any2md — Convert Anything to Markdown

A toolkit for converting any media, document, or data format to markdown. AI inference runs locally on Apple Silicon via MLX — no cloud APIs.

Install

# From source
git clone https://github.com/roboalchemist/any2md.git && cd any2md
uv pip install -e '.[all]'

# Or install only what you need
uv pip install -e '.[stt]'    # Audio/video transcription
uv pip install -e '.[pdf]'    # PDF extraction
uv pip install -e '.[img]'    # Image OCR via VLM
uv pip install -e '.[web]'    # Web page conversion

# System dependency for audio/video
brew install ffmpeg

Usage

Just pass a file — any2md auto-detects the format:

any2md lecture.mp4                  # audio/video → markdown
any2md podcast.mp3 --diarize        # with speaker diarization
any2md document.pdf                 # PDF → markdown
any2md screenshot.png               # image → markdown (VLM OCR)
any2md https://example.com          # web page → markdown
any2md page.html                    # local HTML → markdown
any2md report.docx                  # office doc → markdown
any2md data.csv                     # CSV/TSV → markdown table
any2md config.json                  # JSON/YAML → markdown
any2md app.db                       # SQLite → schema + data
any2md captions.srt                 # subtitles → markdown
any2md notebook.ipynb               # Jupyter → markdown
any2md message.eml                  # email → markdown
any2md notes.org                    # Org-mode → markdown
any2md paper.tex                    # LaTeX → markdown
any2md command.1                    # man page → markdown
any2md readme.rst                   # RST → markdown

Use explicit subcommands for full control:

any2md audio podcast.mp3 --diarize --model parakeet-1.1b
any2md video lecture.mp4 -f srt -o ~/subtitles
any2md yt "https://youtube.com/watch?v=..." --diarize
any2md pdf document.pdf --pages 1-10 --ocr
any2md csv data.tsv --max-rows 50 --max-col-width 40
any2md db app.sqlite --max-rows 20 --skip-views
any2md sub captions.vtt -f txt
any2md nb notebook.ipynb --no-outputs

Supported Formats

Subcommand	Input	Engine	Dependencies
audio	MP3, WAV, FLAC, OGG, AAC, M4A	Parakeet STT + Sortformer diarization	mlx-audio
video	MP4, MKV, AVI, MOV, WebM	Parakeet STT + Sortformer diarization	mlx-audio
yt	YouTube URLs + all audio/video	Parakeet STT + yt-dlp download	mlx-audio, yt-dlp
pdf	PDF files	pymupdf4llm + optional Qwen3.5 VLM OCR	pymupdf4llm
img	JPEG, PNG, GIF, BMP, WebP, TIFF	Qwen3.5 (mlx-vlm)	mlx-vlm
web	Web URLs	ReaderLM-v2 (mlx-lm)	mlx-lm
html	Local HTML files	ReaderLM-v2 (mlx-lm)	mlx-lm
doc	DOCX, PPTX, XLSX, EPUB, ODT, RTF	markitdown	markitdown
rst	reStructuredText	pypandoc / docutils	—
csv	CSV, TSV	stdlib csv	none
data	JSON, YAML, JSONL	stdlib json, optional PyYAML	none
db	SQLite databases	stdlib sqlite3	none
sub	SRT, VTT, ASS/SSA subtitles	pysubs2	pysubs2
nb	Jupyter notebooks	stdlib json	none
eml	Email .eml, .mbox	stdlib email/mailbox	none
org	Emacs Org-mode	pure regex	none
tex	LaTeX	pure regex	none
man	Unix man pages	mandoc + regex fallback	none

7 of 16 converters are zero-dependency (stdlib only). audio and video are aliases for yt (same engine, clearer naming).

Architecture

src/any2md/
├── cli.py       # Unified entry point — auto-detect + subcommands
├── common.py    # Shared: frontmatter builder, logging, output helpers
├── yt.py        # Audio/video/YouTube transcription + speaker diarization
├── pdf.py       # PDF extraction + optional VLM OCR
├── img.py       # Image OCR via vision-language model
├── web.py       # Web URL → markdown via ReaderLM
├── html.py      # Local HTML → markdown via ReaderLM
├── doc.py       # Office documents via markitdown
├── rst.py       # reStructuredText conversion
├── csv.py       # CSV/TSV → markdown tables
├── data.py      # JSON/YAML/JSONL → smart markdown
├── db.py        # SQLite → schema + sample data tables
├── sub.py       # Subtitles (SRT/VTT/ASS) → timestamped markdown
├── nb.py        # Jupyter notebooks → markdown
├── eml.py       # Email (.eml/.mbox) → markdown
├── org.py       # Org-mode → markdown
├── tex.py       # LaTeX → markdown
└── man.py       # Unix man pages → markdown

All tools produce YAML frontmatter with source metadata followed by the converted markdown body.

Pre-download AI models

python scripts/download_models.py --stt       # Parakeet (audio/video)
python scripts/download_models.py --diarize   # Sortformer (--diarize)
python scripts/download_models.py --vlm       # Qwen3.5 (img, pdf --ocr)
python scripts/download_models.py --reader    # ReaderLM-v2 (web, html)
python scripts/download_models.py --all       # Everything

Requirements

• Python 3.11+
• Apple Silicon Mac (M1/M2/M3/M4) — for AI-powered converters
• ffmpeg (brew install ffmpeg) — for audio/video
• Non-AI converters (csv, data, db, nb, eml, org, tex, man) work on any platform

License

MIT