Over time, I've spent a lot of hours figuring out the best way to study complex engineering and math concepts with AI. This repository contains the final iteration of that effort—what I have been calling "Learning Mode".
It acts as an intelligent, rigorous layer between you and your study materials. Instead of just reading a textbook or staring at slides, you drop the material into your workspace and tell the agent: "Explain pages 9 to 12." It defaults to a very detailed, step-by-step analysis of the content, breaking down equations, quotes, and figures, and then augments that with its own internal knowledge. You can also ask it highly specific, niche questions about the material.
If you use CLI-based agents like Claude Code, you will run into a big limitation with learning mode, because terminals are terrible at rendering complex LaTeX equations. Staring at raw \frac{\partial}{\partial x} in a terminal window completely ruins the learning experience.
To solve this, I designed this agent with a smart output-routing mechanism. If you ask a quick question or need a short definition, the agent replies directly in the terminal using plain text and simple Unicode (e.g., x²). But if you ask for a derivation, a proof, or a detailed breakdown of a 5-page PDF section, the agent stays quiet in the terminal. Instead, it writes its full, beautifully LaTeX-rendered response directly into a continuous Session Log file in Obsidian.
It prepends new answers to the top of the Markdown file, meaning you get a beautifully formatted, reverse-chronological study log of your entire session, complete with Obsidian callouts and math blocks.
The agent is designed to manage a structured study environment inside your Obsidian vault (under a Learning/ folder). For each subject you study (e.g., "Differential Geometry"), it maintains a folder structure with your raw sources/, its generated sessions/, and a central Map of Content (Differential Geometry MOC.md).
Beyond just logging study sessions, you can ask the agent to create permanent notes. If you tell it, "Create an Obsidian note explaining that Nodal Cubic curves concept,", it will generate a standalone Markdown note with the explanation and automatically link it back to your course's MOC file via wikilinks ([[Nodal Cubic Curves]]).
I specifically use and recommend Gemini Pro models for Learning Mode. I find Gemini exceptionally good at (1) reading pdfs, it can understand pretty much everything in the pdf even if it is very badly formatted, and (2) at explaining stuff, it makes good analogies and I feel it graps concepts very good.
This repository provides the prompt in two flavors depending on your ecosystem:
learning-mode-agent.md: The standard agent file, andSKILL.md: The Skills version formatted to your existing agent. Which of the methods you use really doesn't matter, maybe the agent one will be more focused on its task. Just replace<SOURCES_ROOT_PATH>and<OBSIDIAN_VAULT_ROOT_PATH>with your own paths, and you're good to go.