Marginalia Brings Citing AI to Local PDF and EPUB Readers

Overview

A new tool called Marginalia is redefining how users interact with digital documents by embedding artificial intelligence directly into local PDF and EPUB readers. Unlike traditional AI summarizers that require uploading files to external servers, this application processes documents entirely within the user’s browser, ensuring strict data privacy. The platform allows readers to engage with textbooks, manuals, and literature through a conversational interface that cites exact passages and navigates back to the source with a single tap.

The service is designed to be accessible across devices, including smartphones and desktop computers, while supporting both English and Spanish queries. Currently, users can access sample materials, such as the University Physics textbook from OpenStax, without creating an account. The initial tier offers three free questions per document, providing a lightweight entry point for students and professionals seeking immediate clarification on complex texts.

Core Functionality

Marginalia distinguishes itself by integrating the AI assistant into the margin of the reader rather than placing it in a separate chat window. This design choice ensures that the reading experience remains immersive and uninterrupted. Users read a paragraph, ask a question about it, and receive an answer that is visually anchored to the text. If the AI identifies information not present in the document, it explicitly states so, reducing the risk of hallucinations going unnoticed.

"It is not a chat next to a PDF. It is a reader with the AI in the margin: you read, you ask, you check, all without leaving the page."

The citation system is a central feature of the application. Every answer generated by the AI includes specific references to the source material. Users can tap on these citation numbers to jump directly to the relevant passage, which is then highlighted within the document. This transparency allows readers to verify the AI’s claims against the original text, fostering a more critical and informed engagement with the material.

For users who prefer not to use an account, the application operates with a strict privacy policy. Files are stored locally in the browser, and no document content is uploaded to any server. Only the user’s question and the minimal textual context required to generate an answer are transmitted during the query process. This architecture ensures that sensitive or proprietary documents remain under the user’s control.

Enhanced Reading Experience

Beyond AI interaction, Marginalia addresses common usability issues associated with viewing PDFs on mobile devices. The application features a Text View mode that reflows PDF content to fit the screen width, eliminating the need for zooming or horizontal scrolling. This reflow technology preserves the original layout elements, such as two-column formats, tables, figures, and code blocks, while allowing users to adjust font sizes and switch between light, sepia, and dark themes.

The text view also generates a table of contents for documents that lack one, improving navigation for large files. Users can revert to the original page layout with a single tap, ensuring that complex diagrams or specific formatting are never lost. This flexibility makes the tool particularly useful for academic reading on phones, where standard PDF viewers often provide a poor experience.

Privacy and Account Options

The application offers a tiered approach to privacy and synchronization. For casual users, the local-only mode is sufficient. However, those who wish to sync their progress, highlights, notes, and conversations across multiple devices can create a free account.

When using an account, the document text is processed by the server to build a search index, but the content is not stored permanently. This trade-off allows users to pick up where they left off on a different device while maintaining the privacy of their primary reading sessions. A key requirement for synced accounts is that the user must open the file on each new device at least once to ensure the local index is established.

Users can also highlight text in four distinct colors and add marginal notes. These annotations are linked to the specific passages they reference, meaning that if the font size or theme is changed, the highlights remain correctly positioned on their respective sentences. This ensures that the user’s personal insights and study materials are preserved regardless of viewing preferences.

Getting Started

The platform is currently available for testing with selected public domain books. For example, users can explore University Physics covering topics like Einstein’s Theory of Gravity and Schwarzschild radius. The interface allows users to type their own questions or select from suggested prompts related to the current chapter.

Initial interactions with the sample text are free and do not require registration. The application demonstrates its capability by answering complex physics questions with precise citations, showing how the concept of time dilation near an event horizon is explained in the source material.

"The prediction is that if an object is sufficiently dense, it will collapse in upon itself and be surrounded by an event horizon from which nothing can escape."

For those with their own files, the process is straightforward. Users simply open a PDF or EPUB from their device storage. The application handles the ingestion locally, making the content searchable and askable immediately. This plug-and-play approach, combined with the absence of installation requirements, lowers the barrier to entry for individuals looking to leverage AI for better document comprehension without compromising their data security.