I built an open-source reading companion called Grasper. It helps you read hard science books one page at a time, understand each page properly, and only move forward when you actually get what you just read.
Here is what it is, why I built it, how it works, and what it does not do.
The Problem
People read hard books about physics, engineering, math, and computer science. They read a page, think they understand it, then move on. A day later, most of it is gone.
This is not a discipline problem. It is how memory works.
The forgetting curve is a well-known result. If you do not actively recall something, most of it fades very fast. On top of that, many people struggle to finish self-paced learning. More than 80% of people abandon non-mandatory self-paced courses. One of the main reasons is lack of structured support.
There is also a proven fix: testing yourself. Research shows that quizzing improves learning outcomes. It is not just about reading more. It is about checking understanding while you read.
Most AI reading tools do the opposite. They give you summaries. They make reading faster, but they do not make understanding deeper.
Grasper is different. It slows you down on purpose. It makes you understand one page before it lets you move to the next one.
What Grasper Does
Grasper is a reading tool for hard science books.
You open a book in PDF, EPUB, or plain text. Grasper shows you one page at a time. You read that page. Then you can ask Grasper to explain it.
When you click “Explain This Page”, Grasper creates a plain-language teaching document for that page.
It includes:
- What the page is about
- Why it matters
- A worked example
- How it works
- Three common wrong ideas
- A short study card
It does not give you a lazy summary. It teaches the page from first principles.
Then Grasper asks you five quick questions. These questions are tied to the explanation. You need to answer 4 out of 5 correctly on the first try before the next page unlocks.
If you get one wrong, it tells you which part you need to re-read.
This is the core of Grasper: read one page, understand it, prove that you understood it, then move on.
How It Works
The flow is simple.
- Open a book.
- Grasper parses it into pages.
- You read one page.
- You ask Grasper to explain that page.
- Grasper sends that page to the AI provider you choose.
- The AI returns a first-principles explanation and five questions.
- You answer the questions.
- If you pass, the next page unlocks.
- If you fail, you re-read the weak part and try again.
Your progress is saved locally. If you close the app and open it later, you are exactly where you stopped.
Real Math Support
Hard science books have equations. A lot of reading tools treat equations like random text. That does not work for math, physics, or engineering.
Grasper renders equations properly using KaTeX.
It also has a button that explains each symbol in words. So if you see an equation and you do not know what a symbol means, you can check it directly.
This matters because equations are not decoration in these books. They are the actual content.
Diagram Support
Science books also use figures, graphs, and diagrams. Sometimes the figure is more important than the paragraph next to it.
Grasper can show the figure and use a vision model to explain what the picture means.
So you are not just looking at the diagram. You are learning what it is showing.
Understand Chat
Each page can have its own chat thread.
This chat is seeded with the page explanation. It follows the same teaching rules. You can ask follow-up questions about that page only.
This is useful when the explanation helps, but you still have one small confusion left. Instead of leaving the page with a gap in your understanding, you can clear it right there.
The chat survives restarts. It is saved locally.
Grasp Ledger
Grasper keeps a running note of concepts you have understood.
This is called the Grasp Ledger.
It tracks what you have learned so the tool can avoid repeating too much context. It also automatically trims itself so it does not grow too large.
The point is not just to read pages. The point is to build knowledge that stays with you.
Privacy Model
Your book stays on your machine.
Grasper does not upload your whole book to a server. It only sends the single page you are asking the AI to explain.
Your API key is also stored on your machine, not on some remote server.
You can use any OpenAI-compatible provider. So you choose the AI backend. You choose the key. You control what gets sent.
This matters for people reading personal textbooks, research PDFs, or copyrighted material they already own.
Under the Hood
Grasper is built mainly with Rust, Svelte, and Python.
The desktop app uses Tauri. The frontend uses Svelte. The local database uses SQLite.
The project is split into several parts:
-
grasper-core: shared types, equation extraction, parser traits -
grasper-parser: PDF, EPUB, and TXT parsing -
grasper-ai: OpenAI-compatible AI client, prompts, settings -
grasper-memory: SQLite storage for progress, chat, and grasp ledger -
grasper-cli: terminal tool -
grasper-app: the desktop app built with Tauri and Svelte -
scripts: Python-based PDF pipeline
The CLI lets you test each part from the terminal.
Commands include:
grasper parsegrasper explaingrasper equationsgrasper diagramsgrasper chat
This makes it easier to debug parsing, explanation, math extraction, diagram explanation, and chat without opening the full app every time.
What Grasper Is Not
Grasper is not a speed-reading tool.
It is not a “summarize this book in five minutes” tool.
It is not a tool that gives you the illusion of learning.
If you want to skim a book, Grasper is not for you.
Grasper is for people who want to actually understand hard material. It is slower than normal reading. That is intentional. The slowness is the point.
Who It Is For
Grasper is for people reading:
- Physics
- Engineering
- Math
- Computer science
- Other hard technical books
It is for self-learners, students, and engineers who want a private reading companion that forces real understanding.
The Honest Catch
Grasper depends on the AI model you choose. A better model gives better explanations. A weaker model gives weaker explanations.
Also, hard books are often messy. PDFs can have bad layouts, scanned pages, strange equations, or unclear figures. Parsing them perfectly is not always easy.
Grasper is also intentionally strict. If you do not pass the comprehension gate, you do not move forward. Some people will like this. Some people will hate it.
That is by design.
Final Point
Most AI tools make reading easier. Grasper makes learning stronger.
It keeps your book local. It explains one page at a time. It checks your understanding. It saves your progress. It supports real math, diagrams, and per-page chat.
It is open source.
GitHub link: https://github.com/Sumama-Jameel/grasper
Top comments (0)