdown.
• No batch. Converting a folder means writing your own script, and that script has no answer for the file that hangs silently on item 47 of 200. MDFlux batches a folder with progress, cancellation, and per-file diagnostics.
• Silent failures. A MarkItDown call can hang for minutes with no signal, or return an empty string with no explanation. Every long op in MDFlux streams progress, is cancellable, and warns on empty output with a diagnostics panel that gives a real next step.

Plus a few smaller things: cleanup modes (Off, rule-based, or AI via local Ollama or your own API key), local audio transcription (faster-whisper), self-provisioning of all dependencies on first launch (hash-pinned, no system binaries), and local-first by default (no cloud, no API key).

Repo + download (Windows x64 portable zip): https://github.com/ibrahimqureshae/mdflux

MIT, contributions welcome. If any of the wrapper pieces (OCR wiring, batch concurrency, self-provisioning, the Tauri + sidecar architecture) would be useful to fold back upstream or worth a write-up, happy to discuss. Thanks for the great work of contributors here!