HankPDF
github.com
- Free plan
- Windows
- macOS
- Linux
- API
- Self-hosted
What HankPDF is
HankPDF is a free, Apache-2.0 licensed tool that compresses scanned PDFs into smaller, searchable files, with emphasis on medical and legal documents. Processing runs locally; the project says normal use does not upload files, open network connections, collect telemetry, or phone home. The command-line tool creates compressed PDFs, splits output into chunks for size limits, and exports selected pages as JPEG, PNG, or WebP. A Python package exposes a compress() API. The README targets 8–15× typical compression and up to 200× for text-heavy scans, while noting that already-efficient files may not shrink. Output checks compare OCR text, use tile-level SSIM, and audit structure; the project says it refuses instead of silently corrupting content. Python installation requires Tesseract and qpdf binaries. jbig2enc is optional and can improve text-only compression, while the Docker image includes native tools. HankPDF is distributed through PyPI, GHCR, or source, and supports Windows, macOS, Linux, API, and self-hosting.
Who it suits
HankPDF suits people handling scanned medical or legal PDFs who want local processing and command-line or API access. It also fits teams that need self-hosted deployment.
Good to know, and what to watch
- Local processing without uploads or telemetry
- Compresses, splits, and exports selected pages
- Python compress() API available
- Output undergoes OCR and structural checks
- Apache-2.0 license permits commercial use
- Requires separate Tesseract and qpdf binaries
- Already-efficient PDFs may not shrink
- No offline desktop app
- jbig2enc is optional but separate
Plans
Every plan and priceHankPDF
Open source
The verdict
HankPDF focuses on private, local compression of scanned PDFs with verification checks and multiple output modes. Its command-line and API approach is practical for technical workflows, but setup requires native dependencies.
What the maker publishes
- API
- The Python package provides a compress() API as well as the hankpdf console command. source
- Compression
- The README gives a target of 8–15× typical compression and up to 200× on text-dominant scans, while warning that already-efficient files may not shrink. source
- Content checks
- Outputs are checked with OCR-text comparison, tile-level SSIM, and a structural audit; the project says it refuses rather than silently corrupting content. source
- Dependencies
- A Python install requires separate Tesseract and qpdf binaries; jbig2enc is optional but improves text-only compression, and the Docker image includes these native tools. source
- Installation
- The project is distributed through PyPI and GHCR, or can be installed from its source repository. source
- License
- HankPDF is licensed under Apache-2.0, and the README states it is cleared for commercial use. source
- Local processing
- It runs on the user's machine and states that PDFs are never uploaded, with no network, telemetry, or phone-home behavior during normal use. source
- Maker location
- The maker's GitHub organization profile lists Hank AI, Inc. in the US. source
- Output options
- The CLI can produce compressed PDFs, split output into chunks for size limits, and export selected pages as JPEG, PNG, or WebP images. source
- Platforms
- The README documents native installation on Windows, macOS, and Linux, plus Docker use on those systems. source
- Purpose
- HankPDF compresses scanned PDFs into compact, searchable PDFs, with a focus on medical and legal documents. source
- Release integrity
- The security policy says Docker images are signed with cosign and carry SLSA provenance and an SPDX SBOM; PyPI publishing uses checksum and GitHub OIDC provenance. source
- Security
- The security policy says compression does not open network sockets or transmit user data and describes subprocess resource limits for hostile PDF parsing. source
- Support
- The README directs users who remain stuck to open a GitHub issue and attach the output of hankpdf --version. source
Checked against the maker's pages on 11 Oct 2026.