datalab-to / marker

What This Repo Claims

"Convert PDF to markdown + JSON quickly with high accuracy."

No cloud. No API key. Runs locally. That's the pitch. We tested it specifically as a resume parser for a local AI job seeker stack — parse a PDF privately, feed the Markdown into ollama, tailor a cover letter without any document leaving your machine.

What I Tested

Environment: macOS Apple Silicon, repoverifiertest user, Python 3.13, marker-pdf latest

Task: Convert a test PDF with headers, body text, and a table to clean Markdown — fully locally, no API key.

Install

pip install marker-pdf

One command. No system dependencies, no Java, no LibreOffice, no CUDA. Installed cleanly on Python 3.13 Apple Silicon.

The Code

from marker.converters.pdf import PdfConverter
from marker.models import create_model_dict

converter = PdfConverter(artifact_dict=create_model_dict())
rendered = converter("test.pdf")
print(rendered.markdown)

Three lines. No auth, no config, no API key.

The Output

Section 1: Introduction

This is a test PDF to verify docstrange extraction.

Section 2: Data Table

Name | Age | City

Alice | 30 | London

Bob | 25 | Paris

Headers preserved. Body text clean. Table extracted correctly. First run, no errors.

How It Compares

We tested docstrange before this — same use case, same machine. docstrange's local mode requires CUDA (no Apple Silicon support), and cloud mode hit rate limits on the first document after Google auth.

marker worked on the first try with zero configuration.

The Gotcha

First run downloads models (~1-2GB). Subsequent runs are fully offline. If you're on a slow connection, budget a few minutes for the initial model download.

Also worth noting — very complex layouts with nested tables and forms may not work perfectly, though passing --use_llm and --force_ocr flags largely solves those edge cases. For standard resumes and documents, output is clean.

Verdict: SOLID

Does exactly what it claims. Local, private, no API key, clean Markdown output. Earns its place in any local AI document pipeline. This is the docstrange replacement for anyone who needs PDF parsing without cloud dependency.