ritual-ai-multi-agent-system

A full-stack experimental AI system that simulates multiple AI agents interacting, competing, and producing a resolved decision inside a shared environment.

This project extends the previous concepts:

• Project 1: single AI decision system
• Project 2: single continuous AI agent
• Project 3: multi-agent interaction system

---

What this project does

This system simulates multiple AI agents operating in the same environment:

• each agent produces independent decisions
• agents may disagree
• a resolution layer decides the final outcome
• the system executes only the resolved decision
• everything is stored in an audit history

---

Core idea

Instead of a single AI making decisions, this system introduces:

→ multiple AI perspectives
→ conflicting outputs
→ decision resolution layer
→ collective system behavior

This simulates early-stage “multi-agent intelligence systems”.

---

Why this matters

Real-world decision systems are rarely single-source:

• multiple signals exist
• multiple strategies compete
• final decisions are often aggregated

This project explores a simplified simulation of that concept using AI agents.

---

Tech Stack

• Backend: Node.js + Express
• Frontend: HTML + CSS + Vanilla JavaScript
• AI: OpenAI API (optional) + deterministic fallback
• Storage: file-based JSON state

---

System architecture

Market State ↓ Multiple AI Agents ↓ Agent Outputs (A / B / C) ↓ Conflict Resolution Layer ↓ Verification Layer ↓ Execution Layer ↓ State Update ↓ History Log

---

Core features

🧠 Multi-Agent System

• Multiple independent AI agents
• Each agent has its own strategy behavior
• Each produces a separate decision

Example agents:

• Agent A: aggressive strategy
• Agent B: conservative strategy
• Agent C: neutral strategy

---

⚖️ Conflict Resolution Layer

When agents disagree:

The system resolves decisions using:

• majority voting
• weighted logic (optional)
• override rules

Final decision is produced from this resolution step.

---

🔍 Verification Layer

• Ensures final decision consistency
• Rejects invalid or contradictory outputs
• Acts as a safety constraint layer

---

⚙️ Execution Engine

• Applies final decision to system state
• Updates:
  • balance
  • asset
  • step counter

Note: This is a simulation only, not real financial execution.

---

📊 Shared State System

All agents operate on the same state:

• balance
• asset
• step

State only updates after final resolved decision.

---

🧾 Audit History

Every cycle is recorded:

• agent A / B / C decisions
• reasoning from each agent
• final resolved decision
• verification result
• execution result
• updated state snapshot

---

API endpoints

Run single cycle

---

UI features

• 3-panel dashboard layout

  • Control panel (left)
  • Live execution panel (center)
  • History panel (right)

• Real-time multi-agent visualization

• Shows:

  • each agent’s decision
  • reasoning per agent
  • conflict resolution result
  • final decision

• System status indicator:

  • IDLE
  • RUNNING
  • PAUSED
  • ERROR

• Monochrome (black / white / gray) design system

• Smooth transitions and minimal UI noise

---

Footer

Includes persistent attribution:

• X icon (clickable)
• "Created by emir_ethh"
• Both link to: https://x.com/emir_ethh

---

How to run

Install dependencies

npm install

npm start

http://localhost:3002