# AI glossary

> Plain-English definitions of the vocabulary teams hit when they start building with language models: agents, harnesses, context windows, memory, tool calling and retrieval.

Edited by Derek Wonk. Last updated 2026-08-19. Every definition cites primary sources (model provider docs, framework repositories, published specifications) on the term's own page.

## Terms

- [agent](https://ai.developertools.io/agent/): A model running in a loop: choose an action, see the result, decide again, until a stopping rule fires.
- [context window](https://ai.developertools.io/context-window/): The per-request token ceiling that the prompt, the history, the tool definitions and the reply all share.
- [embedding](https://ai.developertools.io/embedding/): A fixed-length vector that stands in for text, arranged so that similar meanings sit close together.
- [hallucination](https://ai.developertools.io/hallucination/): Fluent, confident output that the input does not support, delivered in the same tone as a correct answer.
- [harness](https://ai.developertools.io/harness/): The program around the model: it owns the loop, the tools and the state, and resends everything each call.
- [memory](https://ai.developertools.io/memory/): State kept outside the request so a later one can use it. The model has none; the harness has all of it.
- [model](https://ai.developertools.io/model/): A frozen set of weights plus its tokenizer and serving config, selected by the name string in your request.
- [prompt](https://ai.developertools.io/prompt/): Everything the model receives in one request: instructions, history, retrieved text and tool definitions together.
- [RAG](https://ai.developertools.io/rag/): Fetch relevant text at request time and put it in the prompt, so the answer is grounded in your own material.
- [reasoning](https://ai.developertools.io/reasoning/): The model writes intermediate steps before answering. They are tokens: often hidden, always billed, always occupying context.
- [session](https://ai.developertools.io/session/): A fiction your code maintains: the history, budgets and tool state you reassemble and resend on every request.
- [token](https://ai.developertools.io/token/): The unit a model reads and writes: a fragment of text. Limits and prices are counted in these, not in words.
- [tool calling](https://ai.developertools.io/tool-calling/): The model returns a structured request to run one of your functions. Your code runs it and sends the result back.

## Browse

- [All terms](https://ai.developertools.io/): the full index, grouped by first letter.
- [Terms starting with A](https://ai.developertools.io/letter-a/): 1 term.
- [Terms starting with C](https://ai.developertools.io/letter-c/): 1 term.
- [Terms starting with E](https://ai.developertools.io/letter-e/): 1 term.
- [Terms starting with H](https://ai.developertools.io/letter-h/): 2 terms.
- [Terms starting with M](https://ai.developertools.io/letter-m/): 2 terms.
- [Terms starting with P](https://ai.developertools.io/letter-p/): 1 term.
- [Terms starting with R](https://ai.developertools.io/letter-r/): 2 terms.
- [Terms starting with S](https://ai.developertools.io/letter-s/): 1 term.
- [Terms starting with T](https://ai.developertools.io/letter-t/): 2 terms.
- [About](https://ai.developertools.io/about/): Who writes this AI glossary, what counts as a source, how often entries are re-checked, and how to report something that has drifted or is simply wrong.