camillehd/grasp

Grasp takes an academic paper and turns it into an interactive knowledge graph you can explore and question. You upload a PDF, and the tool maps out every concept the paper relies on as separate nodes, shows how those concepts connect to and depend on each other, and lets you click any node to open a chat where you can ask questions about it in plain English. It also builds a narrative learning path that walks you through the paper's ideas in order, from the basics up to its main contribution, chapter by chapter. Everything runs locally on your own computer rather than on someone else's server. To set it up you need Python 3.12, Node.js version 18 or newer, and a free Anthropic API key, since Grasp uses Claude to analyze the papers you upload. Analyzing one paper typically costs between $0.30 and $0.80 in API credits, though new Anthropic accounts start with some free credit. Setup is a one time step: download the project, run the included setup script, and paste in your API key when asked. After that, using Grasp means starting a backend server and a frontend server in two separate terminal windows, then opening a local web address in your browser. Once running, you drop a PDF into the upload area and wait roughly thirty to sixty seconds while it is analyzed. The resulting graph color codes nodes by type, such as the paper's core contribution, its methods, background concepts, datasets, and how results are measured. You can click through the graph, chat about any node, or switch to the Learning Path view for a guided walkthrough from start to finish. Behind the scenes, five AI agents work in sequence: one parses the PDF into sections and references, one identifies eight to twenty key concepts and writes plain-English descriptions of them, one maps which concepts depend on which others, one fills in background explanations for anything the paper assumes you already know, and a final one checks the dependency links for accuracy. The README states that your PDFs are processed in memory and not stored permanently, though the paper text itself is sent to Anthropic's API to build the graph. This tool suits students, researchers, or anyone who wants a guided, visual way to work through a dense academic paper instead of reading it cold from start to finish.