System Architecture
Restricted access · Qualified viewers only
Request Access
Enter Your PIN

← Request access instead

Admin Access

System Architecture

Barbarian Agent Agency — Infrastructure Overview · April 2026

Hardware

Local Server SERVER

  • All runtime workloads
  • Docker: Postgres, FastAPI, n8n, Flask
  • Cloudflare Tunnel (zero inbound exposure)
  • Dedicated on-premise infrastructure

Laptop — MRC_XPS_2025 LAPTOP

  • All code repos + development
  • Cloudflare Pages Deploy via Wrangler
  • Claude Code sessions
  • No Python, no Docker

Infrastructure

Docker (Server)

  • barbarian_db — Postgres
  • barbarian_api — FastAPI
  • barbarian_n8n — n8n
  • barbarian_webhooks — Flask

FastAPI

  • Port 8000, tunneled via Cloudflare
  • Portfolio simulate, prices,
  • returns, securities, treasury
  • Intel scalers API (Phase 2)

Postgres

  • barbarian schema, 19+ tables
  • 1.46M+ price history rows
  • 227 securities
  • Equities, fixed income, crypto,
  • commodities, volatility, FX

n8n (19 workflows)

  • 5 BAA Queue runners
  • 5 Scout intel pipelines
  • 2 Hermes health + audit
  • Analyst + Coach Brief
  • 4 legacy Barbarian-*

Cloudflare

  • 5 Pages sites
  • 1 Worker
  • Tunnel → data.riskdimensions.io
  • DNS for all domains

GitHub

  • riskdimensions org
  • 4 repos
  • rd-memory → auto-pull server
  • riskdimensions-site → CF Pages

Connectivity — Three Ways In

Local LAN 192.168.1.x

  • Home network direct access
  • Zero auth, zero latency
  • Daily development path

Tailnet Mesh LIVE 4/22

  • Private WireGuard mesh
  • Tailnet riskdimensions.io (100.x.x.x)
  • Laptop + server + phone
  • RDP / psql / SMB / webhooks from anywhere
  • Zero inbound ports

Public Tunnel CF TUNNEL

  • data.riskdimensions.io
  • Outbound-only from server
  • Fronts FastAPI for Scorecard, PA,
  • future RD Data API clients
  • Zero inbound ports

Products

AI Agents — Barbarian Agent Agency (BAA)

Intelligence Pipeline — The Compounding Moat

The power is not in any single agent or piece of code. It is in the orchestration — and in Mark's ongoing engagement that compounds it over time.

The Three-Pass Intel Loop LIVE

INGEST (Scout)
13 RSS sources + FMP news every 4h. Pass 1: deterministic taxonomy tags — sentiment, player lens, theme, notable flag.
SCORE (Coach via Haiku)
Pass 2: Haiku reads the canon + Mark's frameworks, scores 1–5, writes Mark View editorial line at ≥4.
SURFACE (Consumers)
Score ≥4 → Scorecard ticker + Monitor. Score ≥3 → daily brief. Score 5 → Substack seed + RIA flash.

Memory Architecture — Compound Intelligence

Master Memory (rd-memory)

Institutional knowledge base. CLAUDE.md, agent specs, frameworks, first principles, brand assets. Loaded into every AI session automatically.

Agentic Memory

Session-by-session learning. Feedback, preferences, project state, and validated decisions written to persistent memory files. Compounds across every Claude Code session.

Organic Input

  • Mark's decisions + feedback
  • Session corrections + confirmations
  • Architecture changes + new specs
  • Coach synthesis + framework updates

External Input

  • Scout: FMP news + 13-source RSS (every 4h)
  • Clerk: email ingestion + classification
  • Analyst: FMP price feeds (daily)
  • Notion: CRM, todos, NBP intake

Three-Location Sync

Laptop (C:\rd-memory)  →  GitHub (riskdimensions/rd-memory)  →  Server (C:\RD\rd-memory) · Hermes pulls daily 6AM