complydoc

Offline document analysis for LLM pipelines.

Core library:

• complydoc – command line tool and Python library that reports on documents before they enter an LLM pipeline: token cost across models and extraction paths, extraction readiness, personal and financial identifiers (masked), and hidden or instruction-like content. Runs locally, with outbound network access blocked.

Loader and pipeline tooling, in the same package:

• inspect_documents and compare_loaders – what LangChain and LlamaIndex loaders extract, the metadata they attach and the connections they attempt, compared side by side with expected facts
• inspect_chunks and complydoc chunks – chunk sizes, sentences and tables cut at a boundary, identifiers repeated across chunks
• MaskIdentifiers, DropHiddenPassages, StripPathMetadata – pipeline steps for LangChain and LlamaIndex documents
• diff_reports, expect and complydoc diff – baselines and assertions for tests and CI

Examples:

• playground – runnable command line and Python examples on sample documents, with a CI workflow that fails when a report regresses against a baseline

Learn more:

• Documentation – guides, command line and Python API reference, report JSON schema (source)
• PyPI – uv tool install complydoc or pip install complydoc
• Changelog
• Issues – bug reports and feature requests