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

# Access Controls

> Understand how AIvis protects users, roles, groups, documents, models, Agents, and tools.

AIvis access control combines identity, roles, groups, document visibility, model availability, Agent configuration, and tool authorization. Prompt text is not an access-control mechanism.

## Identity and roles

| Control            | Purpose                                                                                                                     |
| ------------------ | --------------------------------------------------------------------------------------------------------------------------- |
| Authentication     | Confirms the user or service account identity before API access.                                                            |
| Administrator role | Grants access to deployment configuration, model credentials, connectors, user management, and security-sensitive settings. |
| User role          | Grants standard chat, search, and Agent use according to available workspace configuration.                                 |
| Service identity   | Allows internal services and automation to call approved APIs without using a human account.                                |

## Groups

Groups should represent real access boundaries, such as departments, projects, environments, or data-owner scopes. Use groups to restrict connector visibility, document sets, model availability, and Agent usage when broad access would expose sensitive content.

## Document access

| Mode                    | Behavior                                                                   | Risk to review                                                          |
| ----------------------- | -------------------------------------------------------------------------- | ----------------------------------------------------------------------- |
| Public                  | Any user in the deployment can use the indexed content.                    | Do not use for source data that is private or restricted.               |
| Private or group-scoped | Only selected users or groups can use the indexed content.                 | Test both allowed and denied users after connector setup.               |
| Source-permission-aware | Access follows supported source-system permission signals when configured. | Verify connector support and fail-closed behavior before relying on it. |

## Model and Agent access

Administrators should review which users and Agents can use high-cost, high-risk, or specialized models. Agents should receive only the tools, knowledge sources, and model options required for their purpose.

## Tool and action access

Actions can read data, call external services, or change state. Treat them as privileged capabilities:

* Keep a tool owner and change history.
* Validate tool inputs before execution.
* Deny calls outside the current user, workspace, Agent, or deployment policy.
* Review high-impact tools before enabling them for broad Agent access.

## Verification checklist

* A denied user cannot retrieve restricted connector content.
* A user cannot select a private model outside their allowed group or Agent scope.
* An Agent cannot call a tool that is not in its allowlist.
* Admin-only APIs reject non-admin users.
* Access changes are reflected in search, chat, and Agent paths.