A free desktop editor for macOS and Windows, and two SDKs: one to manipulate PDFs, one to extract text for AI and RAG pipelines. The engine is pure Rust and runs locally.
Text + bounding boxes for RAG and LLM pipelines · OCR · Markdown / JSON output · Compat layer that follows part of the PyMuPDF API (independent implementation)
Extraction guideA native, local-first PDF editor for Mac and Windows. Free for everyone, including commercial use.
AcroForm PDFs — government forms, bank documents, contracts — fully supported. Every field type, locally, no account needed.
The PDF engine is written in Rust. The interpreter and heavy decoders are #![forbid(unsafe_code)]. The parser is safe Rust with one audited self-referential cache. Safe Rust rules out buffer overflows and use-after-free.
No C. No C++. No third-party PDF engine. Pure Rust, from scratch.
uploaded. Ever. To anyone.
Adobe Acrobat Pro: individual plan, billed annually and paid monthly, Dutch store, VAT included, as listed on adobe.com on 13-10-2026.
macOS 11+ · Windows 10+
Free for everyone — including at work.
The same Rust engine, as two libraries: one to change PDFs, one to read them into an AI pipeline.
Written in Rust. One part of the SDK changes PDFs, the other reads them into your pipeline.
Free under the AGPLv3, or a commercial licence if you would rather not publish your own source.
Higher than pymupdf4llm 1.28.2 on all 4 table measures, and higher than Docling 2.126.0 on 2 of 5, with 2 level. Each value is the PDFluent score minus theirs.
Balanced tier (quality=balanced) with the table model: opt-in, Preview in SDK 1.1, for Python, Node, Rust and the CLI. Measured 14-10-2026; the fully-correct-tables row 08-10-2026 on an earlier build. Pages come from the tuning (development) split of each dataset, so our side is an upper bound, not a held-out result. Paired bootstrap per document, 95% intervals; level means the interval includes zero. The table model (Microsoft TATR v1.1) was trained on PubTables-1M and FinTabNet.c; overlap with these pages has not been checked. Docling 2.126.0, pymupdf4llm 1.28.2. Reproducible with pdfluent-bench (coming).
Lower than each of the six. The range runs from the lightest to the heaviest of them.
Lower than each of the six. Models for the balanced tier are a separate, opt-in download.
LiteParse 2.14.7 started in 115 ms, Docling 2.126.0 in 58 s; the other four fall in between.
About 200× faster than Docling 2.126.0, single-thread. pymupdf4llm 1.28.2 took 222 ms per page.
Each figure carries its date, the tool versions and the machine. We rerun them for releases and update this block when the numbers change.
Methodology and full results