PDFChef

PDFChef is a document workspace that runs entirely in your browser. You import several files, drag individual pages out of them into brand-new documents and nested folders, preview the result, and export it — with no intermediate exports and nothing leaving your machine. It's deliberately more than a merge/split tool: it behaves like a file manager for document pages.

The idea#

Sources are immutable. Your work is a composition layer of references to source pages, so the original document is never modified — you build the output you want by pointing at pages, not by destructively editing files. The whole loop is direct manipulation: select, drag, drop, preview, export. Selection works the way you'd expect from a real file manager — a range field (4-49, 1-3,50-100), click/Shift/Ctrl, and marquee drag.

Architecture#

The codebase is strictly layered, and the layering is enforced by ESLint so dependencies only ever point downward:

• domain — pure, framework-free logic: ranges, split, composition, and a command/undo model. Every user action is a named command with a derived inverse, which is what makes session-wide undo/redo tractable. Built with TDD.
• adapters — the format layer: PDF via pdf.js and pdf-lib, plus images, text and OOXML (DOCX/PPTX/XLSX), with rendering and text extraction.
• services — the Zustand store, IndexedDB persistence, thumbnails, import/export, and search/OCR.
• ui — the React shell, workspace, preview and export views, and i18n.

Text extraction runs in a worker so the main thread stays responsive.

Decisions that mattered#

Everything is local-first: no account, no server, no cloud database. Documents live in IndexedDB and never make a network request — pdf.js workers, fonts and the tesseract.js OCR data all ship with the app, so it works fully offline as an installable PWA. Annotations (text, shapes, highlights, checkmarks, signatures) are rendered so the preview matches the export exactly, rather than being baked in a separate pipeline that could drift. Direct folder export uses the File System Access API where it exists (Chromium), but everywhere else a full-featured ZIP export covers the same ground — no capability silently disappears depending on the browser. Pure logic is verified with Vitest; flows are checked end-to-end with Playwright. The UI is bilingual (English and German), English by default.