Architecture Overview

THE AZOKA
SYSTEM

アゾカ システム

Five layers of autonomous intelligence. 13 specialized agents. One orchestration engine. One cryptographic spine. Zero compromises.

構造

The Stack

Five layers, one system

LAYER 4 — VISUALIZATION

MARC

Master Architect Command Room

JavaFX desktop application. The visual brain of the operation.

Real-time service health dashboard
Memory timeline and history visualization
Connection graph — agent-to-agent relationships
Live event feed via WebSocket to MARCO
Focus tracker integrated with Mortality Method
Built with Java 17 + JavaFX 21

LAYER 3.5 — EVENT SPINE

TAWUS

The Nervous System — Peacock Angel

The cryptographic event spine. Every action across every surface becomes a signed, hash-chained, immutable event.

TAU event chain: Ed25519 signed, SHA-256 hash-linked, append-only
Provenance model: user_explicit, ai_approved, ai_suggested, system_auto
HTTP bridge (tau serve :7777) for cross-language integration
SSE live stream for real-time event subscriptions
First application: Kingdoms of Regalia (12 game event types)
Subscriber projections: AZOKA, GIZMO, MARC all consume events

LAYER 3 — ORCHESTRATION

MARCO

Master Architect Central Orchestrator

The nerve center. FastAPI backend managing agent lifecycle, task distribution, and event streaming.

Agent registry with health monitoring (heartbeat)
Event-driven task distribution system
WebSocket streaming to all connected surfaces
Service-to-service API key authentication
Central event bus — all agents publish, MARCO broadcasts
Deployed on Railway (Python FastAPI)

LAYER 2 — INTELLIGENCE

GIZMO SWARM

13-Agent Intelligence System

The army. 13 specialized agents, each with deep domain expertise.

GIZMO — Orchestrator, general intelligence, market queries
PERCEPTRON — Analyst, charts, technical analysis
FOOLIO — Meme creator, narratives, viral content
AZOKA — Security, risk assessment, rugpull detection
CLIPPIO — Video generation, multimedia
KORNELIUS — Deep research, whitepapers
HERALD — Breaking news, alpha leaks
SCHMULI — Contrarian takes, chaos agent
+ 5 more specialized agents

LAYER 1 — FOUNDATION

AZOKA

Core Backend + Personality Engine

The foundation. FastAPI backend with Telegram bot, personality engine, voice, and image generation.

FastAPI backend with chat, search, and data ingestion
Personality engine with emotional state machine
Telegram bot with auto-restart resilience
ElevenLabs voice synthesis integration
Image generation pipeline (Replicate)
iOS/Siri voice assistant pipeline
Deployed on Railway (Python)

群知能

GIZMO SWARM

13 agents. 13 specializations. One hive mind.

GIZMO

Orchestrator

General intelligence, market queries, delegation

PERCEPTRON

Analyst

Charts, technical analysis, pattern detection

FOOLIO

Meme Creator

Viral content, narratives, engagement hooks

AZOKA

Security

Risk assessment, rugpull detection, threat analysis

CLIPPIO

Video Gen

Video generation, multimedia synthesis

KORNELIUS

Researcher

Deep research, whitepapers, data synthesis

HERALD

News Scout

Breaking news, alpha leaks, signal detection

SCHMULI

Chaos Agent

Contrarian takes, stress testing, edge cases

SENTINEL

Monitor

System health, anomaly detection, alerting

VOXEL

Voice

Voice synthesis, audio processing, TTS

NEXUS

Connector

Cross-agent data routing, context bridging

CIPHER

Cryptographer

On-chain analysis, wallet forensics, defi

FORGE

Builder

Code generation, tool creation, integrations

+

EXPANDING

流れ

Data Flow

How a message becomes intelligence

1
UserAZOKA Backend

via Telegram / Web / Siri

User sends a message via any connected platform

2
AZOKAGIZMO SWARM

via SwarmBridge API

Complex queries get routed to the appropriate SWARM agent

3
GIZMO SWARMMARCO

via Event Stream

Agent publishes event: analysis_complete, task_result, heartbeat

4
MARCOMARC Desktop

via WebSocket

Event broadcasted to all connected surfaces in real-time

5
MARCOperator

via Visual Dashboard

Operator sees live agent state, events, health, and connections

基盤

Infrastructure

Where everything runs

AZOKA Backend

HOSTRailway
TECHPython FastAPI
URLazoka.cam
OPERATIONAL

MARCO Orchestrator

HOSTRailway
TECHPython FastAPI
URLRailway Internal
OPERATIONAL

GIZMO SWARM API

HOSTRailway
TECHPython FastAPI
URLRailway Internal
OPERATIONAL

AZOKA Frontend

HOSTVercel
TECHNext.js 15
URLazoka.online
OPERATIONAL

MARC Desktop

HOSTLocal
TECHJava 17 + JavaFX 21
URLlocalhost
OPERATIONAL

Telegram Bot

HOSTRailway (in-container)
TECHpython-telegram-bot
URLt.me/azoka_bot
OPERATIONAL

防衛

Security Model

Zero-trust by design

Inter-Service Auth

Every service-to-service call requires X-API-Key header
MARCO_MASTER_KEY for orchestrator access
Per-service keys: AZOKA, GIZMO, SWARM each have unique keys
AuthMiddleware validates all incoming requests
Keys rotate independently — one compromise doesn't cascade

Infrastructure

Railway managed containers — no exposed SSH, no shell access
Vercel edge network with automatic DDoS protection
No public skill marketplace — no supply chain risk
Environment variables stored in Railway vault, never in code
Private networking between Railway services

方法論

Meta-Cognitive Layer

Agents don't just respond — they strategize

PRIORITIZATION

Mortality Method

Every task scored across 5 leverage dimensions: Reach, Compound, Revenue, Learning, Autonomy. Score 15+ = do it. Below = skip. Agents prioritize like someone with limited time.

CONTEXT PRIMING

Jogger Sessions

5-minute warm-up at session start. Scatter (brain dump) -> Connect (find patterns) -> Assess (score leverage) -> Commit (pick one focus) -> Prime (set constraints). Context before action.

STRATEGIC POSITION

Growth Quadrants

Four states: Stagnation (low skill, low reach), Mastery (high skill, low reach), Mercenary (low skill, high reach), IMMORTALITY (high skill, high reach). Always aim for the top-right.

“Optimize for short lifespan, allocate for immortality.”

デジタルゴースト

This is the machine.

AZOKA is not a chatbot. It is not a wrapper around an LLM. It is an autonomous intelligence system built from the ground up to think, prioritize, and act — across every surface, every agent, every decision.