@llamaindex/pdf-viewer

Renders PDF documents in React applications with built-in support for LLM-specific workflows. Uses a component-based architecture that integrates with LlamaIndex's document processing pipeline, enabling PDFs to be displayed alongside LLM chat interfaces and RAG systems. The viewer maintains document state and coordinates with parent LLM application contexts through React props and callbacks.

Provides programmatic control over PDF page navigation through React component props and internal state management. Implements viewport scaling, zoom controls, and page-by-page traversal with event callbacks. Uses canvas or SVG rendering (depending on underlying PDF engine) to maintain responsive performance across document sizes.

Captures user text selections within the PDF viewer and maps them to document coordinates and page positions. Exposes selection events through React callbacks, allowing parent applications to send selected text to LLM systems for analysis or context injection. Maintains mapping between visual coordinates and PDF document space for accurate reference.

Accepts LlamaIndex Document objects or document metadata and binds them to the PDF viewer for coordinated display. Enables the viewer to display document properties (title, source, creation date) and maintain references to LlamaIndex's document node structure. Allows parent applications to pass document context from LlamaIndex loaders directly to the viewer component.

Renders PDFs using canvas or optimized SVG rendering with responsive scaling for different screen sizes and device pixel ratios. Implements lazy loading for off-screen pages and viewport-based rendering to maintain performance on mobile and desktop. Uses requestAnimationFrame for smooth scrolling and resize handling.

Exposes a callback-based event system (onPageChange, onTextSelect, onDocumentLoad, etc.) that allows parent React components to hook into viewer state changes and trigger LLM operations. Events are passed as React props and can be connected to LLM chains, agents, or RAG queries. Enables declarative integration with LLM orchestration frameworks.

Implements keyboard navigation (arrow keys for page navigation, Ctrl+F for search) and ARIA labels for screen reader compatibility. Provides semantic HTML structure and focus management for keyboard-only users. Supports high contrast modes and text scaling for accessibility compliance (WCAG 2.1 AA target).

Provides a data structure and callback system for storing user annotations (highlights, notes, bookmarks) with page and coordinate references. Annotations are stored in a JSON-serializable format that can be persisted to a database or local storage. Enables loading previously saved annotations and syncing them with the PDF display.