Metadata-Version: 2.4
Name: filegoblin
Version: 1.15.0
Summary: Converts PDFs, Office documents, web pages and more into clean Markdown for LLMs
Home-page: https://github.com/funkyfunc/filegoblin
License-Expression: MIT
License-File: LICENSE
Project-URL: Source, https://github.com/funkyfunc/filegoblin
Requires-Python: >=3.8
Provides-Extra: models
Requires-Dist: filegoblin-models==1.15.0; extra == "models"
Description-Content-Type: text/markdown

# filegoblin

Converts PDFs (including scans and math), Word, PowerPoint, Excel, HTML and web pages, EPUB, email, notebooks, subtitles and audio into clean Markdown for LLMs. It's a single native binary with no Python dependencies. This package installs it and puts `filegoblin` on your PATH.

```bash
pip install filegoblin             # converter, without the PDF vision models
pip install "filegoblin[models]"   # plus the PDF vision models (~510 MB)
```

The **models** extra adds the bundled PDF vision models:
- PP-DocLayoutV3 for layout and reading order
- pix2text MFR for formulas, written out as LaTeX
- Table Transformer for tables
- PP-OCRv6 for OCR on scanned pages

`--vision smart`, the default, sends the pages that need them through the models: scans, math, tables, columns and figures. Without the models, filegoblin uses its fast geometric converter, plus Apple Vision OCR on macOS.

The models come from PyPI like any other package, so they install wherever `pip install` works, including behind a corporate package mirror. You can also add them later with `filegoblin --install-models`, which downloads them from the GitHub release. `filegoblin --models-status` shows which models it finds.

```bash
filegoblin report.pdf > report.md
filegoblin https://example.com/article
filegoblin --mcp        # MCP server for Claude and other agents
```

Documentation: https://github.com/funkyfunc/filegoblin

