Architecture

The engine is organized as a Rust workspace with narrow internal crates and a single public facade crate.

Pipeline

1. pdf_objects parses the file structure, page tree, and streams.
2. pdf_content tokenizes and parses page content streams into low-level operations.
3. pdf_text interprets text state to extract text items, glyph geometry, and search matches.
4. pdf_targets normalizes rectangle, quad, and quad-group authoring input into canonical page-space geometry.
5. pdf_redact plans and applies redactions against text, vector, image, and annotation content.
6. pdf_writer rewrites the document as a deterministic full save.
7. pdf_wasm exposes the same API to browser code.

Layer boundaries

Authoring layer

Examples:

• drag rectangles
• text selections
• search results
• future regex or OCR matches

This layer belongs in UI code or higher-level orchestration.

Canonical target layer

All authoring tools are compiled into page-space geometry targets:

• rectangles
• quads
• quad groups

This keeps the apply pipeline independent from specific UI concepts.

Apply layer

The apply pipeline works from geometry and page content. It does not care whether a target came from a drag interaction or a text search term.

Design rules

• Page-space geometry is the canonical input to redaction application.
• UI authoring concepts are kept outside the engine.
• Unsupported features fail explicitly when they affect correctness.
• Output preserves unredacted text when content can be safely rewritten.
• Browser integrations should treat preview rendering as separate from redaction logic.