Key terms and concepts in relational AI, memory systems, and identity architecture. 50 terms and growing.
This glossary provides comprehensive definitions for key concepts in artificial intelligence, with a focus on relational AI, memory systems, and identity architecture. Whether you're building AI companions, game NPCs, or any AI that needs to remember and relate to users, these definitions will help you understand the foundational concepts.
Agent memory refers to memory systems designed specifically for AI agents—autonomous systems that take actions and pursue goals over time. Unlike conversational memory, agent memory must track goals, plans, action history, and environmental state across extended task execution.
AI alignment is the challenge of ensuring AI systems behave in accordance with human values and intentions. It encompasses making AI helpful, harmless, and honest—doing what we actually want rather than literally following instructions in unintended ways.
An AI companion is an artificial intelligence designed primarily for relationship and emotional connection rather than task completion. Unlike assistants focused on productivity, companions prioritize knowing their users, providing emotional support, and maintaining genuine ongoing relationships.
AI memory refers to the ability of artificial intelligence systems to store, retrieve, and utilize information from past interactions. Unlike stateless AI that treats each conversation independently, AI with memory maintains context across sessions, enabling more personalized and coherent experiences over time.
In AI, APIs (Application Programming Interfaces) are how developers access AI models programmatically. Instead of using a chat interface, APIs let you integrate AI capabilities into applications, workflows, and products through code.
An autonomous agent is an AI system that can independently pursue goals by taking actions, observing results, and adapting its behavior—without requiring human input for each step. Agents can handle complex, multi-step tasks by reasoning and acting in loops.
Chain of thought (CoT) is a prompting technique that improves AI reasoning by asking the model to show its work step-by-step. Instead of jumping to answers, the model explains its reasoning process, leading to better accuracy on complex problems.
A context window is the maximum amount of text (measured in tokens) that an AI model can process at once. It defines how much conversation history and information the AI can 'see' when generating a response. Larger context windows allow for longer conversations but have computational limits.
Conversational context encompasses all the information that informs an AI's understanding of the current conversation—including message history, user profile data, retrieved memories, and situational awareness. Rich context enables relevant, personalized responses.
An embedding is a numerical vector representation of data (text, images, audio) that captures its semantic meaning. Embeddings enable AI systems to understand similarity and relationships between pieces of content, powering search, recommendations, and memory retrieval.
Emotional intelligence in AI refers to a system's ability to recognize, understand, and respond appropriately to human emotions—and in advanced systems, to have its own emotional states that influence behavior. This enables AI to be empathetic, supportive, and emotionally authentic.
Episodic memory in AI stores specific events and experiences—the 'what happened' of past interactions. Unlike semantic memory (facts), episodic memory captures particular conversations, events, and their context, enabling AI to recall specific past experiences.
Few-shot learning is an AI technique where models learn to perform tasks from just a few examples provided in the prompt. Instead of training on thousands of examples, you show the model 2-5 demonstrations and it generalizes to new inputs.
Fine-tuning is the process of further training a pre-trained AI model on a specific dataset to specialize it for particular tasks or domains. It adapts general-purpose models to specific use cases while requiring far less data than training from scratch.
Grounding in AI refers to connecting model outputs to verifiable information sources. Grounded AI responses are based on retrieved documents, databases, or other factual sources—reducing hallucination and enabling fact-checking.
Guardrails are safety mechanisms that constrain AI behavior—preventing harmful outputs, enforcing policies, and keeping AI within acceptable bounds. They're the practical implementation of AI safety, from content filters to output validators.
The identity layer is the infrastructure that gives AI persistent selfhood—combining memory, emotion, personality, and initiative into a coherent identity that persists across all interactions. It's what transforms stateless AI into AI that feels like someone.
Inference in AI refers to using a trained model to make predictions or generate outputs. While training teaches the model, inference is when you actually use it—sending inputs and receiving outputs. Inference speed and cost are key considerations for AI deployment.
A Large Language Model (LLM) is an AI system trained on vast amounts of text data to understand and generate human language. LLMs power modern AI assistants, chatbots, and content generation tools, demonstrating remarkable abilities in conversation, reasoning, and creative tasks.
Latency in AI refers to the time between sending a request and receiving a response. Low latency (fast responses) is crucial for interactive applications, while high latency degrades user experience. Latency depends on model size, infrastructure, and request complexity.
Long-term memory in AI systems refers to persistent storage of information that survives beyond individual conversations. Unlike context windows which reset between sessions, long-term memory allows AI to recall interactions from days, months, or years ago.
Machine learning is a branch of AI where systems learn patterns from data rather than being explicitly programmed. Instead of writing rules, developers provide examples, and algorithms discover the underlying patterns—enabling AI to improve through experience.
Parameters are the learned values in a neural network—the numbers that define what the model knows. Large language models have billions of parameters, and model size (parameter count) correlates with capability, though efficiency matters too.
Multimodal AI refers to systems that can process and generate multiple types of data—text, images, audio, video. Instead of being limited to one modality, multimodal models understand and create across formats, enabling richer AI interactions.
Natural Language Processing (NLP) is the field of AI focused on enabling computers to understand, interpret, and generate human language. NLP powers chatbots, translation services, search engines, and any AI that works with text or speech.
A neural network is a computational system inspired by biological brains, composed of interconnected nodes (neurons) that process information in layers. Neural networks learn patterns from data and are the foundation of modern AI, powering everything from image recognition to language models.
Personality persistence is an AI's ability to maintain consistent character traits, communication style, and behavioral patterns across all interactions. Rather than being a generic assistant, persistent personality means the AI has its own stable identity that users come to know and recognize.
Personalization adapts AI behavior to individual users based on their preferences, history, and context. Personalized AI remembers what you like, adjusts its communication style, and provides relevant responses based on knowing you specifically.
Proactive AI refers to artificial intelligence that initiates interactions rather than only responding to user input. Instead of waiting to be asked, proactive AI reaches out with relevant information, check-ins, or conversation—creating a sense of genuine presence and care.
Prompt engineering is the practice of crafting effective inputs (prompts) to get desired outputs from AI models. It involves understanding how to structure instructions, provide context, and guide AI behavior through careful prompt design.
Relational AI describes artificial intelligence designed to form and maintain meaningful relationships with users over time. Unlike task-focused AI, relational AI prioritizes connection, understanding, and ongoing relationship development.
Retrieval-Augmented Generation (RAG) is an AI architecture that combines information retrieval with text generation. Before generating a response, the system retrieves relevant documents or information from an external knowledge base, then uses this retrieved content to inform the AI's response.
RLHF is a training technique that fine-tunes AI models based on human preferences. Human raters compare model outputs, and the model learns to produce responses humans prefer—making AI more helpful, harmless, and aligned with human values.
Semantic memory in AI stores facts, concepts, and knowledge—the 'what is true' rather than 'what happened.' It holds user preferences, world knowledge, and learned facts independent of specific events where they were learned.
Semantic search is a search methodology that finds results based on meaning rather than exact keyword matching. Using embeddings and vector similarity, semantic search can find conceptually relevant content even when the exact words differ.
Stateless AI refers to AI systems that have no memory of previous interactions. Each request is processed independently with no context from past conversations—every interaction is effectively a first meeting with no relationship history.
Streaming in AI refers to receiving model outputs token-by-token as they're generated, rather than waiting for the complete response. This reduces perceived latency, enables real-time interaction, and allows early stopping if the response is going wrong.
A system prompt is the initial instruction set given to an AI model that defines its behavior, personality, and constraints. It shapes how the AI responds to all subsequent user messages—like programming the AI's character and rules before conversation begins.
Temperature is a parameter that controls randomness in AI model outputs. Low temperature (0-0.3) produces focused, deterministic responses; high temperature (0.7-1.0+) increases creativity and variety. Adjusting temperature is key to getting the response style you want.
A token is the basic unit of text that AI language models process. Text is split into tokens—which might be words, parts of words, or characters—before being fed to the model. Token counts determine context limits, costs, and processing time.
Tool use or function calling allows AI models to interact with external tools, APIs, and systems. Instead of just generating text, the AI can search the web, run code, query databases, or call any function—dramatically expanding what AI can accomplish.
Training data is the dataset used to teach AI models. The quality, quantity, and composition of training data fundamentally determine what a model learns—its capabilities, biases, and limitations. 'Garbage in, garbage out' applies strongly to AI.
The Transformer is a neural network architecture that revolutionized AI by enabling models to process sequences in parallel and learn relationships between distant elements. It's the foundation of all modern large language models including GPT, Claude, and Llama.
Promitheus provides the identity layer that makes AI memory, emotion, and personality work together.
Maintained by the Promitheus Team
Last updated: January 2026 · 50 terms