PDF Pals

Performs optical character recognition on scanned PDF documents entirely on the user's Mac without transmitting content to cloud services. Uses native macOS vision frameworks or embedded OCR engines to convert image-based PDF pages into machine-readable text, enabling downstream text analysis and search. The local-first architecture ensures sensitive documents (legal contracts, medical records) remain on-device throughout the OCR pipeline.

Enables natural language queries against PDF content through a chat interface powered by local or integrated LLM inference. The system likely embeds extracted text into vector representations, indexes them for semantic search, and uses retrieval-augmented generation (RAG) to answer questions grounded in the document. Queries are processed locally or via privacy-respecting API calls, maintaining the local-first data philosophy.

Provides seamless integration with macOS file system, Finder, and system services through native APIs (likely NSDocument, UTType, and Cocoa frameworks). Enables drag-and-drop PDF import, system-level file associations, and integration with macOS services menu. Avoids browser-based overhead by using native Swift/Objective-C implementation, enabling faster file operations and tighter OS integration than web-based alternatives.

Stores all processed PDFs, extracted text, chat histories, and user data exclusively on the local Mac file system without automatic cloud synchronization or backup. Data remains under user control with no transmission to remote servers unless explicitly initiated. This architecture eliminates cloud dependency but requires users to manage their own backups and device-level security.

Extracts text from PDF documents (both native text-based and OCR'd scanned PDFs) and builds a local full-text search index enabling fast keyword queries across document content. Likely uses inverted index data structures (similar to Lucene or SQLite FTS) to enable sub-millisecond keyword searches without re-scanning the original PDF. Supports both exact phrase matching and fuzzy/partial matching depending on implementation.

Enables users to add annotations (highlights, underlines, comments, sticky notes) directly to PDFs and stores all markup locally without cloud synchronization. Annotations are embedded in the PDF file or stored in a local sidecar database, preserving them across sessions. Implementation likely uses PDF annotation standards (PDF/A or incremental updates) to maintain compatibility with other PDF readers.

Enables users to query or compare content across multiple PDF documents simultaneously through the chat interface, using semantic embeddings to find related concepts and passages across different files. The system likely maintains separate vector indices for each document and performs cross-document similarity searches or synthesis queries that require information from multiple sources. This capability extends beyond single-document RAG to multi-document reasoning.