# Core concepts

A short glossary of the terms we use. Each concept gets a deeper page elsewhere in the docs.

## Agent

A configurable AI persona that converses with visitors. Has a prompt, a voice, optional avatar, optional slides, and optional knowledge base. One agent → one published URL → one embeddable widget.

## Slide

A single page in a fullscreen experience the agent walks visitors through. Slides have a title, freely positioned elements (text, images, videos, quizzes), an optional avatar dock configuration, and an optional script the agent reads.

## Knowledge base

A library of documents (PDF, web pages, plain text) the agent can reference during a conversation. Visitors ask questions; the agent grounds answers in the KB.

## Tool

A capability the agent can invoke during conversation — calling an API, advancing slides, sending an email, looking up a customer record. Tools are how agents take action, not just talk.

## Visitor

A person interacting with a published agent. Identified anonymously by default; can be tied to your CRM via a visitor identity hook.

## Conversation

One session between a visitor and an agent. Has transcript, recording (if enabled), tool-call history, and a summary.

## Workflow

The graph of slides, tools, and branching logic that defines an experience. The visual editor renders it as nodes and edges.

## Template

A pre-built starting point — agent + slides + prompts — you can clone and customize. See the [recipe gallery](/docs/recipes).