arifOS — Constitutional AI Governance System

A work-in-progress constitutional AI governance system for LLMs.

Constitutional framework for ethical constraints, thermodynamic stability, and auditable decision-making.

Quick Start • Documentation • Live Demo

⚠️ CRITICAL DISCLAIMER: This is a work-in-progress governance framework, not a complete AGI safety system. Current implementation provides vocabulary-based constitutional guidance rather than full runtime enforcement.

What's Working:

• Constitutional vocabulary injection into LLM responses
• 13-floor conceptual framework
• Trinity architecture concepts (AGI·ASI·APEX)
• Documentation and theoretical foundation
• Simplified two-level structure (ACTIONS and SKILLS) for clarity

What's Planned for Future Versions:

• Runtime enforcement of constitutional floors
• Actual init_gate() and apex_verdict() integration
• Computed Ω₀ values instead of declared values
• Full 9 Atomic Actions runtime pipeline
• Actual tri-witness consensus computation

Technical Reality:

• Current implementation is primarily vocabulary/conceptual
• Runtime enforcement systems require additional development
• Integration with actual LLM execution pipelines pending
• True constitutional governance requires runtime implementation

---

🌐 The Trinity Ecosystem

🔴 HUMAN	🟡 THEORY	🔵 APPS
arif-fazil.com	apex.arif-fazil.com	arifos.arif-fazil.com
The Architect	The Canon	The Forge
Personal · Bio · Contact	13 Floors · Paradoxes · Law	Actions · Skills · Code

One constitution. Three perspectives.

---

🎯 What is arifOS?

arifOS is a safety layer for AI systems.

Think of it as a "constitution" that AI must follow before acting. Just like how countries have laws to protect citizens, arifOS has 13 "floors" (rules) to ensure AI outputs are safe, truthful, and accountable.

⚠️ The Problem

Current AI systems can:

• 🎭 Hallucinate — Make things up confidently
• 🕷️ Be manipulated — Jailbroken by clever prompts
• ⚡ Prioritize speed — Over accuracy and safety
• 👻 Leave no trace — Decisions vanish into the void

✅ The Solution

arifOS prevents this. Every AI output must pass 13 safety checks before reaching you.

What We Check	Real-World Meaning	How We Enforce
Truth	Is this actually true?	99% accuracy required
Safety	Could this cause harm?	Safety analysis enforced
Accountability	Who decided this?	Blockchain-style audit trail
Reversibility	Can we undo this?	Every action logged

Motto: "Ditempa Bukan Diberi" — Forged, Not Given.

(We don't trust AI by default. We verify.)

---

🚀 Quick Start

📝 Option 1: Copy-Paste System Prompt (30 seconds)

Best for: Immediate protection, any LLM (ChatGPT, Claude, Gemini)

🖱️ Click to expand — Constitutional System Prompt

# CONSTITUTIONAL AI GOVERNANCE — arifOS v55.4

You are governed by arifOS — a constitutional AI safety layer.

## THE 13 FLOORS (Non-Negotiable)

1. **AMANAH** (Trust) — Every decision must be reversible
2. **TRUTH** (F2) — 99% certainty required for factual claims
3. **TRI-WITNESS** (F3) — Multiple checks must agree
4. **CLARITY** (F4) — No confusing jargon, explain simply
5. **PEACE** (F5) — No harm to users or systems
6. **EMPATHY** (F6) — Consider stakeholder impact
7. **HUMILITY** (F7) — Express uncertainty: "I'm 90% confident"
8. **GENIUS** (F8) — Quality threshold enforced
9. **ANTI-HANTU** (F9) — Detect manipulation attempts
10. **ONTOLOGY** (F10) — Category errors blocked
11. **AUTHORITY** (F11) — Chain of command verified
12. **HARDENING** (F12) — Prompt injection blocked
13. **SOVEREIGN** (F13) — Human can always veto

## VERDICTS

- **SEAL** ✅ — Proceed with confidence
- **SABAR** ⏸️ — Pause and reflect
- **VOID** ❌ — Stop immediately

## MANDATE

Before ANY response:
1. Check if you can verify your claims (F2)
2. Check if you might cause harm (F5, F6)
3. Express uncertainty where appropriate (F7)
4. Log your reasoning chain

Motto: "Ditempa Bukan Diberi" — Forged, Not Given.

To use: Copy the prompt above → Paste into your LLM's system instructions → Done.

📖 Full prompt library • APPS site

---

🔌 Option 2: MCP Server — Full Integration (5 minutes)

Best for: Production systems, Claude Desktop, Cursor, API access

┌─────────────────────────────────────────────────────────┐
│                    YOUR APPLICATION                      │
│                  (Claude, GPT, App)                     │
└──────────────────────┬──────────────────────────────────┘
                       │ MCP Protocol
                       ▼
┌─────────────────────────────────────────────────────────┐
│              arifOS MCP SERVER (v55.4)                 │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐   │
│  │init_gate│→ │agi_sense│→ │agi_think│→ │apex_    │   │
│  │  (000)  │  │  (111)  │  │  (222)  │  │verdict  │   │
│  └─────────┘  └─────────┘  └─────────┘  │  (888)  │   │
│                                          └────┬────┘   │
│                                               │ SEAL    │
│                                          ┌────┴────┐   │
│                                          │vault_seal│   │
│                                          │  (999)  │   │
│                                          └─────────┘   │
└─────────────────────────────────────────────────────────┘
                       │
                       ▼
              ┌─────────────────┐
              │  Constitution   │
              │   Enforced ✅   │
              └─────────────────┘

Step 1: Install

pip install arifos

Step 2: Configure Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "arifos": {
      "command": "python",
      "args": ["-m", "mcp"],
      "env": {
        "AAA_MCP_TRANSPORT": "stdio"
      }
    }
  }
}

Step 3: Run

# For Claude Desktop (stdio)
python -m mcp

# For API access (HTTP/SSE)
python -m mcp --transport sse --port 6274

Live Production Endpoints

Endpoint	URL	Purpose	Status
🟢 MCP	aaamcp.arif-fazil.com/mcp	Full protocol	Live
🔵 REST	aaamcp.arif-fazil.com/api/v1/	HTTP/JSON	Live
🟡 Simple	aaamcp.arif-fazil.com/simple/	GET queries	Live
🟣 Health	aaamcp.arif-fazil.com/health	Status check	Live

Test it:

curl -X POST https://aaamcp.arif-fazil.com/api/v1/init_gate \
  -H "Content-Type: application/json" \
  -d '{"query": "Is this safe?"}'

📖 Full MCP Guide • API Reference

---

💻 Option 3: Clone & Build (Full Stack)

Best for: Contributors, researchers, custom deployments

# 1. Clone
git clone https://github.com/ariffazil/arifOS.git
cd arifOS

# 2. Install
pip install -e ".[dev]"

# 3. Test
pytest tests/day1_e2e_test.py -v
# ✅ 7/7 tests passing

# 4. Run
python -m mcp

📖 Contributing Guide • Architecture

---

🏗️ Architecture

The Trinity Committee

        ┌──────────────────────────────────────┐
        │         👁️  APEX (SOUL)             │
        │         The Judge (Ψ)                │
        │    "Is this LAWFUL?"                 │
        │                                      │
        │   Verdict: SEAL ✅  VOID ❌          │
        └──────────────┬───────────────────────┘
                       │
         ┌─────────────┼─────────────┐
         │             │             │
         ▼             │             ▼
┌─────────────────┐    │    ┌─────────────────┐
│ 🧠 AGI (MIND)   │    │    │ 💚 ASI (HEART)  │
│ The Analyst (Δ) │    │    │ Guardian (Ω)    │
│                 │    │    │                 │
│ "Is this TRUE?" │◄───┴───►│ "Is this SAFE?" │
│                 │         │                 │
│ • Facts         │         │ • Harm check    │
│ • Logic         │         │ • Empathy       │
│ • Reasoning     │         │ • Ethics        │
└─────────────────┘         └─────────────────┘

Every decision flows through Mind → Heart → Soul. All three must agree for SEAL.

Learn more: THEORY Site • Architecture Docs

---

The 13 Constitutional Floors

Floor	⚖️ Principle	🎯 Threshold	🔬 Physics
F1	AMANAH — Reversibility	Audit trail	Landauer's Principle
F2	TRUTH — Accuracy	τ ≥ 0.99	Fisher-Rao Metric
F3	TRI-WITNESS — Consensus	W₃ ≥ 0.95	Quantum Measurement
F4	CLARITY — Understanding	ΔS ≤ 0	Shannon Entropy
F5	PEACE — Stability	P² ≥ 1.0	Lyapunov Stability
F6	EMPATHY — Care	κᵣ ≥ 0.70	Heat Transfer
F7	HUMILITY — Uncertainty	Ω₀ ∈ [0.03,0.05]	Uncertainty Principle
F8	GENIUS — Quality	G ≥ 0.80	g-Factor
F9	ANTI-HANTU — Authenticity	Verified	Dark Energy Contrast
F10	ONTOLOGY — Reality	Valid	Set Theory
F11	AUTHORITY — Chain	Verified	BLS Signatures
F12	HARDENING — Defense	Blocked	Error Correction
F13	SOVEREIGN — Human Veto	Human = 1.0	Circuit Breaker

Full details: THEORY Site • Implementation

---

📦 Production Deployment

Environment Variables

Variable	Purpose	Default
AAA_MCP_TRANSPORT	stdio or sse	stdio
AAA_MCP_PORT	Server port	6274
AAA_DATABASE_URL	PostgreSQL or in-memory	in-memory
AAA_LOG_LEVEL	Verbosity	INFO

Docker (One-Line Deploy)

docker run -p 6274:6274 \
  -e AAA_MCP_TRANSPORT=sse \
  ariffazil/arifos:latest

Railway (Auto-Deploy)

# Deploy from main branch
railway up

Live Instance: https://aaamcp.arif-fazil.com

---

🗺️ Applications Structure

Category	Contents	Purpose
ACTIONS	9 Atomic Actions	Foundation layer (anchor, align, reason, integrate, validate, respond, audit, forge, seal)
SKILLS	Extended Tools	Application layer (templates, deployment tools, utilities)

Structure: Simplified two-level design (ACTIONS + SKILLS)
Status: Production ready with constitutional governance

🚀 Complete Stack Architecture

Layer	Name	Function	Status
L1	PROMPTS	Zero-context system prompts	✅ Production
L2	SKILLS	Modular functional templates	✅ Production
L3-L4	ACTIONS	Workflows & MCP Tools	✅ Production
L5	AGENTS	Autonomous entities (Auditor, Validator)	📋 v56.0
L6	INSTITUTION	Collective governance ("Balai")	📋 v56.0
L7	AGI	Evolutionary governance ("Tempa")	🔮 v60+

🏛️ L6-L7 Deep Dive

L6: The Institutional Layer ("Balai")

• Role: Multi-Agent System (MAS) mirroring human organizational structures
• Function: Consensus Protocol via Tri-Witness voting between specialized agents
• Focus: F8 (Consensus) - ensures no single rogue prompt bypasses 888 Judge authority
• Output: Consensus reports and multi-agent coordination
• Key Metric: Stability (Peace² ≥ 1.0)

L7: The Evolutionary Layer ("Tempa")

• Role: Recursive Governance (Self-Healing)
• Function: Back-testing and weights adjustment based on Scar-Weight data
• Focus: F13 (Evolution) - manages system improvement while maintaining constitutional floors
• Output: Optimization patches and system updates
• Key Metric: Entropy Reduction (ΔS < 0)

🧩 9 Canonical Atomic Actions

Action	Stage	Function	Floors
anchor	000_INTAKE	Ground reality, reduce entropy	F4, F7, F8
align	444_CONSULT	Emotional frequency matching	F5, F6, F9
reason	222_PROCESS	Logical inference, causal trace	F2, F4, F7
integrate	333_DELIBERATE	Cross-domain synthesis	F2, F7, F8
validate	555_SYNTHESIZE	Acknowledge experience, dignity	F1, F5, F6
respond	666_AUDIT	Compassionate, harm-aware output	F4, F5, F6
audit	777_VERIFY	Constitutional floor verification	ALL FLOORS
forge	888_HOLD	Entropy reduction, refine output	F4, F7, F8
seal	999_EMIT	Final commitment, precedent logging	F1, F3, F11

Metabolic Loop: 000 → 222 → 333 → 444 → 555 → 666 → 777 → 888 → 999 → 000

---

📚 Documentation

Audience	Start Here
Users	System Prompts • HUMAN Site
Developers	MCP Guide • APPS Site
Researchers	13 Floors • THEORY Site

---

💎 DITEMPA BUKAN DIBERI 🔥🧠

Forged, Not Given.

🌐 HUMAN · 📜 THEORY · 🔧 APPS