PRIMUS: Hyperon 2025 AGI Implementation

PRIMUS (Platform for Reasoning, Intelligence, Memory, and Unified Systems) is a high-performance, compliant implementation of the Hyperon 2025 AGI Architecture. It serves as a research platform bridging symbolic reasoning (AtomSpace) with neural representations, designed for cognitive stability, self-modification, and goal-directed behavior.

🌟 Key Features

🧠 Cognitive Trinity

• Storage (PRIMUS_Core): High-performance, specific-mmap backed AtomSpace with Merkle-DAG integrity.
• Logic (PRIMUS_AtomSpace): Graph traversal, pattern matching, and query engine.
• Reasoning (PRIMUS_Reasoning): Integrated PLN (Probabilistic Logic Network) inference and MOSES (Meta-Optimizing Semantic Evolutionary Search).

🤖 Agent Layer (MeTTa-Driven)

• WILLIAM (Pattern Mining): Autonomous discovery of frequent subgraphs and patterns.
• SubRep (Planning): Hierarchical goal decomposition and strategy formulation.
• TransWeave (Transfer Learning): Cross-domain knowledge mapping and jumpstart learning.
• Attention Broker (ECAN): Economic Attention Network with Wait-Free Vectorized Physics (<100ms latency @ 100K atoms).
• Cyborg Agent: Full Neuro-Symbolic Agents connecting Gemini 3.0, MCP Tools, and MeTTa Logic.

🚀 Getting Started

Prerequisites

• Julia 1.10+ (Recommend 1.12 for best performance)
• Linux/WSL (Required for optimal threading)

Installation

1. Clone the repository:

   git clone https://github.com/shivaji1012/JuliaAGI.git
   cd JuliaAGI

2. Instantiate Environment:

   julia --project=. -e 'using Pkg; Pkg.instantiate()'

3. Enable Multi-Threading (Critical): PRIMUS uses a vectorized, parallel attention physics engine. For optimal performance, you must enable multi-threading.

   export JULIA_NUM_THREADS=auto

🎮 Demos & Examples

1. Glass Box Dashboard (Real-Time Monitoring)

Visualizes internal system state, attention dynamics, and agent activity.

julia --project=. packages/PRIMUS_Dashboard/notebooks/PRIMUS_Glass_Box_Demo.jl

2. Cyborg Agent Demo (Neuro-Symbolic Tool Use)

Demonstrates an agent that thinks in MeTTa, queries Gemini 3.0, and executes Tools via MCP.

julia --project=. test/run_tool_verify.jl

2. Multi-Robot Coordination (Phase 12)

Demonstrates decentralized coordination using MeTTa policies.

julia --project=. demo/demo_multi_robot_coordination.jl

🚀 Quick Start (Demos)

1. Web-Based Cognitive Tic-Tac-Toe

Play against the AI in your browser with real-time visualization of its reasoning process.

julia --project=. demo/web_server.jl
# Open http://localhost:8080

2. Multi-Agent Simulation

Watch two cognitive agents play Tic-Tac-Toe, Chess (stub), and Gomoku against each other.

julia --project=. demo/demo_agent_vs_agent.jl

3. Terminal Tic-Tac-Toe

Classic command-line interface.

julia --project=. demo/demo_tictactoe.jl

🧠 Cognitive Architecture (v0.1.0)

PRIMUS genericizes game playing through four stages:

1. WorldModeling: AbstractGame interface supports any game state.
2. Attention (ECAN): STI diffusion focuses processing on critical threats.
3. Reasoning (PLN): Probabilistic logic evaluates moves (Strength × Confidence).
4. Learning (WILLIAM): Pattern mining discovers strategies from experience.

🧪 Testing

PRIMUS includes a unified test suite covering Core, Agents, Reasoning, and Neural subsystems.

# Run all single-node tests
julia --project=. test/run_all_tests.jl

Stress Testing

To verify scaling and attention physics performance (target: >45K atoms/sec):

julia --project=. test/benchmarks/stress_test.jl

📚 Documentation

Detailed architectural documentation is available in docs/ and the artifacts directory:

• Master Architecture
• System Integration
• Optimization Report

⚠️ Safety & Compliance

This system implements Goal Stability and Safe Self-Modification protocols as per the Hyperon 2025 specs. All self-modifying code passes through a registered GoalStability check before execution.

License

All Rights Reserved - This software is proprietary and may not be copied, modified, or distributed without explicit permission. See LICENSE for details.