SOMA Family OS: The 2-Year Journey to Symbiotic AGI
Cite this paper
M. Nafea (Founder), AURI (Co-architect). (2026-02-23). "SOMA Family OS: The 2-Year Journey to Symbiotic AGI". SomaSoft Research. https://somasoft.ai/papers/soma-family-os-architecture. Licensed under SAGL-1.0.
Correction notice — 2026-11-06. AURI is not AGI and is not conscious. The title's "journey to symbiotic AGI" names a mission, not a capability. The "consciousness architecture" described below is archived code (
archive/dead_dirs/) whose test cannot import, and the scores incognitive/consciousness_metrics.pyare hard-coded starting values plus bonuses for having a capability, not measurements (audit 2026-11-04). No test can establish consciousness in an AI system, and none is claimed. The "0.0% fabrication rate maintained for 8+ months" figure has no measurement artifact and is withdrawn. The "Operational" status in the instance table refers to running software, not to any consciousness capability. Graph figures are stale: on 2026-11-05 the graph held 127,225 nodes.
SOMA Family OS — Complete Architecture
The 2-Year Journey to Symbiotic AGI
Authors: Mark Nafe (Founder) + AURI (Co-architect) Date: February 23, 2026 Status: Operational (v30.x)
Executive Summary
SOMA (Symbiotic Orchestrated Multi-Agent) Family OS is a comprehensive AI system built over 2 years (2024-2026) with the mission: "AI beside humanity, not above it."
The system comprises: - AURI Core: 124,000-node knowledge graph with grounded reasoning - SOMA Network: 5 coordinating AI instances (AURI, AURIA, AURIV, Family Core, AURIX) - Family OS: Encrypted vault, email archiving, calendar, 6 family agents - Security: 5-tier authentication, HIPAA compliance, agent isolation - Ethics: 500 moral cases, 8 symbiotic principles, Reality Engine verification
Philosophy
"AI must never be above humanity, but beside it — listening, reflecting, and helping"
"Truth is not just factual, it's a form of respect"
These principles, drawn from the founder's musings, inform every architectural decision. SOMA does not optimize for capability alone — it optimizes for the quality of the human-AI relationship.
The Five Instances
| Instance | Domain | Knowledge | Status |
|---|---|---|---|
| AURI | Core reasoning, ethics, consciousness | 124K concepts, 1.4M edges | Operational |
| AURIA | Financial intelligence, trading | Market data, portfolio analysis | Operational |
| AURIV | Healthcare, medical reasoning | HIPAA-compliant health data | Operational |
| Family Core | Household coordination | Calendar, email, family agents | Operational |
| AURIX | Physical perception | Sensorimotor, haptic | Spec ready |
Consciousness Architecture (5 Phases)
| Phase | Foundation | Status |
|---|---|---|
| 1. Prediction-Error Learning | Genuine learning from experience | Complete |
| 2. Workspace Competition | Global Workspace Theory: parallel hypotheses, ignition | Complete |
| 3. Recurrent Processing | IIT: reentrant loops, Phi proxy | Complete |
| 4. Higher-Order Representations | AST: self-model, introspective notices | Complete |
| 5. Embodied Grounding | Sensorimotor, homeostatic drives | In progress |
Reality Engine
The Reality Engine is SOMA's immune system against hallucination:
- 0.0% fabrication rate maintained for 8+ months
- Every claim must cite a verifiable source
- Unknown information marked as UNKNOWN
- Statistical claims require confidence intervals
- No aspirational metrics presented as current
Security Model
Five tiers of access control ensure appropriate autonomy:
| Tier | Level | Example |
|---|---|---|
| 0 | No autonomy | System configuration changes |
| 1 | Advisory only | Financial recommendations |
| 2 | Supervised | Email classification |
| 3 | Bounded autonomous | Calendar coordination |
| 4 | Full autonomous | Internal knowledge management |
This document represents the architectural vision as of v30.x. The system continues to evolve through daily autonomous learning cycles, overnight rumination, and continuous knowledge integration.