CortexOS: Business Second brain

CortexOS architecture

Inspiration

CortexOS: Second brain for businesses inspired by Andrej Karpathy's Second Brain

What it does

cortexOS is a personal knowledge base system that turns raw documents into a structured, LLM-maintained wiki. You drop files into a raw/ folder, and Claude Code automatically processes them into structured pages inside a wiki/ folder. The system keeps knowledge organized, interlinked, and queryable without any manual curation.

How we built it

Built on top of Claude Code as the orchestration layer. Four core skills drive the system: cortexOS (main orchestrator), cortexOS-ingest (processes raw files), cortexOS-lint (cleans and normalizes wiki pages), and cortexOS-query (handles knowledge retrieval). Routed through OpenRouter for model access.

Challenges we ran into

OpenRouter introduced noticeable latency into the Claude Code pipeline. Rather than rebuilding the routing layer, we worked around it and kept shipping. Getting the skill chaining logic right across the four components also required iteration.

Accomplishments that we're proud of

A fully autonomous knowledge base that maintains itself. Drop a document in, the system ingests, structures, and links it without human intervention. The four-skill architecture is also cleanly separable, which made it installable as a standalone GitHub repo via npx skills add.

What we learned

Claude Code is a serious orchestration layer, not just a coding assistant. Skills as modular agents that chain together is a powerful pattern for building autonomous document workflows.

What's next for CortexOS

Distributing cortexOS as a single installable GitHub repo. Expanding ingestion to handle more file types. Eventually building a query interface on top so users can ask questions directly against their wiki.