Glossary

Plain-English definitions of the terms used across the AI Tools Academy course. Each term has its own anchor, so lessons can link straight to it.

The terms

Action

An action is a single step that an automated workflow carries out, such as sending an email, creating a calendar entry, or updating a spreadsheet row. Actions usually run in response to a trigger. In automation tools you build a flow by chaining a trigger to one or more actions.

Agent

An agent is an AI system that can take a goal and work towards it over several steps, deciding what to do next and using tools along the way, rather than just answering a single question. For example, an agent might read a document, search the web, and then draft a summary without being prompted at each stage. Agents still work within limits you set, and it is sensible to review what they do.

Agentic

"Agentic" describes software that behaves like an agent: it plans, takes actions, and adapts based on what happens, instead of simply responding once. A feature described as agentic can carry out multi-step tasks on your behalf. The term is a spectrum, so some tools are more agentic than others.

Artifact

An artifact is a self-contained piece of work that an AI tool produces in a separate panel beside the chat, such as a document, a table, or a small app. Keeping it separate makes it easier to read, edit, and reuse than text buried in a conversation. "Artifact" is the term used in Claude; other tools have similar features under different names.

Canvas

A canvas is a dedicated editing space, alongside the chat, where you and the AI can work on a document or piece of code together and revise it in place. Instead of regenerating everything, you can point at a section and ask for changes. Several tools offer a canvas-style view under names such as Canvas.

Connector

A connector is a piece of setup that links an AI tool to another system, such as your email, calendar, or a file store, so the AI can read from or act in that system with your permission. Connectors are what let a tool go beyond the chat window and use your actual data. You should only connect tools and accounts you trust and are permitted to link.

Connector directory

A connector directory is a catalogue within an AI tool that lists the available connectors you can turn on, such as links to popular apps and services. It is where you go to see what a tool can plug into and to enable a connection. Availability varies by tool and by your organisation's settings.

Context window

The context window is the amount of text an AI model can consider at once, including your prompt, any documents you have added, and the conversation so far. It is measured in tokens. If a conversation or document is larger than the context window, the model cannot see all of it at the same time, so older or overflowing parts may be left out.

Copilot

Copilot is Microsoft's brand for its AI assistants built into products such as Windows, Microsoft 365, and Edge. In a workplace setting it can help draft documents, summarise emails, and work with your files inside the Microsoft apps you already use. Exactly what it can do depends on your licence and your organisation's configuration.

Custom instructions

Custom instructions are standing preferences you set once so the AI applies them to future conversations, such as your role, your tone of voice, or how you like answers formatted. They save you repeating the same context each time. They are separate from any single prompt and can usually be edited or turned off.

Deep research

Deep research is a mode in some AI tools that carries out a longer, multi-step investigation of a question, searching many sources and pulling the findings into a structured, referenced report. It takes longer than a normal answer because it is doing more work. You should still check the sources it cites before relying on the result.

DLP

DLP stands for Data Loss Prevention: technology and rules that stop sensitive information, such as customer records or card numbers, from leaving an organisation or being pasted into places it should not go. In an AI context, DLP controls may block or warn when someone tries to share confidential data with a tool. It is one of the ways employers keep data safe.

Fine-tuning

Fine-tuning is the process of further training an existing model on extra, specific examples so it performs better on a particular task or style. It changes the model itself, unlike prompting or grounding, which leave the model as it is. Fine-tuning is more involved and is usually done by technical teams rather than everyday users.

Flow

A flow is an automated sequence that connects a trigger to one or more actions, so that when something happens, a series of steps runs on its own. For example, "when a form is submitted, save the details to a sheet and send a confirmation email" is a flow. Flows are the building blocks of automation tools.

Gem

A Gem is a customised version of Google's Gemini assistant that you set up for a specific purpose, with its own instructions and focus. For example, you might make a Gem that always writes in your team's tone or helps with a particular task. Gems are Google's equivalent of the custom assistants offered by other tools.

GPT

In everyday use, "GPT" refers to a family of large language models made by OpenAI, and a "GPT" (with a capital G) can also mean a customised version of ChatGPT that you build for a specific job. The letters stand for "generative pre-trained transformer", which describes how the model is built. Context usually makes clear whether someone means the model type or a custom assistant.

Grounding

Grounding means giving an AI model reliable source material, such as your own documents or trusted data, so its answers are based on that material rather than only on its general training. Grounding reduces the chance of made-up answers and lets the model cite where information came from. It is a key idea behind approaches like RAG.

Hallucination

A hallucination is when an AI model produces something that sounds confident and plausible but is actually wrong or invented, such as a fake statistic or a non-existent reference. It happens because the model predicts likely-sounding text rather than looking facts up. This is why you should check important claims, especially names, numbers, and quotes.

Knowledge cut-off

The knowledge cut-off is the date after which a model's training data stops, so it can't know later events without web search. Ask about something that changed after the cut-off and, from memory alone, the model may give a confident but out-of-date answer. Connecting it to live sources, or checking against a trustworthy page, is how you get current facts.

Large language model

A large language model (LLM) is an AI system trained on very large amounts of text so it can understand and generate human-like language. It works by predicting likely next words, which lets it answer questions, summarise, and draft content. Most of the chat assistants covered in this course are built on large language models.

MCP

MCP stands for Model Context Protocol: an open standard for connecting AI tools to external systems, data sources, and tools in a consistent way. It acts like a common plug, so a tool that supports MCP can work with many different services without custom code for each one. In practice, MCP is one of the technologies that powers connectors.

Memory

Memory is a feature that lets an AI tool remember useful details about you or your work across conversations, such as your preferences or ongoing projects, so you do not have to repeat them. Unlike a single chat, memory persists between sessions. You can usually review, edit, or clear what has been remembered.

Model

A model is the underlying AI system that does the actual work of understanding and generating content. Different models vary in their capabilities, speed, and cost, and a single product may let you choose between several. When people talk about "which model" they mean which specific version of the AI you are using.

Multimodal

Multimodal describes an AI model that can work with more than one type of input or output, such as text, images, audio, and sometimes video, rather than text alone. For example, a multimodal model could read a photo of a receipt and turn it into a table. Most leading assistants are now multimodal to some degree.

NotebookLM

NotebookLM is a Google tool that lets you upload your own sources, such as documents and notes, and then ask questions, get summaries, and generate outputs grounded in just those sources. Because it works from the material you provide, its answers stay close to your content and can cite where they came from. It is popular for research and study.

Project

A project is a workspace within an AI tool that groups related chats, files, and instructions together around a particular piece of work. Keeping everything in one project gives the AI consistent context and keeps your material organised. Different tools use the word "project" for slightly different features, but the idea is broadly the same.

Prompt

A prompt is the instruction or question you give an AI tool to tell it what you want. A clear prompt, with enough context and a sense of the format you expect, usually gets a better result. Writing good prompts is a skill you build with practice, and much of this course is about doing exactly that.

Prompt injection

Prompt injection is a type of attack where hidden or malicious instructions are placed in content the AI reads, such as a web page or document, in an attempt to make the AI do something it should not. Because the AI can struggle to tell instructions from ordinary content, this is a real security concern. It is one reason to be careful about what data and links you let a tool process.

RAG

RAG stands for Retrieval-Augmented Generation: an approach where the AI first retrieves relevant information from a trusted source and then uses it to generate its answer. This grounds responses in real material and helps reduce hallucinations. Many document-based and search-based AI features work this way behind the scenes.

Reasoning model

A reasoning model is a type of AI model designed to work through problems in a more deliberate, step-by-step way before giving an answer, which can help with harder tasks such as maths, planning, or multi-part questions. It often takes a little longer in exchange for more careful results. Not every task needs one, and for simple requests a standard model may be quicker.

Skill

A Skill is a packaged set of instructions and resources that teaches an AI tool how to do a particular task well, which it can draw on when the task comes up. Skills let you extend what a tool can do without starting from scratch each time. The exact form and name vary between tools.

System prompt

A system prompt is a behind-the-scenes instruction that sets the overall behaviour, role, and boundaries for an AI assistant, before you type anything. It shapes how the assistant responds across the whole conversation. You usually do not see it directly, but it is why a tool has a consistent style and set of rules.

Temperature

Temperature is a setting that controls how varied or predictable an AI model's output is. A lower temperature makes answers more focused and consistent, which suits factual work, while a higher temperature makes them more varied and creative. Not every tool exposes this setting, but where it exists it lets you tune the balance.

Token

A token is roughly a chunk of a word: the small unit that AI models use to read and generate text. A short word may be one token, while a longer word can be split into several. Tokens matter because context windows, limits, and sometimes pricing are measured in them rather than in words.

Trigger

A trigger is the event that starts an automated flow, such as a new email arriving, a form being submitted, or a set time of day. Once the trigger fires, the flow's actions run. Choosing the right trigger is the first step in building any automation.

UK GDPR

UK GDPR is the United Kingdom's data protection law, which sets out how organisations must handle people's personal data, including keeping it secure and using it only for legitimate purposes. It matters for AI because putting personal data into a tool counts as processing it, and must be done lawfully. When in doubt, follow your organisation's data protection guidance before sharing any personal information with an AI tool.