AI glossary: terms explained simply
52 AI terms, from tokens and context windows to RAG, agents and MCP, explained in one or two plain sentences.
52 terms
- Agent
- A system where an AI model takes actions toward a goal — using tools, memory and a loop of plan, act and check — rather than only answering.
- Agentic workflow
- A process where AI handles some decisions between fixed steps, such as classifying a lead before an automation routes it.
- AI Inbox
- A Gmail feature for paid Google Workspace users that surfaces suggested to-dos and summaries of key topics.
- API
- A way for software to talk to an AI model directly, usually billed per token rather than by subscription.
- Artifact
- In Claude, a standalone piece of output — a document, app, diagram or code — shown alongside the conversation.
- Audio upload
- Attaching a recording to an assistant so it can transcribe, summarize and answer questions about it.
- Benchmark
- A standard test used to compare models. Useful as a signal, but real tasks often rank models differently.
- Chain-of-thought
- The step-by-step reasoning a model works through before answering. Reasoning models do much of this internally.
- Context window
- The maximum amount of text (in tokens) a model can consider at once, including your prompt, files, history and its reply.
- Custom instructions
- Standing preferences an assistant applies to every new chat — who you are and how you like answers.
- Deep Research
- An agent mode that browses many sources over several minutes and returns a cited report.
- Deep Research agent
- A research mode that runs many searches and returns a long report with citations.
- Dot
- An always-on ChatGPT agent with its own cloud computer that keeps working on a goal between conversations.
- Embedding
- A list of numbers representing the meaning of text, so similar ideas sit close together. Used for search and RAG.
- Few-shot prompting
- Including a few examples in your prompt so the model copies the pattern.
- Fine-tuning
- Further training a model on examples so it learns a specific style, format or narrow task.
- Frontier model
- One of the most capable models available at a given time, usually from a major AI lab.
- Gem
- A custom version of Gemini with saved instructions. Google is replacing Gems with Skills.
- Grounding
- Connecting a model’s answer to specific sources — documents or search results — so claims can be checked.
- Guardrails
- Rules and checks that limit what an AI system can say or do, such as requiring approval before sending email.
- Hallucination
- A confident answer that is false or unsupported — invented facts, quotes or citations.
- Image editing (multi-turn)
- Changing an AI image step by step while keeping the subject consistent between edits.
- Inference
- Running a trained model to produce an answer. Training builds the model; inference uses it.
- Intelligent UI
- ChatGPT’s October 2026 feature that mixes text with diagrams, comparisons and interactive tools in answers.
- Knowledge cutoff
- The date after which a model’s training data stops, so it may not know about later events without search.
- Large language model (LLM)
- An AI model trained on large amounts of text to predict and generate language.
- Latency
- How long a model takes to start and finish responding.
- MCP (Model Context Protocol)
- An open standard that lets AI assistants connect to tools and data sources in a consistent way.
- Memory
- Information an assistant keeps between conversations, such as your preferences or ongoing projects.
- Multimodal
- Able to work with more than one type of input or output — text, images, audio, video.
- n8n
- A workflow automation platform (self-hosted or cloud) where you connect triggers, apps and AI steps on a visual canvas.
- Open-weight model
- A model whose trained weights are published so others can run and adapt it.
- Parameters
- The internal numbers a model learns during training. More parameters often, but not always, means more capability.
- Project
- A workspace in ChatGPT or Claude that keeps chats, files and instructions together for one piece of work.
- Prompt
- The instructions and context you give a model.
- Prompt chaining
- Breaking a task into several prompts where each output feeds the next.
- Prompt injection
- Hidden or malicious instructions inside content (a web page, email or file) that try to make an AI do something unintended.
- RAG (retrieval-augmented generation)
- Finding relevant passages from your documents and giving them to the model with the question so answers are grounded.
- Rate limit
- A cap on how many requests or messages you can send in a period of time.
- Reasoning model
- A model that spends extra computation thinking through a problem before answering, trading speed for accuracy.
- Role prompting
- Asking the model to adopt a specific expertise and standards, such as “a senior editor”.
- Skill
- In Gemini, reusable custom instructions for a specific job — the replacement for Gems.
- Structured output
- Asking a model to reply in a fixed format such as JSON or a table so software can use the result.
- System prompt
- Instructions set by an app or developer that shape a model’s behavior before your message.
- Temperature
- A setting that controls randomness: lower is more focused and repeatable, higher is more varied.
- Token
- A chunk of text a model reads and writes — roughly four characters of English on average.
- Tool use
- A model calling external functions — search, a calculator, a calendar — to complete a task.
- Training data
- The text and other data a model learned from. Models may not know about events after it was collected.
- Transcription
- Turning spoken audio into written text.
- Vector database
- A database that stores embeddings so it can find passages similar in meaning to a question.
- Voice mode
- Talking to an assistant out loud and hearing spoken replies in real time.
- Workflow automation
- Software that runs a sequence of steps automatically when a trigger happens.