Agentic OS — Sistema Operacional Claude Code
lanzmarketingagency-eng/agentic-os-template
SO de IA com 3 camadas, memória RAG e pilares swap para qualquer negócio
Ver no GitHubTemplate de arquitetura modular para Claude Code que funciona como um SO pessoal de IA. Usa roteamento em 3 camadas (Map, Rooms, Tools), memória RAG com Obsidian e pilares de capacidade intercambiáveis, permitindo clonar e adaptar o chassis para qualquer tipo de trabalho em uma hora.
Instalação
npx convez-ai lanzmarketingagency-eng/agentic-os-template1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Criar sistema operacional de IA personalizado para equipes ou freelancers
- Automatizar workflows de criação de conteúdo (ex: carrosséis Instagram)
- Montar chassis reutilizável que escala entre clientes e projetos sem reexplicar contexto
- Implementar memória de projeto com graphify + vault para compounding
Veredito

Arquitetura bem pensada e modular que resolve o problema real de reutilização de contexto em Claude Code. O template é concreto (inclui exemplo funcional de carrossel Instagram) e documentação clara, mas exige disciplina para manter os contextos atualizados. Ideal para times que querem escalar Claude como ferramenta principal, não para quem quer algo plug-and-play.
README
A tradução para o português deste README ainda está na fila.
Agentic OS — a shareable Claude Code operating system
Turn Claude Code into your personal AI operating system: a folder architecture that tells Claude where it is, what to do, and where to put the work — so you stop re-explaining yourself and start compounding.
This is a template. Clone it, replace the placeholders, swap the blue pillars for your line of work, and you have an OS tuned to your business in an hour.
Open
docs/architecture.excalidrawat excalidraw.com to see the full map.
The idea in one picture
You / Client
│
CLAUDE CODE ← the conductor (reads CLAUDE.md first)
│
┌─────────────────┴───────────────────────────────┐
│ FOUNDATIONS (always on) CAPABILITIES (swap me) │
│ • Memory (graphify+vault) • Research │
│ • Productivity • Content Creation ★ │
│ • …your pillars… │
│ OPS (workflows, automations) INTEGRATIONS │
└───────────────────────────────────────────────────┘
Foundations stay fixed for everyone. Capabilities swap per role, client, or project. A content creator, a freelancer, and a SaaS team all run the same chassis — they just change the blue pillars. ★ = the worked example shipped in this repo.
Three layers (the routing system)
- Map —
CLAUDE.mdat the root. Claude reads it first, every task. Identity, routing table, naming conventions. - Rooms — each workspace folder has a
CONTEXT.mddescribing its purpose and process. Claude loads only the room it's working in. - Tools — skills / scripts / MCP servers wired into the rooms that need them.
The
content-creationpillar ships with a working tool (below).
This keeps Claude's context small and on-task: it never reads your finance notes while writing a carousel.
Quickstart
git clone <this-repo> my-os && cd my-os
- Open the folder in Claude Code.
- Edit
CLAUDE.md— fill in the[PLACEHOLDERS](who you are, your #1 goal). - Edit each
CONTEXT.mdto describe your own process. - Give Claude a task. It reads the map, routes, and works.
That's it. No framework, no database, no agents to configure.
The worked example: Content Creation → Instagram Carousel
The capabilities/content-creation/ pillar ships with a working tool so you can
see a capability end-to-end. It turns one JSON file into branded 1080×1350 slides
plus a shareable mockup.html.
cd capabilities/content-creation/skills/instagram-carousel
pip install pillow # requires Python 3.10+ and Google Chrome
python carousel.py examples/demo-carousel/content.json --brand examples/demo-carousel/brand.config.json
# → out/demo-carousel/slide-1..6.png + mockup.html
In Claude Code you don't run it by hand: ask for a carousel, Claude reads the
skill's strategy (SKILL.md), writes the slide copy, fills the JSON, and renders.
How to add your own capability (swap a pillar)
- Copy
capabilities/_TEMPLATE/tocapabilities/<your-pillar>/. - Write its
CONTEXT.md(what it does, the process, what good looks like). - Drop any tools into
<your-pillar>/skills/. - Add one row to the routing table in
CLAUDE.md.
Done. Claude now knows the new pillar exists and when to use it.
What stays, what swaps
| Stays (foundations) | Swaps (capabilities) |
|---|---|
| Memory — graphify graphs + Obsidian vault | Research, Content, Sales, Delivery, Finance… |
| Productivity — inbox, mail, calendar, files | whatever your business actually does |
| Ops — automations, scheduled routines | per role / client / project |
Docs
- docs/setup-guide.md — full setup + customization, and why it compounds (start here).
- docs/obsidian-rag-setup.md — build the memory foundation (Karpathy's Obsidian RAG, no vector DB).
- docs/references.md — the source videos.
- docs/architecture.excalidraw — the capability map.
- docs/compounding-effect.excalidraw — the compounding-growth visual.
Credits
The three-layer folder-routing idea (Map → Rooms → Tools) is a community pattern rooted in classic software principles — separation of concerns and modular composition. This template packages it for Claude Code with a working example.
License
MIT — see LICENSE. Use it, fork it, ship your own OS.