Note: this repository does not build standalone yet. The following forward dependencies are not yet published:

• mly-core → https://github.com/andrewdyates/mly — cold-start: dpdf rewrites mly git deps to the public mly slot, but that repo has not been published yet

---

claims:

• type: entrypoint name: dpdf-cli code_path: crates/dpdf-client/src/main.rs
• type: package name: dpdf-core code_path: crates/dpdf-core/
• type: package name: dpdf-pipeline code_path: crates/dpdf-pipeline/
• type: package name: dpdf-server code_path: crates/dpdf-server/
• type: package name: dpdf-wasm code_path: crates/dpdf-wasm/
• type: directory name: docs-dir code_path: docs/

---

dpdf

Pure Rust PDF extraction and document understanding for local, server, and browser workflows.

Author: Andrew Yates andrewyates.name@gmail.com Version: 0.1.0 License: Apache 2.0 Copyright: 2026 Andrew Yates

What Is dpdf?

dpdf extracts structured text, tables, figures, thumbnails, and rendering output from PDF files. The workspace includes:

• dpdf CLI for inspection, extraction, rendering, thumbnails, benchmarking, and pipeline debugging.
• dpdf-server for HTTP extraction services.
• dpdf-wasm for Tier 0 browser extraction with no server round-trip.
• Tiered ML document-understanding pipelines built around local mly-* crates.

Dependency Model

The native extraction workspace (dpdf-core, dpdf-types, dpdf-pipeline, dpdf-client, and dpdf-server) is implemented in Rust and avoids crates.io dependencies. ML tiers depend on local path crates such as mly-core and mly-metal, and the optional browser wrapper crates/dpdf-wasm uses wasm-bindgen.

Installation

dpdf is currently built from source:

cargo build --workspace
cargo test --workspace
cargo build -p dpdf-client --release

The CLI binary is written to target/release/dpdf. If you build with CARGO_TARGET_DIR=target/user, the binary lands at target/user/release/dpdf.

Usage

# Inspect metadata
dpdf info paper.pdf

# Extract page 1 as Markdown
dpdf extract paper.pdf --pages 1

# Render page 1 to PNG
dpdf render paper.pdf --pages 1 -o ./output/

# Generate a first-page thumbnail
dpdf thumbnail paper.pdf

# Inspect pipeline routing and intermediate stages
dpdf pipeline route paper.pdf

For ML-backed extraction, point the CLI at a local model directory:

export DPDF_MODEL_DIR=~/dpdf-models
dpdf extract paper.pdf --tier 1

Reference Docs

• Quick Start
• WASM wrapper
• Throughput benchmark methodology

Workspace Layout

crates/
├── dpdf-core/       PDF parsing, text extraction, rendering, and image codecs
├── dpdf-types/      Shared document data model and serializers
├── dpdf-pipeline/   Tier routing, ML orchestration, and evaluation harnesses
├── dpdf-client/     `dpdf` CLI
├── dpdf-server/     HTTP server surface
└── dpdf-wasm/       Browser-facing Tier 0 wrapper

Additional top-level directories:

• docs/ for user-facing documentation that ships with the public repo.
• tests/ for CLI and integration coverage.

License

Apache License 2.0. See LICENSE.