AI glossary
Short, plain-English definitions of the words that come up when you use or compare AI tools. Each one links to a guide that goes deeper.
- Agent (AI agent)
An AI system that takes actions toward a goal instead of only answering: it can plan steps, use tools such as a browser, files or apps, check the results and keep going until the task is done or it needs you.
- API
An application programming interface: the way software talks to an AI model directly. Developers pay per token through an API instead of a monthly chat subscription.
- Automation (AI workflow)
A chain of steps that runs on its own, such as "when an email arrives, summarize it with AI and post it to Slack". Tools like Zapier, Make and n8n connect apps and add AI steps in between.
- Chatbot
An app you talk to in plain language, such as ChatGPT, Claude or Gemini. Behind the chat window is a large language model.
- Connector
A link that lets an AI assistant read from or act in another service, such as Google Drive, Gmail, Slack or GitHub, so it can work with your own information.
- Context window
How much text a model can consider at once, counted in tokens. It covers your messages, any files you attach and the model's own replies. Past the limit, older material is dropped or summarized.
- Custom GPT
A version of ChatGPT set up with its own instructions, files and tools for one job, which you can keep private or share.
- Deep research
A chatbot mode that spends several minutes searching many web pages and then writes a longer report with sources, instead of answering straight away.
- Diffusion model
The kind of model behind most AI image and video generators. It starts from random noise and removes it step by step until a picture that matches your prompt appears.
- Embedding
A list of numbers that represents the meaning of a piece of text, so software can find passages that mean similar things even when the words differ. Embeddings power semantic search and RAG.
- Fine-tuning
Training an existing model a little more on your own examples so it picks up a style, format or specialty. It changes the model itself, unlike a prompt or a custom GPT.
- GGUF
A file format for language models used by llama.cpp, Ollama and LM Studio. One GGUF file holds a whole model, usually already quantized to fit in less memory.
- GPU
A graphics processor. Its many small cores do the math AI models need much faster than a normal processor, so the amount and speed of GPU memory decides which models a computer can run well.
- Hallucination
When an AI states something false with confidence, such as a made-up quote, statistic or source. Asking for sources and checking them is the main defense.
- Inference
Running a trained model to get an answer. Each reply you get is inference; training is the separate, far costlier step that creates the model.
- Large language model (LLM)
A model trained on huge amounts of text to predict the next word, which lets it write, summarize, translate, answer questions and write code. GPT, Claude, Gemini, Llama and Qwen are LLM families.
- Local AI
Running a model on your own computer instead of a company's servers. It works offline and keeps your data on your machine, but you need enough memory and the best models are usually cloud-only.
- MCP (Model Context Protocol)
An open standard for connecting AI assistants to tools and data sources. A service publishes one MCP server and any assistant that supports MCP can use it.
- Multimodal
A model that handles more than text, such as images, audio, video or files, as input, output or both.
- Open weights
A model whose trained parameters are published so anyone can download and run it, such as Llama, Qwen or DeepSeek. The license may still limit how you use it.
- Parameters
The numbers inside a model that were set during training. Counts like 8B or 70B mean billions of parameters; more usually means more capable but needs more memory.
- Plugin
An add-on that gives an AI assistant new abilities, often bundling instructions, tools and connectors together.
- Prompt
What you type to an AI: the question, instructions, examples and any context. Clearer prompts with the goal, audience and format spelled out get better answers.
- Prompt injection
Hidden or sneaky instructions in a web page, email or file that try to make an AI assistant ignore its user and do something else. It is the main security risk for AI agents that browse or read your inbox.
- Quantization
Shrinking a model by storing its numbers with fewer bits, for example 4-bit instead of 16-bit. The model needs far less memory and runs faster, with a small loss in quality.
- RAG (retrieval-augmented generation)
Looking up relevant passages from your own documents first and giving them to the model with your question, so it answers from your material instead of memory alone.
- Reasoning model
A model that works through a problem step by step before answering. It is slower and costs more per answer but does better on math, code and multi-step questions.
- Skill
A packaged set of instructions, and sometimes scripts or files, that teaches an AI assistant how to do one kind of task the same way each time.
- Speech-to-text (transcription)
AI that turns spoken audio into written text. Meeting note-takers and podcast editors are built on it.
- System prompt
Standing instructions given to a model before the conversation starts, which set its role, tone and rules. Custom instructions and custom GPTs are a kind of system prompt.
- Temperature
A setting that controls how random a model's word choices are. Low values give steadier, more predictable answers; high values give more varied, creative ones.
- Text-to-speech (TTS)
AI that reads text aloud in a synthetic voice. Modern TTS tools such as ElevenLabs and Murf can sound close to a human and clone a voice from a sample.
- Text-to-video
AI that makes video from a script or prompt, either by generating footage, assembling stock clips with a voiceover, or animating a presenter avatar.
- Token
The chunk of text a model reads and writes, often a short word or part of a word. In English, 1,000 tokens is roughly 750 words. API prices and context windows are counted in tokens.
- Unified memory
Memory shared by a computer's processor and graphics chip, as on Apple silicon. It lets a large model use most of the system's memory, which is why such machines can run big models locally.
- Vibe coding
Building software by describing what you want in plain language and letting an AI write the code, with little or no hand-written code. AI app builders like Lovable, Bolt and v0 are made for it.
- Voice cloning
Making a synthetic voice that sounds like a specific person from recordings of them. Reputable tools require consent from the voice's owner.