Agentic OS

Architectural pattern where Claude Code becomes the engine of a personal 'operating system' for your tasks, not just a coding assistant. Three bricks compose it: a shared memory (often Obsidian) that avoids re-explaining everything, a set of standardized skills for consistent outputs, and a dashboard / command center usable even by someone who never opens a terminal. For memory, start with an LLM wiki of Markdown files in Obsidian rather than a full RAG such as LightRAG or RAG-Anything: it is lighter to maintain, and it is enough as long as its index fits in a context window. For daily work, plug in the GWS CLI: Claude Code then drives Docs, mail, Sheets and calendar from the same terminal, and the OS stops being a developer project and becomes your workstation. The dashboard comes last, once those three heads are running, and its job is precise: one button per skill to launch a task without typing a command, next to gauges showing where your usage windows stand. Group skills by everyday domain (productivity, research, content creation) rather than by tool: that is what makes the board readable by someone other than you. The idea: plug Claude Code into shared infrastructure (memory, tools, goals) instead of treating it as an isolated chatbot. A common setup runs in three moves: a brain (Obsidian plus a graph plugin, then Graphify to turn your data into a knowledge graph), a builder framework that generates the OS phase by phase, then publishing it locally or deployed on Vercel or Railway. You can package the whole thing and hand it to your team or to clients.

Strengths

Limitations

Best for

Official site

View on Coeurdar