Reference
AI Glossary
52 terms in plain language. Look one up when a vendor, a lesson, or a colleague uses a word you have been nodding along to.
52 terms
A
- Acceptable Use Policy AUP
- A written policy defining which AI tools employees may use and how.
- Accuracy
- Whether an AI's output is factually correct.
- Agent
- An AI session pre-configured with a specific purpose, context, and tools before you send your first message.
- Agent Team
- A group of Claude Code sessions that work in parallel on the same problem, each handling a separate piece of the work in its own context window.
- Auto Memory
- Notes that Claude writes for itself as it works, stored in a dedicated memory folder for the project.
B
- Bias AI
- AI models reflecting and amplifying biases present in their training data.
C
- Claude Code
- Anthropic's AI tool built for software developers and technical work.
- Claude Cowork
- A mode within Claude Desktop that enables multi-step task automation.
- Claude Projects
- A feature within Claude's interface that creates a persistent, organized workspace.
- CLAUDE.md
- A plain-text file that gives Claude persistent instructions at the start of every session in Claude Code.
- Connector
- A plugin or integration that lets Claude reach into an external tool — reading emails, searching SharePoint, creating calendar entries, and more.
- Constitutional AI CAI
- An alignment technique developed by Anthropic that gives an AI model a set of explicit principles (a "constitution") and has it critique and revise its own outputs against those principles.
- Context Window
- The AI's working memory for a single conversation.
D
- Desktop Extension
- A connector that runs within Claude Desktop and integrates with local applications and files on your computer.
- Diffusion Model
- The dominant architecture for AI image generation.
E
- Embeddings
- A representation of text (or other content) as a list of numbers in a high-dimensional space.
- Extended Thinking
- A mode in the AI where it takes additional time to reason through a problem before answering.
F
- Fine-Tuning
- Additional training applied to a pre-trained AI model using a smaller, focused dataset, adapting it to a specific task, domain, or style without retraining from scratch.
- Foundation Model
- A large AI model trained at enormous scale on diverse data, used as a base for many applications.
G
- Grounding
- Connecting AI responses to specific, verifiable source material — retrieved documents, web search results, or uploaded files — rather than relying solely on the model's training.
H
- Hallucination
- When an AI confidently states something that is factually incorrect or invented.
- Hook
- An automation trigger in Claude Code that fires at a specific moment during a session — before a tool runs, after a command completes, when a session starts, when permission is requested.
I
- Inference
- Using a trained AI model to process new input and produce output.
J
- Jailbreaking
- Attempts to bypass an AI's safety guardrails through creative prompting — getting the model to say or do things it's designed to refuse.
K
- Knowledge Base Project
- The collection of documents uploaded to a Claude Project that Claude can reference during conversations.
L
- Large Language Model LLM
- A neural network trained on an enormous amount of text that learns statistical patterns about language.
M
- MCP Server
- Model Context Protocol Server.
- Memory
- Claude's cross-conversation storage of things you've told it — your role, preferences, recurring projects, and context.
- Model
- The specific AI engine in use — for example, Claude Sonnet, Claude Opus, or Claude Haiku.
- Multimodal AI
- AI that processes more than one type of input — text, images, audio, video, documents — and can produce output in multiple formats.
N
- Neural Network
- A computing architecture loosely inspired by the human brain.
P
- Permissions
- The specific access rights granted to an AI connector or tool.
- Plan Limits
- Restrictions built into AI subscriptions that control usage — how many messages per session, which models you can access, how large a file you can upload, and how much context a single conversation can hold.
- Prompt
- The input you give to an AI — your message, question, or instruction.
- Prompt Injection
- A security attack where malicious instructions are embedded in content that an AI is asked to process.
R
- RAG Retrieval Augmented Generation
- A technique that allows an AI to search a large document collection and retrieve relevant sections before responding — rather than loading everything into the context window at once.
- Research Mode
- A Claude feature that conducts multiple rounds of web research — typically 5 or more searches — to produce a comprehensive, synthesized report on a topic.
- RLHF Reinforcement Learning from Human Feedback
- The most widely used technique for aligning AI responses with human preferences.
S
- Semantic Search
- Search that finds results based on meaning rather than exact keyword matching.
- Skill
- A discrete, reusable capability within an agent.
- Skills Library
- An organized collection of skills available to an agent or team.
- SSO Single Sign-On
- An authentication method that lets users log in with their organization's identity provider (Microsoft, Google, Okta, etc.) instead of a separate password.
- Subagent
- A specialized AI worker that Claude delegates tasks to within a Claude Code session.
- Synthetic Data
- Artificially generated data that mimics real data in structure and statistical properties.
- System Prompt
- Hidden instructions given to an AI at the start of a session that define its role, behavior, and constraints — before the user sends any message.
T
- Token
- The unit used to measure AI input and output for context window and API billing purposes.
- Training Model Training
- The process by which an AI company uses conversations and interactions to improve its models.
- Transformer
- The mathematical architecture underlying nearly all modern large language models.
V
- Vector Database
- A database designed to store and search embeddings (numerical representations of meaning).
- Voice Mode
- A feature in Claude Mobile that enables fully spoken, two-way conversations — you speak, the AI listens and responds in audio.
Z
- Zero Data Retention ZDR
- A data handling arrangement where the AI provider agrees not to store any conversation content after processing.
- Zero-Shot Prompting
- Asking an AI to perform a task without providing any examples.