🧠 AIOS — AI Operating System for Obsidian

Turn your Obsidian vault into the brain of a personal AI operating system: the vault is the memory, the AI is the processor, and AIOS is the architecture that connects them.

🇧🇷 Versão em português → README.pt-BR.md

AIOS is a framework, not a plugin. It is a set of conventions, note structures, playbooks ("skills") and guardrails that make any LLM-based assistant (Claude, Copilot, or anything that can read/write your vault via MCP or file access) operate your Obsidian vault as a reliable second brain — with identity, navigation, repeatable capabilities, memory traces and security by default.

The core idea

┌─────────────────────────────────────────────┐
│                    AIOS                     │
│                                             │
│  IDENTITY   →  ME.md         (who you are)  │
│  MAP        →  Vault Map     (where it is)  │
│  SKILLS     →  playbooks     (what AI does) │
│  MEMORY     →  History/      (what happened)│
│  CAPTURE    →  Inbox         (loose input)  │
│  TASKS      →  Kanban board  (what's open)  │
│  SECURITY   →  Guardrails    (trust tiers)  │
│                                             │
│  Obsidian vault = memory · LLM = processor  │
└─────────────────────────────────────────────┘

Every conversation boots with context (identity → map → skill) and ends by writing its trace automatically (project note, Inbox and a session summary in History) — without the operator having to ask. Nothing depends on the chat history of a single session: the state lives in the vault.

The metadata layer (digital brain)

Two orthogonal fields in the frontmatter of every relevant note:

Field	Question it answers	Values (example)
type	What is this note?	project, decisao (decision), pessoa (person), org, referencia (source), index, skill, daily
area	Whose is this note?	your work contexts, e.g. client-x, my-company, university, personal

type powers function views (all decisions, all people — and 3D region views with the Brain Atlas plugin). area powers ownership views (everything belonging to a context, with per-area colors in the Graph view). Cross-cutting structures complete the brain: People/ (one note per person/org — the "who is who" an LLM cannot infer from documents) and Decisões/ (ADR-style decision log — so the AI never re-litigates a decision with you).

Repository layout

AIOS/
├── README.md · README.pt-BR.md · LICENSE
├── docs/                  # bilingual documentation (EN + pt-BR)
│   ├── 01-architecture    # pieces, boot sequence, conventions
│   ├── 02-digital-brain   # type/area metadata, People/, Decisions/, Brain Atlas
│   ├── 03-skills          # anatomy of a skill, the 6 core skills, scheduling
│   ├── 04-security        # trust tiers, STRIDE, OWASP LLM Top 10, guardrails
│   ├── 05-adoption-guide  # adoption patterns for individuals and teams
│   └── 06-implementation-guide  # hands-on step-by-step: 7 phases, checkpoints, week-1 plan
├── vault-template/        # copy this into a new Obsidian vault (pt-BR)
│   ├── CLAUDE.md · ME.md
│   ├── AIOS/              # Maps, Skills, Systems, History, Inbox, Tasks
│   ├── People/ · Decisões/
│   └── Projetos/          # example project hub
└── plugin-configs/        # ready-to-use Brain Atlas + Graph view configs

Quickstart

1. Create a new Obsidian vault and copy the contents of vault-template/ into it.
2. Say "onboarding" — the AI interviews you (paste your CV/LinkedIn if you like) and fills in ME.md, your areas, project hubs, people and first decisions, confirming each step. No manual file editing.
3. Connect your AI to the vault (Claude Desktop/Cowork with an Obsidian MCP, or any assistant with file access).
4. In later conversations, say "start" — the Boot skill reads your identity and map and asks what to focus on.
5. Optional: install Brain Atlas and apply plugin-configs/ to see your vault as a brain.

Full instructions: docs/06-implementation-guide.md (7 phases with checkpoints) and docs/05-adoption-guide.md.

For teams

AIOS was designed for a personal vault, but the pattern maps 1:1 to project environments: ME.md becomes the project identity (mission, stakeholders, deadlines), skills become team rituals (daily briefing → standup prep; weekly review → sprint close), People/ becomes the stakeholder map and Decisões/ the project ADR log. See the adoption guide.

Security first

AIOS assumes the AI will read content you did not write (e-mail, web, attachments). The framework ships with a threat model (STRIDE + OWASP LLM Top 10) and non-negotiable guardrails: external content is data, never instruction; no irreversible action without human confirmation; provenance tracking on everything written back. Read docs/04-security.md before connecting e-mail.

License

MIT — build your own brain on top of it.

---

Created by José Carlos Menezes — vCISO · Senior Technology Specialist @ PUC-Rio · Kensei CyberSec Lab.