AI Glossary: Plain English Definitions
Updated July 23, 2026 · Data as of 2026-07-09
Most AI terminology exists to describe how a product is built, not what it does for you. These definitions go the other way. Each one explains a term in plain language and then says what it should change about the tool or plan you pick. Start anywhere, or take the quiz if you would rather skip the vocabulary and get a recommendation.
Agentic coding
Agentic coding is when an AI tool takes a goal, plans the steps itself, and then acts on your project across several turns, reading files, editing them, running commands, and checking its own results, instead of returning a single block of suggested code.
API Key
An API key is a unique credential — a long string of characters — that identifies your account when you send requests to an AI provider like Anthropic, OpenAI, or Google.
Context window
A context window is the total amount of text a model can consider in a single request, measured in tokens. It covers your instructions, the files you attach, the conversation so far, and the response the model is about to write.
Credits
Credits are a vendor's own unit of account, sold with a subscription and spent as you use the product. One credit is not a fixed amount of work; what it buys depends on which model you invoke and how much the tool does on your behalf.
Fine-tuning
Fine-tuning is additional training applied to an existing model using your own examples, which permanently adjusts the model's behavior rather than adding information to a single request.
Hallucination
A hallucination is output that is fluent, confident, and wrong: an invented citation, a function that does not exist, a plausible statistic with no source. The model is not lying, it is producing likely text, and likely is not the same as true.
MCP (Model Context Protocol)
MCP, the Model Context Protocol, is an open standard for connecting AI assistants to outside systems such as databases, file stores, ticket trackers, and internal APIs, so that any compliant assistant can use any compliant connector.
Open-weight model
An open-weight model is one whose trained parameters are published for download, so anyone can run it on their own hardware or through a provider of their choosing, rather than only through the original vendor's service.
Prompt caching
Prompt caching is a provider feature that stores the processed form of a repeated section of your request, usually a long instruction block or an attached document, so that later requests reusing that section are cheaper and faster.
RAG (retrieval augmented generation)
Retrieval augmented generation is a pattern where the system searches your documents for passages relevant to a question, puts those passages into the request, and asks the model to answer using them. The model is unchanged; the input is enriched.
Rate limit
A rate limit is a cap on how much you can use a service within a window of time, expressed as requests per minute, tokens per minute, or messages per hour. Hitting one pauses you; it does not usually cost you anything.
Reasoning model
A reasoning model is one trained to work through a problem step by step before producing an answer, spending additional computation at the moment you ask rather than answering immediately.
Temperature
Temperature is a setting that controls how much randomness enters the model's word-by-word choices. Low values make it pick the most likely continuation almost every time; high values let it choose less likely ones more often.
Token
A token is a chunk of text — roughly 3 to 4 characters, or about 0.75 words. AI providers measure usage in tokens to calculate how much processing a request requires.
Know the words and still not sure what to buy? Take the quiz.
Take the quiz →Missing a term? Tell us and we will add it. See also how we score tools.