A drag-and-drop front end for FFmpeg, Pandoc, and friends

What it does FileWizard is a self-hosted web app for three chores that normally mean three different tools: converting files, OCR-ing PDFs and images, and transcribing audio. It reimplements none of them — it orchestrates the standard CLI stack (FFmpeg, LibreOffice, Pandoc, Ghostscript, Calibre, ImageMagick, Tesseract, faster-whisper, plus a drawer of others) behind one dark drag-and-drop UI. Jobs run in the background with live status and persistent history; 180+ formats work out of the box, 700+ in theory.

The interesting bit This is glue code and refreshingly unashamed of it — the value is that someone else did the plumbing: a job queue, per-tool command templates, per-process CPU and memory limits, and secrets that never get sent to the browser. Adding a converter means writing a command template in settings.yml, not writing a converter.

Key highlights

• Conversion, OCR (many languages, selectable per job), and Whisper transcription with .txt, .srt, and .vtt output — all on your own hardware.
• Extendable to any CLI tool via settings.yml; command templates stay read-only in the browser unless you explicitly unlock edits.
• Background jobs with real-time status, persistent history, and automatic cleanup of finished jobs after a set number of days.
• Ships as Docker images for amd64 and arm64 (Raspberry Pi 4/5, Apple Silicon); CPU-only by default, with a cuda variant for GPU transcription.
• FastAPI backend, vanilla HTML/JS/CSS frontend — no framework payload.

Caveats

• In LOCAL_ONLY mode there is no login: anyone who can reach the port can use the app and change its settings. The README warns this in bold; believe it.
• Converters process untrusted files, and the README itself says to run the container with only the volumes it needs.
• Fidelity varies — complex layouts through LibreOffice and PDF input/output via Calibre are flagged as imperfect in the project’s own tools table.

Verdict If you run a home server or NAS and regularly convert, OCR, or transcribe files, this is a genuinely handy appliance — arm64 support means it’ll even run on the Raspberry Pi in your drawer. If you’re exposing it beyond your own machine, note that local mode ships without a door lock; OIDC is the lock.