- “if you’re wiring your entire life into AI tools that didn’t exist six months ago, systems built on someone else’s infrastructure, then what happens when that tool changes? will you reach for your ideas only to realise they were never actually yours?”
- “we need not just file over app, we need file over AI.”
- “don’t just learn about AI — you need to teach AI to learn about you.”
nick milo presents an “ai os” — a three-layer system built on plain text files that isn’t locked into any specific ai tool or app: your ideaverse (knowledge), maps (connective files), and tools (replaceable apps). the core of the system is two portable markdown files — me.md (who you are, how you think, how you want ai to work with you) and a vault map (how to navigate your knowledge base) — plus a skills folder of process files. by keeping your knowledge and ai instructions in portable text files you own, your workflow survives any tool change, model deprecation, or company pivot.
What are the crucial points in this article or video that make it iconic, ideas I want to remember for the rest of my life?
- file over AI — just as obsidian’s “file over app” philosophy protects your notes from app lock-in, your ai processes should live in portable plain-text files you own, not locked inside any specific tool’s settings or prompts. when the next ai paradigm shift hits, your system should still be standing.
- me.md as portable identity — a single 50–80 line markdown file that tells any ai tool who you are, how you think, and how you want to collaborate. point every tool-specific file (claude.md, etc.) to it. one source of truth that travels with you everywhere.
- skills are just markdown — process documentation lives in text files in your own folder, not as app-specific plugins. because they’re plain markdown, they travel to any ai tool that reads text. you own the process, not the app.
nick milo’s core message is that you should build ai workflows you actually own — portable, future-proof systems in plain text that survive tool changes, model deprecations, and company pivots. the deeper intention: your thinking and knowledge should be yours, not someone else’s infrastructure.
-
ai os — three-layer system: ideaverse (your knowledge in plain files), maps (connective tissue files), tools (replaceable apps/models)
-
file over AI — extension of obsidian’s “file over app” philosophy applied to ai workflows
-
ace folder structure — atlas (knowledge), calendar (time-based notes), efforts (projects) — gives ai clear section guides
-
me.md — portable identity file: who you are, how you think, how you want ai to interact with you. points to other core maps via file paths
-
vault map — the ai’s manual to navigate your ideaverse: folder structure, note types, file naming conventions, how to create notes on your behalf
-
skills as markdown files — process documentation (not app plugins) stored in an aios skills folder; travels to any ai tool that reads text
-
ideaverse — your personal universe of knowledge in plain text, the center of the ai os
-
maps and manuals layer — the connective tissue between your knowledge and ai tools; gives ai fast context without scanning thousands of files
-
multi-ideaverse connections — future direction: connecting personal, team, and family ideaVerses through the maps layer while keeping each private
- create
me.mdat 50–80 lines covering: summary of who you are, first principles, values, how you think, working preferences, communication style - create a vault map file describing: folder structure and purpose, note types and where they go, file naming conventions, how ai should create notes
- point tool-specific files (e.g.
claude.md) tome.md— all tools converge on one portable identity file - build a dedicated aios folder for all ai-related maps and skills (except me.md, which lives at root)
- create one or two core skills to start (e.g. how to structure daily notes, how to summarise research), then expand
- review
me.mdand vault map every few weeks — treat like gardening: add, refine, remove what’s outdated - mark ai-generated content clearly with an emoji, tag, or dedicated folder so you always know what’s yours
- keep deeply personal content (journals) outside of what ai can access
- always back up your files
- [0:00] opening premise — what happens when your ai tool changes or disappears?
- [1:31] “file over AI” introduced as the core principle
- [2:14] three layers of the ai os explained — ideaverse, maps, tools
- [5:10] the two core files every ai os needs: me.md and vault map
- [6:42] what me.md looks like — 50–80 lines, portable identity
- [8:18] vault map — manual for ai to navigate your ideaverse structure
- [10:22] skills as the third map layer — markdown process files in your own folder
- [12:29] privacy considerations and the future of local ai models
- how should
me.mdevolve over time — what’s worth updating vs. what should stay fixed as a stable identity anchor? - what’s the right level of detail in a vault map — enough to guide ai without it becoming a maintenance burden?
- how do you manage skill sprawl as your library of process files grows — do you need a skill for your skills?
- when does connecting to other ideaVerses (team, family) become more complex than it’s worth?
- nick milo’s course: linking your ai (builds your own thinking partner, future-proof for any ai)
- obsidian ceo’s “file over app” philosophy — the foundation this system builds on
- claude code / claude co-work — the ai tool nick currently uses with this system
- notebooklm — mentioned as an alternative tool in the tools layer
- nick milo’s ace folder system setup video (linked in the video)
- monthly gardening session: review
me.mdand vault map — prune what’s outdated, add anything that’s shifted in how you work or think. treat it like the monthly review, not a big project - when any AI generates content in your notes, mark it clearly — you already do this in the daily context block (“generated by claude code”), extend this convention consistently across all note types
- before starting any AI session in a new tool, load
me.mdfirst — make it the instinct, not an afterthought
- when you catch yourself re-explaining a workflow to AI for the second time, that’s a skill worth writing — save it to the skills folder instead of repeating yourself next session
- when creating a new note type or vault section, update the vault map the same day — keep it current while context is fresh
- create
me.mdat the vault root — a portable identity file (50–80 lines): who you are, how you think, working preferences, communication style, and links to key maps. separate fromCLAUDE.mdwhich is project-specific;me.mdis the one file that travels to any AI tool - update
CLAUDE.mdto point tome.mdfor identity — keepCLAUDE.mdproject/codebase-specific (ryeones.com), outsource the “who I am” layer tome.md - create a vault map file (e.g.
private/x/vault-map.md) — a polished, AI-readable version of what’s currently spread across.claude/context/vault.mdandCLAUDE.md, covering folder structure, note types, naming conventions, and how to create notes - build an
ai-os/skills/folder at the vault root with a skills index — your skills currently live in.claude/skills/(claude-code-specific); a portable copy means they work with any tool you adopt - use this framework as the foundation for the “ai infrastructure for all brands” things 3 task — build
me.mdequivalents for seeksophie and fomties so any AI brought into those projects gets the right context immediately