Skip to main content

DocSift

Convert documents once. Give agents only what they need.

DocSift converts PDFs and Office documents into clean, structured, AI-ready Markdown and JSON — locally, with no cloud APIs. Docling handles PDFs; MarkItDown handles the breadth formats; both sit behind one interface.

v0.1.0

Install

DocSift needs at least one conversion engine:

pip install "docsift[markitdown]"   # Word, Excel, PowerPoint, HTML, CSV, EPUB
pip install "docsift[docling]"      # PDFs (large download: ML layout models)
pip install "docsift[all]"          # both

pip install docsift alone installs the CLI but no engine, and conversion will fail with an install hint.

For local development from a clone:

uv sync --all-extras

Usage

docsift convert report.pdf
docsift convert report.pdf --engine markitdown
docsift --version
docsift compare report.pdf
docsift compare report.pdf --output ./comparison

compare runs every engine on the same document and writes <name>.compare.json (machine-readable metrics) and <name>.compare.md (human-readable report) alongside per-engine output folders.

Output defaults to ./output/ and can be changed with --output DIR.

Chunking and cleaning

Conversion cleans the Markdown (repeated headers/footers, page numbers, image references) and splits it into token-budgeted chunks with heading context:

docsift convert report.pdf --max-tokens 800 --overlap 100
docsift convert report.pdf --keep-image-refs
docsift convert report.pdf --keep-furniture
docsift convert report.pdf --no-cache

Results are cached in ~/.cache/docsift (override with DOCSIFT_CACHE_DIR); an unchanged file with unchanged settings returns instantly.

Known limitations in 0.1.0

  • PDF chunks come from Docling's HybridChunker and are built from the raw document, so the cleaning stages (header/footer removal, page-number stripping) apply to the exported Markdown but not to the chunk text on that path.
  • --overlap has no effect on PDFs chunked by Docling; it applies to the fallback Markdown chunker used for other file types.
  • The result cache in ~/.cache/docsift has no automatic eviction; delete the directory to reclaim space.

License

MIT

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

docsift-0.1.0.tar.gz (29.3 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

docsift-0.1.0-py3-none-any.whl (24.8 kB view details)

Uploaded Python 3

File details

Details for the file docsift-0.1.0.tar.gz.

File metadata

  • Download URL: docsift-0.1.0.tar.gz
  • Upload date:
  • Size: 29.3 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.12.1 {"installer":{"name":"uv","version":"0.12.1","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for docsift-0.1.0.tar.gz
Algorithm Hash digest
SHA256 7fe3d40f3938abb843fc118f2dfe9789ed615804893040c9f0e763b9feae1a31
MD5 a16870beae60e1a22bb634214cb8ec8d
BLAKE2b-256 1fc0670b07f6eb348ef49b7645cbe36e1871bbc6a8c0dff43c7b859da1eacdc2

See more details on using hashes here.

File details

Details for the file docsift-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: docsift-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 24.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: uv/0.12.1 {"installer":{"name":"uv","version":"0.12.1","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

File hashes

Hashes for docsift-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 085fdb95f74d54e4dd38c2536dfba27b2923b7e93916de81375277593f96bdf8
MD5 2753c9f8676132d3dcc91ec3b9e2876e
BLAKE2b-256 d5e802760020acf52475823b61c834e7c8134a87db6c948b590ffcb06d6635e0

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page