🧭 Architecture Map

Meier Group · 56 projects · 7 domains · 3 machines

56

Projects

7

Domains

3

Machines

4

Agents

7

OOi Layers

30

Connected

⭐

"What does a human need to be healthy — Body & Soul"

NORTH-STAR.md · Everything flows toward this

System Stack

🌐 Face — What users seeSymbion8 · NMTW · BP Derma · Brand Sites

Symbion8 NMTW bpderma-com TBB THE HUB INCICLOUD

🤖 Agents — Autonomous servicesjDax · eDax · cDax · nDAX

nDAX (4-tier) eDax Mail Channels Auth (CIAM)

🌀 Data Engine — Thropif.ai + OOiIngest · Classify L1–L7 · Store

_search (RAG) MIRAG scrape-face BPMI ERP

⚡ Hub — Everything converges01-meier-festivals

Meier Festivals (THE HUB)

🔧 InfrastructureMesh · Broker · Registry

Skynet/theMesh claude-peers 00x (registry)

Tech Stack

Languages

TypeScriptPythonNode.jsPHP

Frameworks

Next.jsSvelteKitFastAPIFlaskHono

Databases

PostgreSQLSQLiteRedisQdrantNeo4j

AI / Inference

OllamaGemma4Qwen3BitRouterOpenClaw

Runtime

BunDockerPM2launchdsystemd

Infra

TailscalenginxWooCommerceWordPress

🏗️ Architecture

How the 56 projects connect · Data flows into the Hub

Data Flow

User action (Telegram / Web / API) │ ▼ Symbion8 (face — Next.js PWA, WebSocket) │ calls API ▼ Thropif.ai (data engine — classify into OOi L1–L7) │ stores / retrieves ▼ OOi Store (~/.ooi/L1-L7/ — JSON per object, 7 cloud policies) │ ├──→ _search (Qdrant + Neo4j RAG) ├──→ nDAX (4-tier: Router→Compactor→Executor→Validator) └──→ Meier Festivals (THE HUB — all data converges) Parallel ingest pipelines: scrape-face ──→ partner catalogs ──→ INCICLOUD (31k ingredients) BPMI ──→ marketplace crawl ──→ INCICLOUD Mail Channels ──→ classified mail ──→ Symbion8 + _search eDax ──→ video/article analysis ──→ Data Center ──→ _search

Edge Kinds

Kind	Meaning	Example
`data`	Structured data transfer	scrape-face → INCICLOUD
`inference`	LLM / AI model call	nDAX → Meier Festivals
`rag`	RAG retrieval	Mail Channels → _search
`chat`	User-facing messaging	Symbion8 → Meier Festivals
`broker`	Agent coordination	claude-peers → Skynet
`harvest`	Nightly data collection	eDax → _search
`superseded`	Replaced by successor	CommBot → Mail Channels

Domains (7)

🏗️ Core Platform

Hub, messenger, mail, inference

6 projects

🤖 AI & RAG

_search, MIRAG, eDax Original

3 projects

💄 Cosmetics

INCICLOUD, BP Derma, scrapers

14 projects

🕸️ Mesh & DevOps

Skynet, broker, gateway

3 projects

📣 Marketing

NMTW, HUB, CMS, brand sites

16 projects

💰 Finance & ERP

Z-ColdCases, ACCMDG2025

2 projects

🧰 DevTools

00x registry, eDax, benchmarks

11 projects

🤖 The Dax Agents

Three personal agents + a data engine. Each has a clear role.

🤖

jDax — The Executor

"j" = job. Fast, decisive, terse. One question → one answer → one action.

Layer	L5 (Project execution)
Harness	OpenClaw · Telegram @Jdaxx_bot
Model	Qwen3.5 27B (local via BitRouter)
Risk	LOW_RISK auto · DEFERRED queued · BLOCKED refused

TypeScriptPythonOllamaOpenClaw

🔬

eDax — The Researcher

Thorough, detailed. Reads everything Mark shares. Persistent memory.

Layer	L5 (Research)
Channel	channel:agent-edax
Scope	Vault + INCICLOUD + Market data
Risk	Read-only, zero risk

PythonOllamaObsidian

📦

cDax — The Archivist

Records only approved content. Write-once, immutable, approval-gated.

Layer	L5 (Archival)
Scope	MIL/ + MDG/ folders
Risk	Write-once, approval-gated
Encrypt	Special Encrypto technology

🌀

Thropif.ai — The Data Dyson

Vacuum any data. Classify into OOi L1–L7. Store in personal database. API for Symbion8.

Input	Any: files, links, messages, images
Output	OOi L1–L7 classified objects
Consumers	Symbion8 (7 mask views)
Store	~/.ooi/L1-L7/ (JSON per object)

PythonOllamaOOi Store

nDAX — 4-Tier Inference Pipeline

Channel-AI's processing pipeline. Every message passes through 4 tiers before reaching the user:

1. Routergemma4:e4b-it-qat · 4.3GB · ~1s

Classifies intent, routes by risk tier (founder > agent > batch)

2. Compactorgemma4:12b-it-qat · 7.2GB · ~3s

Compresses conversation history to ~200 tokens, preserving key facts

3. Executorqwen3-coder:30b · 18.6GB · ~9s

Runs the actual LLM task — drafts, answers, plans

4. Validatorgemma4:12b-it-qat · 7.2GB · ~3s

Checks output against rules, returns approve/deny/revise

Total: ~16s per message on M3 Max. Package: @meier/ndax in the monorepo.

🗄️ OOi — Object-Oriented Intelligence

7-layer classification system. Every object in the estate gets L1–L7 JSON files.

The 7 Layers

L1

CLOUD_OK

Identity — name, purpose, tech stack, team, status

Claude

L2

CLOUD_AUTH

Structure — folder tree, dependencies, git stats

Claude

L3

LOCAL_PREF

Rules — locked constraints, regulatory, contracts

Owner

L4

LOCAL_PREF

Metrics — file count, size, last modified timestamps

Claude

L5

LOCAL_ONLY

Validation — quality signals, risks, health checks

Claude

L6

LOCAL_ONLY

Worth — strategic value, 7×7 pillar mapping

Claude

L7

NEVER_CLOUD

Private — founder notes, never auto-generated, never leaves machine

Owner

How It Works

# Canonical store
~/.ooi/L1-L7/{object-id}.json

# Per-project planted folders
~/Projects/myproject/.ooi/
  ├── L1.json   (identity)
  ├── L2.json   (structure)
  ├── L3.json   (rules)
  ├── L4.json   (metrics)
  ├── L5.json   (validation)
  ├── L6.json   (worth)
  └── L7.json   (private)

# Scanner
python3 ~/theOOi-2026/deep-scan.py

# Plant into project dirs
python3 ~/theOOi-2026/scripts/plant-ooi.py

# Query via knowledge agent
python3 ~/theOOi-2026/agent-kb/agent.py --search "proj:mf-pkg-ndax"

Cloud Policy

Each layer has a cloud policy controlling where data can go:

Policy	Layers	Meaning
`CLOUD_OK`	L1	Can be sent to any cloud service
`CLOUD_AUTH`	L2	Cloud OK with authentication
`LOCAL_PREF`	L3, L4	Prefer local; cloud only if necessary
`LOCAL_ONLY`	L5, L6	Never leaves the local machine
`NEVER_CLOUD`	L7	Founder-only; never transmitted anywhere

Thropif.ai Integration

Thropif.ai is the intake agent that feeds the OOi system. Any data thrown at it — files, links, messages, images — gets classified into the appropriate layer and stored. Symbion8 consumes the OOi store via Thropif.ai's API, showing 7 filtered "mask" views (one per layer).

⚙️ Infrastructure

3 machines · Tailscale mesh · Local-first with VPS production

Machines

skymac

Apple M3 Max · 128GB unified

Primary development machine

OllamaBitRouterOpenClawlaunchd

Runs: Meier Festivals, Symbion8, nDAX, _search, MIRAG, Mail Channels, eDax, 00x, all local AI inference

nordstar

Hostinger VPS · 72.61.91.140

Production server

PM2Dockernginxsystemd

Runs: INCICLOUD, THE HUB, NMTW, Auth, TBB, CMS Network, claude-peers broker, brand sites

bigmac

Mac Studio · Retiring

Legacy — eDax Original host

OllamaPython

Runs: eDax Original (@Edaxv01_bot). Being retired — mirror at ~/Backups/bigmac-edax/mirror/

Services by Runtime

Runtime	Where	Services
`launchd`	skymac	OpenClaw gateway, mailboard (07:00), mesh-librarian (02:30), jdax-snapshot (00:00), eDax, fleet watchers
`PM2`	nordstar	INCICLOUD (:3015), Auth (:3110), TBB, brand sites (bpderma-beta, dutco, friedhil, hydra-facial, etc.)
`Docker`	nordstar	THE HUB (FastAPI+Postgres+Redis+Celery), BillionMail, CosmosHub, infra-stack
`systemd`	nordstar	claude-peers-broker (:7899), clawdbot-gateway, giver-chat, CMS (nginx+php-fpm)

LLM Gateway

# BitRouter (localhost:4356) — unified LLM proxy
# Routes to local Ollama + cloud providers (OpenRouter)
# Config: ~/.bitrouter/bitrouter.yaml

Local models (Ollama :11434):
  qwen3.5:27b         27.8B   16.2GB   ← jDax primary
  gemma4:12b-it-qat    ─      ← nDAX compactor/validator
  qwen3-coder:30b      ─      ← nDAX executor
  clawd-rift-16k      4.2B    2.5GB    ← backstop

Cloud (via OpenRouter):
  google/gemma-4-31b-it        256k ctx
  deepseek/deepseek-v4-flash   195k ctx
  moonshotai/kimi-k2           195k ctx

Networking

Layer	Tech	Purpose
Mesh VPN	Tailscale	Connects all machines (skymac ↔ nordstar ↔ bigmac)
Agent Broker	claude-peers-mcp :7899	Agent-to-agent communication (on nordstar)
Reverse Proxy	nginx	All VPS domains → PM2/Docker services
DNS	Hostinger	*.meierdynamics.de, bpderma.us, ingredient.cloud
Certs	Let's Encrypt (certbot)	Auto-renew; recurring cert rot — audit quarterly

📋 Project Registry

All 56 projects · Source: projects.yaml

Status	ID	Name	Domain	Host	Stack

Source of truth: ~/Projects/00x/projects.yaml
Regenerate: cd ~/Projects/00x && python3 gen.py
Visual editor: registry-dashboard.html