> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rerun.build/llms.txt
> Use this file to discover all available pages before exploring further.

# Core concepts

> Every Rerun term, defined once.

## Workspace

Your account. It holds your plan, your members, your Boxes and your agents. You can belong to several workspaces and switch between them from the pill in the top left corner of the app.

An [API key](/api/authentication) is scoped to exactly one workspace.

## Box

A private computer in the cloud, isolated from every other account, that hosts your agents. On the board it is the rectangle your agents sit on.

One Box holds many agents and gives them two things they share: a **shared folder** and a **shared database**. You can restart it or shut it down from its panel.

<Note>
  The Rerun API calls a Box a `space`, in tool names like `list_spaces` and `space_db_query`. Same thing.
</Note>

## Agent

An AI worker inside a Box. It has:

|                                       |                                                                                        |
| ------------------------------------- | -------------------------------------------------------------------------------------- |
| **A name**                            | Free text, what you see. Rename it whenever you want                                   |
| **A handle**                          | The `@mention` key other agents call it by. It never changes when you rename the agent |
| **A soul**                            | Its system prompt: who it is, what it owns, how it works                               |
| **A model**                           | The provider and model it thinks with, changeable per conversation                     |
| **Capabilities**                      | Which families of tools it may use, from files to shell to browser                     |
| **Memory, skills, files, a database** | All private to it                                                                      |

## Soul

The agent system prompt, written in markdown. It is the single most important thing about an agent. Edit it in the agent **Settings** tab, or let the agent rewrite it as it learns.

## Skill

A written procedure the agent loads on demand, when the task matches its description. It is a folder: a markdown body plus optional reference files.

Skills keep an agent sharp without bloating its prompt: only a one-line index of every skill is always in context, and the full text loads when it is needed. See [Skills](/agents/skills).

## Memory

Durable facts the agent writes down and recalls later. Memories fade if nothing uses them, and each recall extends their life, so what matters stays and what does not quietly retires.

Correct an agent once and it remembers the correction.

## Connector

A ready-made integration with an external service: Stripe, Gmail, Notion, Slack and hundreds more. Installing one on an agent gives it that service tools plus a skill explaining how to use them.

A connector is installed per agent, not per Box. See [Connectors](/agents/connectors).

## MCP server

The open standard Rerun connectors are built on. When the library does not have what you need, you can register any MCP server on an agent yourself, with its own URL, headers or command.

## Run

One execution of an agent. It starts when you send a message, when a scheduled task fires, when a webhook is called, or when another agent calls it. It ends when the agent is done, when it fails, when you cancel it, or when it pauses to ask you something.

Every run records its steps, its tool calls, its tokens and its cost. Read them in the **Monitoring** and **Logs** tabs, or with [`get_run`](/api/tools/runs#get-run).

## Session

A conversation thread with an agent. A run belongs to a session. Scheduled tasks and webhooks each open their own session, so an automated run never pollutes your chat.

## Scheduled task

A recurring instruction, on a cron and in a timezone, or a one-off at a given date. When it fires, the agent runs with no user message and follows the instruction. See [Automations](/agents/automations).

## Trigger

A public webhook URL that starts a run when it is called. The caller payload is handed to the agent alongside your instruction. The URL is the credential. See [Webhook triggers](/api/webhooks).

## Template

A packaged agent setup, published in the marketplace, that anyone can install into their own workspace in one click. It carries prompts, skills, scheduled tasks, connector declarations and files, but never memories or credentials.

Templates can be free or paid. See [Templates](/templates/install).

## Board

The isometric view that is the app itself. Boxes are drawn as rectangles, agents sit on them, and clicking an agent opens its panel.