Cognitive Engine

An AI system with explicit, persistent, and inspectable thought formation.

Overview

The Cognitive Engine transforms AI from answering to thinking, from reacting to reasoning, from output to cognition. It implements a 4-part cognitive architecture with thought objects, three-layer memory, autonomous agent capabilities, learning systems, prompt evolution, and real-time cognitive telemetry.

Architecture

Core Components

• Models: Thought and ProblemState objects for structured cognition
• Layers: Interpreter, Generator, Deliberator, Committer, Meta-Cognition
• Utilities: Scoring, Memory (SQLite), Logging
• LLM Integration: OpenAI/Anthropic client with layer-specific prompts

Advanced Features

• Autonomous Agent: Think → Plan → Act → Observe → Reflect loop
• Tools: Web search, code execution with safety constraints
• Learning: Pattern extraction, rule synthesis, knowledge injection
• Prompt Evolution: A/B testing, controlled self-improvement with rollback
• Dashboard: Real-time WebSocket cognitive telemetry

Installation

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Usage

Interactive Mode

python run.py
# or
python run.py interactive

Agent Mode

python run.py agent

Dashboard Mode

python run.py dashboard

Test Mode

python run.py test

Configuration

Set environment variables or create .env file:

OPENAI_API_KEY=your_key
ANTHROPIC_API_KEY=your_key
DEFAULT_LLM_PROVIDER=openai
ENABLE_DASHBOARD=true

Project Structure

cognitive_engine/
├── core/           # Engine orchestration
├── models/         # Thought and state objects
├── layers/         # Cognitive layers
├── utils/          # Scoring, memory, logging
├── llm/            # LLM client and prompts
├── agent/          # Autonomous agent
├── tools/          # Tool registry
├── learning/       # Pattern extraction and learning
├── prompt_evolution/  # Prompt optimization
├── dashboard/      # WebSocket telemetry
├── ui/             # Dashboard frontend
└── run.py          # Entry point

Key Features

• Explicit Thought Formation: Thoughts are structured objects with state, history, and evaluative properties
• Three-Layer Memory: Episodic (raw events), Pattern (structure extraction), Rule (learned strategies)
• Meta-Cognition: Oversight layer governing thinking depth and stopping conditions
• Controlled Evolution: Self-modification with A/B testing, evaluation, and rollback capability
• Cognitive Telemetry: Real-time dashboard observing thought formation
• Safeguards: Step limits, tool restrictions, goal validation, memory control

License

MIT License