AI & Prompting Glossary
Clear, plain-language definitions of the AI and prompting terms used across LearnPrompt — from “prompt” and “token” to “agent”, “MCP”, and “prompt injection”. Use the search box to find a term, or link straight to any definition.
46 terms
- Agentic workflow
- A task designed to run largely on its own through an agent's plan-act-observe loop, set up with a clear goal, the tools it may use, and guardrails on what it may do.
- AI agent
- A system that takes a goal and works toward it on its own -- planning steps, using tools, and checking results in a loop -- instead of just answering a single prompt.
- API
- Application Programming Interface -- a way for software to talk to an AI model directly in code rather than through a chat window, used to build AI into apps and automate prompts.
- Chain-of-thought
- A prompting technique that asks the model to reason step by step before answering, which improves results on problems that need logic or several steps.
- Chatbot
- A conversational AI tool you use by typing messages back and forth, such as ChatGPT, Claude, or Gemini.
- Context engineering
- The practice of shaping everything a model can see when it answers -- documents, data, history, and tools -- not just the wording of the prompt.
- Context window
- The amount of text a model can hold in view at once, measured in tokens. Everything it uses to answer -- your prompt, pasted documents, and the conversation so far -- must fit inside it.
- Custom instructions
- Saved preferences that tell a chatbot how to respond to you every time, so you do not have to repeat them in each new chat.
- Delimiter
- A clear marker -- such as triple quotes or tags -- used to separate your instructions from pasted content, helping the model treat that content as data and reducing prompt injection.
- Embedding
- A numeric representation of text that captures its meaning, letting software find passages with similar meaning. The basis of retrieval and vector search.
- Few-shot prompting
- Giving the model a few examples of the input and the output you want before your real request, so it follows the pattern.
- Fine-tuning
- Further training a base model on extra examples so it specializes in a task or style. Unlike prompting, it changes the model itself, not just the input.
- Frontier model
- One of the most capable, cutting-edge AI models available at a given time.
- Function calling
- A capability that lets a model trigger a defined function or tool with structured inputs, so it can take actions or fetch data instead of only producing text.
- Generative AI
- AI that creates new content -- text, images, audio, or code -- by predicting likely outputs, rather than only classifying or retrieving existing data.
- Grounding
- Tying a model's answer to real, supplied facts (documents or data) rather than its general memory, which reduces made-up information.
- Guardrails
- Rules and limits placed around a model to keep its behavior safe and on-topic -- what it must refuse, must not reveal, or must get approval for.
- Hallucination
- When a model states something false or invented as if it were true, because it generated a plausible-sounding continuation rather than a verified fact.
- Inference
- Running a trained model to produce an output from an input. "Inference" means using a model, as opposed to training it.
- Jailbreaking
- Crafting input that gets a model to bypass its own safety rules and do something it would normally refuse.
- Knowledge cutoff
- The date after which a model has no built-in knowledge, because its training data ends there. Newer events require web search or context you provide.
- Large language model (LLM)
- An AI model trained on huge amounts of text to predict likely next words, which lets it answer questions, write, summarize, and more. The engine behind chatbots.
- Latency
- The delay between sending a prompt and getting a response back from a model.
- MCP (Model Context Protocol)
- An open standard that lets any AI app connect to external tools and data sources in a uniform way -- often called the "USB-C" of AI tools.
- Multi-agent orchestration
- A setup where several specialized agents handle different parts of a task, coordinated by an orchestrator that passes work between them.
- Multimodal
- Able to work with more than one type of input or output -- for example text together with images, audio, or video.
- Open-weight model
- A model whose trained weights are publicly released, so anyone can run, inspect, or adapt it. Often called open-source, though licenses vary.
- Parameter
- One of the many internal values a model learns during training. Models are often described by their parameter count (for example, billions), a rough proxy for capacity. (Not to be confused with a fill-in placeholder in a prompt.)
- Persona prompting
- Telling the model to take on a role or expertise ("act as a copy editor") to shape the tone and focus of its answer. Also called role prompting.
- Prompt
- The input you give an AI model -- your question, instruction, and any context -- that it responds to.
- Prompt chaining
- Breaking a task into a sequence of prompts where each step's output feeds the next, instead of asking for everything at once.
- Prompt engineering
- The skill of writing clear, well-structured prompts to get reliable, useful results from an AI model.
- Prompt injection
- When instructions hidden inside content a model is processing -- an email, a web page, a file -- get the model to act against the user's intent.
- Prompt template
- A reusable prompt with [bracketed placeholders] you fill in each time, so you get consistent results without rewriting it.
- RAG (Retrieval-Augmented Generation)
- A technique that fetches relevant passages from your own documents and adds them to the prompt, so the model answers from real sources.
- Reasoning model
- A model designed to work through problems in more steps before answering, trading some speed for stronger performance on complex tasks.
- Retrieval
- Searching a collection of documents for the passages most relevant to a question, usually to feed them into a prompt.
- Structured output
- Output produced in a specified format -- such as JSON, a table, or a bulleted list -- so it can be used reliably by a person or another program.
- System prompt
- Background instructions that set a model's role, rules, and behavior for a whole conversation, separate from each user message.
- Temperature
- A setting that controls how random a model's output is. Lower is more focused and predictable; higher is more varied and creative.
- Token
- The unit of text a model reads and generates -- roughly three-quarters of a word. Usage and context limits are measured in tokens.
- Tool use
- A model's ability to call external tools -- search, code execution, file access, app connections -- to get real results instead of guessing.
- Training data
- The large body of text and other content a model learns its patterns from during training.
- Vector database
- A store of embeddings that lets software quickly find text with similar meaning -- the lookup engine behind retrieval and RAG.
- Vibe coding
- Building software mostly by describing what you want in natural language and letting an AI generate the code, reviewing and steering rather than writing it line by line.
- Zero-shot prompting
- Asking a model to do a task with no examples, relying on what it already learned during training.
No terms match your search.