Find an answer

A plain-language glossary

Short explanations of terms you will encounter while using Grok.

On this pagePromptModel and modeContextTokenHallucination — an unsupported outputGrounding and citationCustom and system instructionsMemory, history and personalisationAPI, key and usageAgent, connector and automationGrok BotGradual rollout

Prompt#

Text or other input describing what the assistant should do. A useful prompt defines the result and relevant materials.

Model and mode#

A model is a trained system that processes inputs and generates outputs. A mode is a way of working within a product; it may change tools or processing. A mode name is not automatically a model name.

Context#

Information available while processing a particular task: instructions, previous messages and accessible materials. Chat history, persistent memory and current context are different concepts.

Token#

A unit into which a system divides text or another input. It is not exactly one word. Token counts can affect context limits and API billing.

Hallucination — an unsupported output#

Plausibly presented information that is incorrect or invented. Examples include a nonexistent source, wrong number or claimed content of an unread attachment.

Grounding and citation#

Grounding ties an answer to available evidence. A citation points to a source. The presence of a link alone does not prove that it supports the entire sentence.

Custom and system instructions#

Custom user instructions express your preferences. System instructions guide the service at another level. This setting does not allow you to change the service’s system rules.

Memory, history and personalisation#

History is a record of conversations. Memory is a mechanism for reusing selected information across conversations. Personalisation adapts answers to the user; its inputs depend on the product. Turning one off may not disable the others.

API, key and usage#

An API is an interface for your application to communicate with a service. An API key is a secret credential. Access and paid usage are different things; check billing in the developer chapter.

Agent, connector and automation#

An agent may carry out a task in several steps using tools. A connector provides access to another source or service. An automation runs a process according to a rule or schedule. Permissions and capabilities differ by product.

Grok Bot#

See how to choose a tool for the practical difference from chat and Projects. Then try a first task with a result you can review.

Gradual rollout#

A new feature becomes available in stages. A product announcement does not prove that a particular Slovak account, plan or app has it.

Sources and verification scope

Practical guide · Editorial update: 23 Oct 2026

    Product claims are based on publicly opened sources. We did not personally test a signed-in Slovak account. Prompts are original examples; sample outputs are illustrative.

    Search the guide

    Try “privacy”, “files” or “subscription”. Accents are optional.