doclify

Document-to-knowledge converter. Extracts text from plain text, PDF and DOCX files, splits it into overlapping chunks, and produces an extractive summary using a pure-NumPy TextRank implementation. An optional LLM client can polish the summary when an API key is configured; otherwise a deterministic offline polisher is used automatically.

Features

• Text extraction for .txt / .md (native), .pdf (pdfplumber) and .docx (python-docx). Optional readers are lazily imported and raise a clear error only when the file type actually needs them.
• Word-based chunking with configurable size and overlap.
• Extractive TextRank summarization built on NumPy (cosine-overlap similarity, power-iteration ranking). Deterministic, no network.
• Thin LLM client (doclify.llm.DocLLMClient) behind an offline fallback: when no API key is present, summaries are polished with a local rule-based routine.
• CLI: python -m doclify process <file> [--summary].

Quickstart

pip install -r requirements.txt
python -m doclify process notes.txt --summary
python -m doclify process notes.txt --summary --json
python -m doclify supported
python -m pytest -q

CLI usage

python -m doclify process FILE [--summary] [--chunk-size N] [--overlap N] [--max-sentences N] [--json]
python -m doclify supported

Example:

python -m doclify process tests/data/sample.txt --summary --chunk-size 256 --overlap 32

Environment variables (all optional)

Variable	Default	Purpose
DOCLIFY_API_KEY	OPENAI_API_KEY	Enables remote summary polish
DOCLIFY_MODEL	gpt-4o-mini	Model used by the LLM client
DOCLIFY_BASE_URL	https://api.openai.com/v1	OpenAI-compatible endpoint

When no key is available the offline rule-based polisher is used, so the tool always works without credentials.

Project structure

doclify/
  README.md
  requirements.txt
  .gitignore
  .env.example
  doclify/
    __init__.py
    extractor.py    # text extraction (guarded optional imports)
    chunker.py      # word-based chunking with overlap
    summarizer.py   # TextRank (pure NumPy)
    llm.py          # optional LLM client + offline fallback
    pipeline.py     # end-to-end process_document()
    __main__.py     # CLI entry
  tests/
    data/sample.txt
    test_smoke.py

License

MIT