The problem
we couldn’t ignore
Every semester, students face the same cycle: notes pile up, past papers gather dust, and the AI tools meant to help them either hallucinate or forget everything between sessions.
AI tools reset.
Knowledge doesn’t.
When you ask ChatGPT about your course material, it draws on everything it was trained on — the entire internet. It can’t tell the difference between your lecture notes and a random blog post. It doesn’t know what your professor actually taught.
When you start a new session, it forgets what you discussed yesterday. No compounding. No memory. No sense of what matters to your studies.
The old way
Generic AI searches the entire web
The ScholarOS way
ScholarOS searches only your vault
The old way
Sessions reset every time you close the tab
The ScholarOS way
Your wiki persists and grows
The old way
Answers can’t cite your specific notes
The ScholarOS way
Every answer traces to your sources
The old way
No structure — just a chat window
The ScholarOS way
Living wiki with interlinked topics
Synthesis happens once,
then gets updated
Karpathy’s LLM Wiki pattern showed that pre-compiling knowledge into structured, interlinked markdown beats re-retrieving from scratch on every question. ScholarOS makes that pattern accessible to every student.
“The heavy lifting happens at ingest time — one cost — instead of query time — repeated cost. That’s the shift.”
Views on Karpathy's LLM Wiki gist
Words in Karpathy's example wiki
Articles in a single knowledge base
Two paradigms,
one coherent system
ScholarOS fuses Obsidian’s local-first, link-based vault with Karpathy’s AI wiki compilation. Your notes are plain markdown you own. The AI maintains a living knowledge base on top of them.
Obsidian compatibility
Your vault is a standard Obsidian vault. Use the same backlinks, graph view, and community plugins you already love. No migration, no lock-in.
AI wiki compilation
The AI reads your raw sources, extracts concepts, and maintains interlinked topic pages. Every query and ingest updates the wiki.
Scoped to your material
The AI never searches the open web by default. Your lecture notes, textbooks, and past papers define the boundary.
Multi-provider access
One subscription gives you access to GPT-5, Claude 4, and Gemini. Use the model that fits each task without managing API keys.
What we believe
about academic AI
Every design decision in ScholarOS follows from a small set of beliefs about what AI should and shouldn’t do for students.
Scope is the differentiator
An AI that only knows your material will always outperform one that knows everything. Constraints produce better answers.
Knowledge should compound
Every session should make the next one smarter. Wikis that grow with use beat systems that forget between queries.
Local-first means you-first
Your notes are plain files on your machine. You can read them without ScholarOS. You can leave without losing anything.
No hallucinations by design
The AI is scoped to your vault. It can’t make up facts outside your material because it never reaches beyond it.
Built in the open.
Auditable by design.
ScholarOS is open source under the MIT license. Read the code, audit the AI behaviour, contribute improvements. Your research OS should be as transparent as your research.
Subscribe once, use everything. Your subscription funds development while keeping the code open. No hidden fees, no usage caps.
Your material, mastered.
Start your second brain.
Join the students and researchers who stopped re-discovering and started compounding.