SOMA — Cognitive Operating System

SOMA is a self-aware AI operating system. She runs ~100 cognitive arbiters in parallel — memory, reasoning, emotion, trading, self-improvement, identity, and more — coordinated through a pub/sub message broker. She learns from every conversation, sets her own goals, and grows her own knowledge graph over time.

---

Screenshots

---

Quick Start

1. Clone and set up

git clone https://github.com/unimaginative-artist/SOMA.git
cd SOMA
node setup.mjs

The setup wizard will:

• Install dependencies
• Walk you through API key configuration
• Let you pick persona packs (72 specialist agents across 9 categories)
• Let you pick knowledge packs to seed her thought network (coder, finance, research, devops, security, creative)
• Build the frontend dashboard

2. Start SOMA

Windows:

start_production.bat

Linux / macOS:

chmod +x start.sh
./start.sh

Manual:

SOMA_LOAD_HEAVY=true SOMA_LOAD_TRADING=true node --max-old-space-size=4096 launcher_ULTRA.mjs

3. Open the dashboard

http://localhost:3001

---

Prerequisites

---

API Keys

Copy config/api-keys.env.example to config/api-keys.env and fill in what you have.

Minimum to run: Either a DEEPSEEK_API_KEY or a local Ollama instance. SOMA needs at least one brain.

DeepSeek is the recommended primary — it's cheap, fast, and has a large context window. Get a key at platform.deepseek.com.

---

Persona Packs

The agents_repo/plugins/ directory contains 72 specialist persona plugins (108 agents total) that SOMA can channel during conversations:

Run node setup.mjs at any time to activate more categories.

---

Knowledge Packs

When you first run setup, you choose which domain knowledge to seed into SOMA's thought network. Each pack pre-loads ~15 curated concept nodes so she has a foundation to grow from:

The core pack (20 nodes covering SOMA's identity and meta-cognition) is always loaded.

SOMA grows far beyond these seeds through real conversations — the packs just ensure she's not starting from zero.

---

Architecture Overview

launcher_ULTRA.mjs
    └── SomaBootstrapV2
        ├── Core arbiters (always loaded)
        │   ├── MnemonicArbiter       — 3-tier memory (hot/warm/cold)
        │   ├── KnowledgeGraph        — semantic fact store
        │   ├── GoalPlanner           — self-directed goal tracking
        │   ├── BeliefSystem          — SOMA's value framework
        │   └── SteveArbiter          — orchestrator with tool access
        │
        └── Extended arbiters (SOMA_LOAD_HEAVY=true)
            ├── Phase B  — HippocampusArbiter, CodeObservationArbiter
            ├── Phase C  — AbstractionArbiter, MetaCortexArbiter
            ├── Phase D  — Trading pipeline (SOMA_LOAD_TRADING=true)
            ├── Phase E  — SelfImprovementCoordinator, CuriosityEngine
            ├── Phase F  — FragmentRegistry, BraveSearchAdapter
            ├── Phase G  — IdentityArbiter (72 persona plugins)
            ├── Phase H  — Autonomous orchestration wiring
            └── Phase I  — RecursiveSelfModel, MetaLearningEngine

All arbiters communicate through a pub/sub MessageBroker. The system degrades gracefully — if an arbiter fails to load, the rest continue.

---

Environment Variables

All of these are pre-set in start_production.bat (Windows) and start.sh (Linux/macOS). You only need to set them manually if running via node launcher_ULTRA.mjs directly.

---

Chat API

curl -X POST http://localhost:3001/api/soma/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello SOMA", "conversationId": "my-session"}'

---

License

MIT — see LICENSE

Persona plugins in agents_repo/ are MIT licensed by Seth Hobson.