Every PDF. Best result.
Your keys.

Most RAG pipelines fail before they reach the model

The problem is not the LLM. The problem is document ingestion. Broken column ordering, missing pages, OCR failures, tables flattened, slide decks returning empty text. Most tools extract once and hope for the best.

Page 1: (ok)
Page 2: (ok)
Page 3: 
Page 4: Amoun  Dscriptin
        $450   Consltng    Widgt
        $1200  Setp
Page 5: (ok)
No quality info. No way to know
which pages are broken.

Page 1: good 0.98
Page 2: good 0.96
Page 3: bad → OCR'd 0.91
Page 4: bad → OCR'd 0.87
Page 5: good 0.97
✓ 5 pages, 94% avg confidence
  2 re-extracted with OCR

How it works

pdfmux doesn't extract once and hope. It runs a self-healing pipeline that audits every page, detects failures, and re-extracts them automatically.

Three lines of Python

import pdfmux

# extract as markdown — auto-audits every page
text = pdfmux.extract_text("report.pdf")

# structured json with locked schema
data = pdfmux.extract_json("report.pdf")

# LLM-ready chunks with token estimates
chunks = pdfmux.load_llm_context("report.pdf")
# → [{title, text, page_start, page_end, tokens, confidence}]

Built for LLM pipelines

pdfmux outputs structured content designed for RAG pipelines, vector databases, agent workflows, and knowledge retrieval systems.

• Section boundaries with page references
• Per-page confidence scoring
• Structured chunks with token estimates
• Table and key-value extraction with normalized values
• Schema-guided mapping via fuzzy matching
• Locked JSON schema (API frozen for 1.x)

Structured extraction

Extract tables, key-value pairs, and map them to your own JSON schema — all rule-based, zero LLM cost. Dates, currencies, and rates are normalized automatically.

# structured json with tables + key-values
$ pdfmux convert invoice.pdf --format json --stdout
# outputs:
{
  "schema_version": "1.2.0",
  "tables": [{
    "headers": ["Description", "Amount"],
    "rows": [["Consulting", "$450.00"],
             ["Setup fee", "$1,200.00"]]
  }],
  "key_values": [
    {"key": "Invoice Date", "value": "2026-02-28"},
    {"key": "Total Due", "value": {"amount": 1650.00, "currency": "USD"}}
  ]
}

# schema-guided — map to your own JSON schema
$ pdfmux convert statement.pdf --format json --schema bank.schema.json
# outputs data mapped to your schema fields via fuzzy matching:
{
  "account_number": "1234-5678-9012",
  "statement_date": "2026-02-28",
  "closing_balance": {"amount": 12450.75, "currency": "AED"},
  "transactions": [
    {"date": "2026-02-01", "description": "ADNOC Fuel", "amount": 185.50}
  ]
}

Command line

# convert — auto-audits every page
$ pdfmux invoice.pdf
✓ invoice.md (2 pages, 95% confidence)

# image-heavy pdf — bad pages re-extracted with OCR
$ pdfmux pitch-deck.pdf
✓ pitch-deck.md (12 pages, 85% confidence, 6 OCR'd)

# quality presets — trade speed for accuracy
$ pdfmux convert report.pdf -q fast      # PyMuPDF only, ~0.01s/page
$ pdfmux convert report.pdf -q standard  # auto-audit + OCR fallback (default)
$ pdfmux convert report.pdf --mode premium  # BYOK LLM for scans & complex layouts

# quick triage — per-page quality without extraction
$ pdfmux analyze report.pdf

# structured json output
$ pdfmux convert report.pdf --format json

# LLM-ready format with token estimates
$ pdfmux convert report.pdf --format llm

# start MCP server for AI agents
$ pdfmux serve

Quality presets

One flag controls the speed/accuracy tradeoff. Default is standard — which handles 90% of real-world PDFs with zero cost.

Scanned or handwritten PDFs? Use --mode premium with any LLM provider. The base install handles 90% of digital PDFs; your LLM of choice handles the rest — stamps, handwriting, mixed layouts, low-contrast scans.

Extractors

pdfmux picks the best extractor per page automatically. Install more backends for better results.

Expected accuracy across document types

For ingestion systems like pdfmux, what matters most is semantic chunk accuracy — correct text in the right order with reliable boundaries for RAG.

Aggregate across a mixed dataset:

Benchmark results

Tested on opendataloader-bench — 200 real-world PDFs across academic papers, financial filings, legal contracts, and scanned documents.

#2 overall, #1 among free tools. pdfmux now beats docling (+2.3%), marker (+3.9%), and every other open-source extractor. It achieves 99% of the paid #1 score at zero cost per page — with the best heading detection of any engine, paid or free.

Built-in

MCP server

Give your AI agent the ability to read PDFs. Three tools: convert_pdf for extraction, analyze_pdf for quick triage, batch_convert for directories.

Works with

LangChain & LlamaIndex

# LangChain
from pdfmux.integrations.langchain import PDFMuxLoader

loader = PDFMuxLoader("report.pdf")
docs = loader.load()  # → list[Document]

# LlamaIndex
from pdfmux.integrations.llamaindex import PDFMuxReader

reader = PDFMuxReader()
docs = reader.load_data("report.pdf")  # → list[Document]

Optional extras

# add OCR for scanned pages (~200MB, CPU-only)
$ pip install "pdfmux[ocr]"

# add table extraction
$ pip install "pdfmux[tables]"

# add LLM vision for complex layouts (Gemini, Claude, GPT-4o, Ollama)
$ pip install "pdfmux[llm]"

# add LangChain or LlamaIndex loader
$ pip install "pdfmux[langchain]"
$ pip install "pdfmux[llamaindex]"

# install everything
$ pip install "pdfmux[all]"

What pdfmux is not

• Not a cloud service. pdfmux is a local Python library. Your documents never leave your machine unless you opt into an LLM provider (Gemini, Claude, GPT-4o, Ollama, or any BYOK model). The base install is fully offline.
• Not a GUI. CLI, Python API, and MCP server for developers. No web interface, no upload form — just pip install pdfmux.
• Not locked to one LLM. Bring your own keys — Gemini, Claude, GPT-4o, Ollama, or any OpenAI-compatible API via a 5-line YAML config. pdfmux routes to the best model per page type automatically.
• Not a replacement for individual extractors. pdfmux orchestrates PyMuPDF, OpenDataLoader, RapidOCR, Docling, and your LLM of choice. It doesn't replace them — it routes each page to the right one and re-extracts failures.

Frequently asked questions

From the blog

View all posts →