# How Aviato fits together (https://docs.getaviato.com/concepts)



| Concept        | What it means                                                                   |
| -------------- | ------------------------------------------------------------------------------- |
| Workspace      | Your team, membership, subscription, and shared administration.                 |
| Project        | One connected application and its configuration.                                |
| Environment    | An isolated connection within a project, such as production or staging.         |
| Database agent | The service that runs next to your database and enforces data access.           |
| Collection     | A database table, MongoDB collection, or custom collection exposed by a plugin. |
| Record         | An item in a collection, identified by its primary key.                         |
| Action         | A named operation defined by your development team.                             |
| Plugin         | Your application code extending the agent through a language SDK.               |
| MCP client     | An AI application connecting to the agent through Model Context Protocol.       |

## The agent and the control plane [#the-agent-and-the-control-plane]

The **database agent** connects to your data sources. Your browser and authorized MCP clients request records from that agent. It applies permissions, runs operations, and queues audit events.

The **control plane** manages sign-in, workspaces, roles, agent configuration, approvals, and audit services. It does not connect directly to a self-hosted database. Some features send selected data through the control plane; see [data flow](/administration/data-flow).

## Two kinds of agent [#two-kinds-of-agent]

Aviato's database agent is infrastructure you deploy. An AI agent is a client, such as an assistant using MCP, which sends requests to it. Connecting an AI agent does not grant unrestricted database access.

## Environments are separate [#environments-are-separate]

A record ID is only meaningful within its collection and environment. A customer numbered `42` in staging is not the same record as customer `42` in production. Check the environment selector before editing, running actions, or reviewing an approval.

## Permissions follow the request [#permissions-follow-the-request]

Workspace roles provide baseline access. Custom project roles can narrow it. MCP scopes further constrain what an AI client can do. A plugin using the agent's Data API operates under the caller's permissions.

## From a collection to an audited change [#from-a-collection-to-an-audited-change]

The [record guide](/guides/records) walks through a real connected Customers collection. A teammate opens a row, saves an edit through the agent, and checks the resulting event in [record activity and the audit log](/administration/audit). The screenshots use a disposable database with fictional example data.