Cognitive Engine Wiki
Modular cognitive architecture for advanced AI agents and research
Brought to you by:
autobotsolution
Menu
▾
▴
Glossary
Glossary
Definitions of terms used throughout the Cognitive Engine documentation.
A
Agent
An autonomous system that can accept goals, create plans, execute actions using tools, observe results, and learn from experience. The agent follows a Think → Plan → Act → Observe → Reflect loop.
AGPL-3.0
GNU Affero General Public License version 3.0. The license under which the Cognitive Engine is distributed. Requires source code availability for network services.
Anthropic
An AI research company that provides the Claude LLM API, which can be used as an LLM provider for the Cognitive Engine.
API Key
A secret key used to authenticate with LLM providers (OpenAI, Anthropic). Must be kept secure and never committed to version control.
Async/Await
Python's asynchronous programming model used for I/O operations to improve performance and scalability.
B
Backpropagation
An algorithm for training neural networks by calculating gradients and updating weights. Not directly used in the Cognitive Engine but relevant to understanding LLMs.
Batch Processing
Processing multiple queries or data points together for efficiency.
Black
A Python code formatter used to enforce consistent code style across the project.
C
Caching
Storing computed results to avoid recomputation. Used for LLM responses, thought graph queries, and memory lookups.
Claude
Anthropic's LLM family, available as an option for the Cognitive Engine's LLM provider.
Cognitive Layer
One of the five processing layers in the Cognitive Engine architecture: Interpreter, Generator, Deliberator, Committer, or Meta-Cognition.
Cognitive Telemetry
Real-time observation and visualization of the cognitive process through the dashboard.
Commitment Layer
The fourth cognitive layer that selects the best thought and converts it to usable output.
Committer
The component implementing the Commitment Layer.
Confidence
A score (0-1) indicating how certain the system is about a thought or answer. Higher confidence means greater certainty.
Confidence Threshold
The minimum confidence required for acceptable output. If no thought meets this threshold after maximum iterations, the best available thought is used.
Configuration
Settings that control the behavior of the Cognitive Engine, typically set via environment variables or a .env file.
Context
Relevant background information that helps interpret a query or problem.
Contributor License Agreement (CLA)
Legal agreement that contributors must sign, granting license rights to their contributions.
D
Dashboard
Web-based interface for real-time visualization of the cognitive process, showing thought graphs, memory updates, and strategy shifts.
Data Augmentation
Technique to increase dataset diversity by creating modified versions of existing data. Not directly used but relevant to ML concepts.
Decision Control
Advanced cognitive feature managing autonomy, authority assessment, and decision override capabilities.
Deliberation
The process of evaluating, testing, and evolving thoughts through internal simulation, stress testing, and comparative scoring.
Deliberation Layer
The third cognitive layer where thoughts are evaluated and evolved.
Deliberator
The component implementing the Deliberation Layer.
Docker
Containerization platform used for consistent deployment across environments.
Docker Compose
Tool for defining and running multi-container Docker applications.
E
Early Stop Confidence
The confidence threshold at which the cognitive process can stop early, even if minimum iterations haven't been reached.
Emotional Simulation
Advanced cognitive feature modeling emotional states, stakeholder attachment, and empathy.
Episodic Memory
The first layer of the memory system that stores raw events and interactions as the foundation for higher-level extraction.
Ethical Alignment
Advanced cognitive feature for moral reasoning, value conflict detection, and welfare optimization.
Evaluation
The process of assessing the quality, correctness, or effectiveness of thoughts, rules, or strategies.
Executor
Component in the agent system that executes actions using available tools.
F
FastAPI
Modern Python web framework used for the dashboard server and potential REST API endpoints.
Flake8
Python style guide enforcement tool used for linting.
Free Software Foundation (FSF)
Organization that created the GPL and AGPL licenses.
G
GPT
Generative Pre-trained Transformer, OpenAI's LLM family available as an option for the Cognitive Engine.
Generator
The component implementing the Generation Layer.
Generation Layer
The second cognitive layer that creates multiple competing thought candidates.
Goal
A high-level objective that an agent aims to achieve through planning and execution.
Goal Validation
Security feature that checks agent goals for safety before execution.
Graph
Data structure consisting of nodes and edges, used for the thought graph to represent relationships between thoughts.
GPT-3.5
Faster, less expensive OpenAI model suitable for many tasks.
GPT-4
More capable but slower and more expensive OpenAI model for complex tasks.
H
Health Check
Endpoint or mechanism to verify that the system is functioning correctly.
Hyperparameter
Configuration parameter that controls the learning process or model behavior (e.g., iteration limits, confidence thresholds).
I
Inner Knowing
Advanced cognitive feature for continuous self-awareness, self-modeling, and identity formation.
Input
Raw data or query provided to the Cognitive Engine for processing.
Interpreter
The component implementing the Interpretation Layer.
Interpretation Layer
The first cognitive layer that transforms raw input into structured ProblemState.
Iteration
One complete cycle through the cognitive layers (Interpretation → Generation → Deliberation → Commitment).
Iteration Count
The number of iterations performed during cognitive processing.
J
Journaling
In SQLite, a mechanism for database recovery and concurrency. WAL (Write-Ahead Logging) mode improves concurrent access.
JSON
JavaScript Object Notation, a lightweight data interchange format used for configuration and data storage.
K
Knowledge Base
Repository of information used to provide context to LLM interactions.
L
Large Language Model (LLM)
A neural network trained on vast amounts of text data that can generate human-like text. The Cognitive Engine uses LLMs for various cognitive operations.
Learning
The process of improving performance over time through experience, pattern extraction, and rule synthesis.
Learning Interval
The number of processing cycles between learning operations.
Learning System
The subsystem responsible for pattern extraction, rule synthesis, and knowledge injection.
LLM Client
The single entry point for interacting with LLM providers (OpenAI, Anthropic).
Lock
Database mechanism to prevent concurrent access conflicts.
Log
Record of events and actions for debugging, monitoring, and auditing.
Logging
The practice of recording events and system state for later analysis.
M
Max Iterations
The maximum number of cognitive iterations to prevent infinite loops. Default: 50.
Memory
The subsystem that stores and retrieves information across sessions. Three-layer architecture: Episodic, Pattern, Rule.
Meta-Cognition
The fifth cognitive layer that governs the thinking process itself, determining when to continue or stop.
Meta-Cognition Layer
The oversight layer that controls iteration depth, stopping conditions, and confidence thresholds.
Min Iterations
The minimum number of cognitive iterations before considering early stop. Default: 3.
Model
A mathematical representation of a system or process. In the Cognitive Engine, refers to Thought and ProblemState objects.
Mutation
The process of modifying a thought to improve it, typically based on identified weaknesses.
Mypy
Static type checker for Python used to catch type-related errors.
N
Neural Network
A machine learning model inspired by biological neurons. LLMs are a type of neural network.
Node
An element in a graph data structure. In the thought graph, nodes represent thoughts.
Nonce
A number used once, often in cryptography to prevent replay attacks.
O
Observer
Component in the agent system that interprets and analyzes execution results.
OpenAI
AI research company that provides the GPT LLM API, which can be used as an LLM provider for the Cognitive Engine.
Open Source
Software whose source code is available for anyone to inspect, modify, and distribute.
Output
The final result produced by the Cognitive Engine after processing.
Overfitting
Machine learning concept where a model learns training data too well and fails to generalize.
P
Parameter
A variable that can be adjusted to control system behavior.
Pattern
A recurring structure or behavior extracted from episodic memory.
Pattern Memory
The second layer of the memory system that stores recurring behaviors and structures.
Pattern Extraction
The process of identifying recurring patterns in episodic memory.
Persistence
The property of data surviving beyond a single session or process.
Planner
Component in the agent system that converts goals into actionable plans.
Premise
The core hypothesis or content of a thought.
ProblemState
A structured representation of a problem including goals, constraints, knowns, unknowns, and context.
ProcessResult
The output of the Cognitive Engine containing the final answer, confidence, iteration count, and reasoning trace.
Prompt
Text input provided to an LLM to elicit a specific type of response.
Prompt Evolution
Experimental feature where the system suggests, tests, and adopts improvements to its own prompts.
Prompt Template
A reusable pattern for generating prompts to LLMs.
Proposer
Component in the prompt evolution system that suggests prompt improvements.
Pydantic
Python library for data validation using type hints.
Pytest
Python testing framework used for the test suite.
Python
Programming language in which the Cognitive Engine is implemented.
Q
Query
A question or request submitted to the Cognitive Engine for processing.
R
Rate Limiting
Restricting the number of requests that can be made in a given time period to prevent abuse.
Reasoning Trace
A record of the cognitive process showing how the engine reached its conclusion.
Recursive
A function or process that calls itself. Used in graph traversal and thought graph analysis.
Refinement
The process of improving a thought by addressing its weaknesses.
Registry
A central repository for managing tools or other resources.
Repository
A storage location for software code and version history (e.g., Git repository).
Revision Lineage
The track of modifications a thought has undergone over time.
Rollback
Reverting to a previous state or version, used in prompt evolution and configuration management.
Rule
A learned strategy derived from patterns that directly modifies reasoning behavior.
Rule Memory
The third layer of the memory system that stores learned strategies and operational rules.
Rule Synthesis
The process of converting patterns into actionable rules.
S
Sandbox
An isolated environment for executing code safely, preventing damage to the host system.
Scoring
The process of evaluating and assigning numerical values to thoughts based on quality metrics.
Scorer
Component that implements scoring functions for thought evaluation.
Secret
Sensitive information such as API keys, passwords, or tokens that must be protected.
Secrets Manager
A service for securely storing and managing secrets (e.g., AWS Secrets Manager, HashiCorp Vault).
Self-Doubt
Advanced cognitive feature for meta-cognitive uncertainty handling and confidence calibration.
SQLite
Embedded SQL database engine used for the memory system.
SQLite3
Python's built-in SQLite database interface.
Stakeholder
An entity affected by or affecting a decision, considered in ethical alignment and emotional simulation.
State
The condition or configuration of a system at a particular point in time.
Stress Testing
The process of challenging a thought to identify weaknesses.
Structured Data
Data organized in a defined format (e.g., objects, dictionaries) rather than unstructured text.
Synthesizer
Component in the learning system that converts patterns into rules.
System Prompt
The initial instructions given to an LLM to establish its role and behavior.
T
Temporal Identity
Advanced cognitive feature for long-term identity continuity and state persistence.
Testing
The process of verifying that software functions correctly.
Thought
A structured entity with state, history, and evaluative properties, representing a hypothesis or idea.
Thought Graph
A graph structure representing interconnected thoughts and their relationships.
Thought Object
The first-class object representation of a thought with properties like premise, confidence, score, and history.
Tool
A capability that agents can use to perform actions (e.g., web search, code execution).
Tool Registry
The central manager for available tools in the agent system.
Trace
A record of events or operations for debugging and analysis.
Transformer
A neural network architecture used in modern LLMs.
Transparency
The property of being open and observable. The Cognitive Engine provides transparency through reasoning traces and the dashboard.
Troubleshooting
The process of identifying and resolving problems.
U
Updater
Component in the learning system that injects learned knowledge into the reasoning system.
User
The person or system interacting with the Cognitive Engine.
Utility
A helper function or module that provides common functionality.
V
Validation
The process of checking that data or configuration meets requirements.
Variable
A named storage location in programming.
Version Control
A system for managing changes to code over time (e.g., Git).
Virtual Environment
An isolated Python environment with its own installed packages.
Visualization
The graphical representation of data or processes, as in the dashboard.
W
Weakness
An identified flaw or limitation in a thought.
WebSocket
A communication protocol providing full-duplex communication channels over TCP, used for the dashboard.
X
XSS (Cross-Site Scripting)
A security vulnerability where malicious scripts are injected into web pages. Prevented by output sanitization.
Y
YAML
A human-readable data serialization language used for configuration files (e.g., docker-compose.yml).
Z
Zero-Knowledge Proof
A cryptographic method where one party proves to another that they know a value without revealing the value itself. Not directly used but relevant to security concepts.
