AI glossary for beginners

17 terms

Agent

A preset AI assistant configured for a particular kind of task. In Verics you can choose an agent when one matches your work better than a general conversation; which agents are available depends on your setup.

API

Short for Application Programming Interface — a way for software to talk to other software. In the context of AI, it is how an application connects to a model provider behind the scenes.

API key

A secret credential that authorises access to a service, such as an AI provider. Depending on your setup, you may add your own provider keys in the Verics settings.

Related: API

Context window

The amount of text a model can consider at once, including your prompt and the recent conversation. When a conversation gets very long, older parts may no longer fit in the window.

ExampleIf a chat runs for a long time, the model may lose track of something said much earlier.

Related: Token, Prompt

Embedding

A numeric representation of text that captures its meaning, used to compare and search text by similarity. It works behind the scenes in features such as searching a knowledge base.

Related: Knowledge base

Fine-tuning

A general AI technique for further training a model on additional data so it specialises in a particular style or domain. It is an industry term rather than a step you need to perform yourself to use an AI assistant.

Hallucination

When an AI produces information that sounds confident but is inaccurate or made up. It is the main reason to check important facts against a reliable source.

ExampleAn AI might invent a plausible-looking source or statistic that does not actually exist.

Knowledge base

Your own saved material that an AI can use as context for its answers. In Verics you can let a conversation draw on your knowledge base when the task relies on your documents.

Related: Embedding, Context window

Model

The underlying AI system that generates responses. Different models vary in speed, depth and capabilities. Verics labels models with plain-language categories such as Fast, Balanced and Deep analysis.

Related: Reasoning model, Multimodal

Multimodal

A model that can work with more than just text — for example images as well as words. Whether a given model is multimodal depends on the model.

Prompt

The instruction or message you give an AI describing what you want. Clearer, more specific prompts generally produce better results.

Example“Write a friendly 80-word email announcing our new Sunday opening hours.”

Related: System prompt

Reasoning model

A model suited to working through problems in careful, step-by-step fashion. Some models expose a reasoning-effort setting so you can trade speed for depth.

Related: Model

System prompt

Background instructions that shape how an assistant behaves across a conversation — its role, tone and rules — separate from your individual messages.

Related: Prompt

Temperature

A setting that influences how varied or predictable a model's output is. Lower values tend to be more focused and consistent; higher values more varied. It may appear among advanced parameters depending on your configuration.

Token

A small chunk of text — often a word or part of a word — that models use to measure and process language. Length limits and usage are commonly counted in tokens.

ExampleA short sentence might be a couple of dozen tokens.

Related: Context window

Tool

An extra capability an AI can use beyond writing text, such as searching the web or reading an attached file, to complete a task.

Related: Web search

Web search

A capability that lets an AI look up information from the web for its answer. Useful for current or external information — though you should still check the sources for anything important.

Related: Tool, Hallucination