Open-Source, Sandboxed Obsidian Agent CLI for Audio Transcription, Mind-Maps, and Deep Research

Hey everyone, I wanted to share something I built to optimize my study workflow. Full disclosure upfront. I built this as part of a student ambassador program with Dedalus Labs, but it’s completely open source, and I designed it specifically to solve my own personal issues with Obsidian note taking. The philosophy behind this project is simple: **AI shouldn't write your final notes, but it should give you a foundation to quickly overview information and help you finalize your own notes much quicker. Specifically, I really wanted to automate initial note linking, drafting, and source aggregation to reduce initial friction of delving into new topics from scratch.** To keep my vault safe, the framework uses a "sandbox" design where all agents are strictly jailed within a single agent folder so they can never modify, break, or read the rest of my vault, which has been an issue with other AI tools. How it handles the heavy lifting: * **Transcription Agent:** Completely skips manual note-taking from audio. It ingests YouTube URLs or voice recordings and outputs polished, timestamped notes. * **Fast Research Agent:** Generates a quick, wiki-style overview on a new topic so I can grasp the basics and obtain sources to further my research. * **Deep Research Agent:** A two-stage setup (Planner + Executor) that turns a broad prompt into a structured, multi-folder curriculum with targeted sub-lessons. This helps me create almost a little primer notebook on a topic that is more in depth than the fast research agent gives. This is helpful for getting my initial notes directory started, and it gives me sources to read and easily evolve the notes beyond the AI's basic introductory analyses. * **Mind Map Agent:** Analyzes a note, hunts down deeper external sources (papers, books), and visually maps them out into a native Obsidian .canvas file. This is just helpful to visualize the ideas of my notes and to be able to break down concepts better and derive questions for various branches. Would love to get any thoughts on improving my system to make the process of lifelong learning of new information more efficient and enjoyable.