The 3-File AI System That Works With ANY MODEL

Summary of Technological Concepts & System Design

Problem with Current AI Workflows

If you “wire your life” into a specific AI tool/provider (e.g., Claude/OpenAI models), your workflow can break when:

• APIs change
• Companies pivot
• Models get deprecated/offline
• Your preferences and structured context are tied to one tool

---

Core Solution: “File over AI” (File-Based, Tool-Agnostic System)

Instead of locking knowledge/process inside a particular AI platform:

• Keep your knowledge and procedures in plain text/Markdown files stored in your own folders.
• Build a system designed to be future-proof and portable across different AI apps and models.

The goal is to avoid depending on any single provider’s internal storage, interfaces, or model behavior.

---

Product/Architecture Concept: “My AI OS” (3-File / 3-Layer System)

Layer 1 — Ideaverse (Your Knowledge Vault)

Your personal “universe” of notes and thinking, organized into three sections:

• Atlas: knowledge
• Calendar: time-related notes
• Efforts (ACE): projects you’re working on

Key idea: the ideaverse consists of plain files, so it remains accessible even if apps change.

---

Layer 2 — Maps & Manuals (Connective Tissue)

A small set of core “map” files helps any AI navigate your ideaverse efficiently—without brute-force scanning.

Why this matters: maps improve context building by directing the AI to relevant files/folders, saving tokens/time/resources.

Two essential map files

1. me.md — Portable identity / working preferences

   • Tells any AI who you are and how it should collaborate with you.
   • Includes preferences such as:
     • preferred tone (e.g., “casual”)
     • relationship style (e.g., “collaborator”)
     • pointers to other map files via file paths
   • Purpose: guidance that should not change often.
   • Portability detail: a tool-specific file (e.g., claude.md) can simply redirect to me.md.

2. Vault map / manual — How to navigate and work inside your ideaverse

   • Acts like a master table of contents / navigation guide.
   • Includes folder/file paths and rules for:
     • how note types should be created
     • where different content goes
     • naming conventions
     • how components interact within the system
   • More changeable than me.md (updates as priorities and vault structure evolve).

---

Layer 3 — Tools (Replaceable Layer)

The actual apps/models can change (Obsidian, Claude, NotebookLM, Gemini, Claude Code, etc.).

Because Layer 1 (Ideaverse) and Layer 2 (Maps/Manuals) are stable, swapping tools shouldn’t break your workflow.

---

“Skills” Concept (Process Documentation as Markdown Files)

A skill is a Markdown file that describes a repeatable AI-assisted process (e.g., daily briefings, summarization, email writing workflow).

• Skills are not tools; they’re text-based mini-manuals that can travel with you to any AI that can read them.
• Skills are stored in a dedicated skills map/folder so AI can find them.
• Suggested approach:
  • start with 1–2 core skills
  • expand over time
  • periodically audit whether maps/identity docs are getting bloated

---

Privacy Guidance & Sharing Model

Using modern frontier tools (e.g., Claude Co-work) implies you’re trusting some tool layer with your data if you want up-to-date capabilities.

Recommended behavior:

• keep deeply personal material (e.g., journals) outside what AI can access
• always backup
• clearly mark AI-generated content (emoji/tag/folders) to distinguish “yours vs AI-assisted”

Long-term expectation: migration toward open-source/local models, supported by the portability of the file-based “AI OS.”

---

Key “Tutorial / Guide” Takeaways

• Create your ideaverse folders first (e.g., in Obsidian: a vault folder plus an ACE structure).
• Build the two Layer-2 core files:
  • me.md (portable identity & collaboration preferences)
  • vault map/manual (navigation + note-creation rules)
• Maintain future-proofing using file paths and redirect tool-specific entrypoints to me.md.
• Maintain maps with a “gardening” rhythm:
  • add/connect/weed
  • occasionally archive or remove outdated parts
• Add additional Markdown “skills” for repeatable tasks, stored centrally for discoverability.

---

Main Speakers / Sources

• Speaker: Nick Milo — creator of the “AI OS” concept and presenter of the system.