Five repos. One family.

Structured PDF extraction API that turns complex files into consistent JSON.

codex-pdf is a contract-first, read-only extraction engine. It provides canonical document facts in a schema-validated CodexDocument payload so downstream systems stop re-parsing the same files in different ways.

• CodexDocument contract with published schemas
• Read-only extraction boundary by design
• CLI workflows: extract, probe, validate, parity
• Consumer-agnostic output for adapter layers
• AGPL open source with typed Python models

Embeddable PDF viewer with separations, TAC, layers, and annotation overlays.

loupe-pdf is an embeddable web PDF viewer purpose-built for prepress review. It surfaces ink separations, total-area-coverage maps, layer toggles, a single-pixel densitometer, and annotation overlays — the things a prepress operator actually needs when looking at a job.

• Per-channel ink separations (CMYK + spots)
• TAC heatmap and densitometer probe
• Layer toggles + annotation overlays
• Embeddable in any Next.js / React host
• Plugin slots for custom toolbars and panels

Detection-only PDF preflight engine — 500+ checks plus the PDF/X-4 conformance suite.

lint-pdf is the open-source preflight engine. It inspects PDFs against 500+ checks across fonts, color, images, transparency, page geometry, and packaging — plus a 91-check PDF/X-4 conformance suite (ISO 15930-7). Detection-only by design: your originals are never modified.

• 500+ engine checks + PDF/X-4 (ISO 15930-7) conformance suite
• Built-in rulesets for GWG sheetfed, GWG digital, PDF/X-4, packaging
• External imports from PitStop, callas pdfToolbox, Acrobat
• FastAPI service + CLI + Python SDK
• Detection-only: your files are never modified

PDF assay and metadata reporting — surface what's actually inside the file.

assay-pdf reports the structural and content characteristics of a PDF without making a value judgment about them. Page boxes, embedded fonts, color spaces, image inventory, security flags, structure tags — the raw inventory that downstream tools (preflight, accessibility, archival workflows) need to do their job.

• Per-page metadata: dimensions, boxes, rotation, content streams
• Font inventory: embedded vs. referenced, subset state, encoding
• Color and image inventory with downsampling detection
• Security and encryption posture
• Structure tag tree for accessibility tooling

Deterministic PDF writer — rewrite, marks, imposition, and trapping with verifiable lineage.

compile-pdf is the producer side of the family — the writer that transforms PDFs through structural rewrite, mark insertion, imposition, and ink-pair trapping. Every operation is deterministic and SHA-256-tracked, so identical inputs and plans produce identical outputs across machines.

• Four producer engines: rewrite, marks, impose, trap
• Deterministic output with SHA-256 lineage tracking
• Real ink-pair extraction for automated trap zones
• CJD (Compile Job Definition) jobs in JSON or XML
• Post-condition verification gates per producer