pdfmuse-haystack

Haystack converter for pdfmuse — a deterministic PDF/DOCX parser for RAG. Same file in → same Document out.

pip install pdfmuse-haystack

Usage

from pdfmuse_haystack import PdfmuseConverter

converter = PdfmuseConverter(mode="markdown")   # or "text" (default)
docs = converter.run(sources=["report.pdf"])["documents"]

In a pipeline:

from haystack import Pipeline
from pdfmuse_haystack import PdfmuseConverter

pipe = Pipeline()
pipe.add_component("converter", PdfmuseConverter(mode="text"))
# ... connect to a splitter / embedder / writer ...

Modes

• "text" (default) — plain reading-order text (fast path).
• "markdown" — structured Markdown (headings + tables).

Extracts text with exact coordinates, tables and structure; no probabilistic models in the core path, so your index is reproducible run-to-run. Scanned/image-only pages surface a NeedsOcr warning (OCR is a pluggable backend, kept out of the core).

MIT OR Apache-2.0 · part of the pdfmuse project.