Ask questions of your own textbook, offline

GroutApp's AI-Memory feature lets you load a PDF or photograph pages from a physical textbook, then ask questions that the software answers using only that book's content, with all processing happening on your own computer. The system works offline once you've added your material, extracting text, building a search index and running the language model entirely on-device.

This matters if you study on a train, in a library with unreliable internet, or simply want to keep your textbooks and questions private. It also means you're not paying per query or waiting for a remote server to respond.

How the textbook question system works

When you add a PDF or a set of photographs to AI-Memory, GroutApp uses on-device optical character recognition to extract the text. It then splits the material by subject, book and chapter, so you can ask questions scoped to a particular section or across an entire book. The software builds a vector index—a mathematical representation of the text that lets it find relevant passages quickly—and stores that index locally.

When you ask a question, the system searches the index for the most relevant pages, passes those pages to a language model running on your machine, and returns an answer. Crucially, it shows you which pages it drew on, so you can verify the response against the original text. This AI textbook assistant approach keeps the model grounded in your actual material rather than inventing plausible-sounding nonsense.

Offline operation and what it costs in accuracy

The trade-off for offline capability is model size and performance. Cloud services like ChatGPT run models with hundreds of billions of parameters on clusters of expensive hardware. A desktop or laptop runs smaller models—typically in the 3 to 8 billion parameter range—that fit in consumer RAM and run at acceptable speed on consumer processors or graphics cards.

These local AI models are less capable at complex reasoning and more prone to misunderstanding ambiguous questions. They handle factual retrieval well when the answer is clearly stated in the text, but struggle with multi-step inference or questions that require synthesising information from distant parts of a book. The models improve each year, and for most textbook questions—definitions, worked examples, theorem statements—they perform adequately.

You should still check the cited pages. The grounding mechanism reduces but does not eliminate fabricated answers, especially when the question is vague or the relevant passage is dense.

Handling physical books and non-searchable PDFs

Many students own physical textbooks or older PDFs that don't contain selectable text. AI-Memory handles both by running OCR on each page image. You photograph the pages with a phone or document camera, transfer the images to your computer, and add them to GroutApp. The OCR extracts the text, and from that point the process is identical to working with a born-digital PDF.

OCR accuracy depends on print quality and page layout. Textbooks with clear fonts and good contrast work well; photocopied notes or books with complex diagrams and interleaved text are harder. The software shows you the extracted text so you can spot obvious errors, though minor OCR mistakes rarely prevent the retrieval system from finding the right passage.

Organising multiple textbooks and subjects

AI-Memory organises material by subject, book and chapter. You might have three physics textbooks, two on mechanics and one on thermodynamics, plus a set of chemistry notes. When you ask a question, you can scope it to a single chapter, an entire book, or all material in a subject. This prevents the system from pulling irrelevant passages from unrelated books.

The organisation also helps when you return to a topic months later. You see which books you've added, which chapters are available, and can quickly navigate to the section you need. The vector index rebuilds automatically when you add new material, so the question-answering system always works across your current library.

Privacy and control of study material

Because everything runs locally, your textbooks and questions never leave your machine unless you choose to sync them. This matters if you're working with proprietary course materials, unpublished notes, or simply prefer not to send your study habits to a third party. There's no usage log on a remote server, no data mining of your questions, and no risk that a service will change its terms or shut down.

You also avoid the cost and unpredictability of per-query pricing. Cloud AI services typically charge by the token—the number of words processed—and a long textbook chapter can consume thousands of tokens per question. With an offline system, you pay once for the software and ask as many questions as you like.

Limits of retrieval-based answers

AI-Memory answers questions by retrieving relevant passages and summarising them. It does not "understand" the subject in the way a tutor does, and it cannot answer questions that require information outside the textbook. If your book doesn't explain a concept, the system cannot invent an explanation.

It also struggles with questions that require combining information from many pages or making judgements about what's important. A question like "compare the three theories of X" works well if the book has a comparison section, but fails if the theories are explained separately across a hundred pages with no explicit comparison. For that kind of question, you're better off reading the relevant chapters yourself and making notes.

The system is most useful for quick factual lookups, clarifying definitions, finding worked examples, and reviewing material you've already studied. It complements rather than replaces reading.

Practical workflow for exam preparation

A typical use case is reviewing for an exam. You've read the textbook during the term, made notes, and now want to test your recall and clarify weak points. You ask AI-Memory to define terms, explain worked examples, or retrieve theorem statements. Because it shows the source pages, you can quickly jump to the full context if the summary isn't enough.

This works well alongside other GroutApp features: you can use the flashcard system for active recall, the practice questions for testing, and AI-Memory for on-demand lookups when you hit a gap in your knowledge. The AI learning studio keeps everything in one place, so you're not switching between a PDF reader, a web browser, and a note-taking app.

What this doesn't replace

Asking questions of a textbook is not the same as being taught. A good teacher anticipates misunderstandings, provides context, adjusts explanations to your background, and knows which details matter for exams. AI-Memory does none of that. It retrieves and summarises text, and it does so without understanding whether you've grasped the underlying concept.

It also doesn't replace working through problems yourself. Reading an answer to "how do I solve this integral" is not the same as solving ten integrals by hand. Use the question feature to check your understanding and find information quickly, but do the exercises.