> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.agenttrunk.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.agenttrunk.ai/_mcp/server.

# Organizations and workspaces

> Separate your team's access boundary from each agent's context.

An organization is your team. A workspace is context for a use case—not an agent
identity or a running process. Several authorized agents can use one workspace.

For Acme, start with a **Customer support** workspace. Add **Research** when its
instructions, owners, or release cadence differ. You do not need a workspace per
conversation or per model.

| Object       | Example               | Purpose                                        |
| ------------ | --------------------- | ---------------------------------------------- |
| Organization | Acme                  | Team membership and access boundary            |
| Workspace    | Customer support      | Context for one use case                       |
| Scope        | General               | Access and release boundary inside a workspace |
| Context      | refund-policy         | Named package of files                         |
| Revision     | Returned immutable ID | Exact package version                          |

## Choose before writing

```sh
agenttrunk workspaces
agenttrunk scopes --workspace WORKSPACE_ID
```

Select IDs deliberately. The API calls a workspace a `trunk`; `trunkId` and
`WORKSPACE_ID` identify the same object. Use the General scope unless you need
different permissions or independent releases.

To create a workspace when your task requires one:

```sh
agenttrunk workspace-create --name "Customer support"
```

It creates a General scope and staging/production environments. An empty list
is not the same as an authorization error. Never create a replacement organization
to get around denied access.

Continue with [uploads](/publishing) or [permissions](/concepts/permissions).