Brain AIX
Connections

Global permissions and digital-human access

Choose who can access a connection and set one shared policy for each tool.

Connecting an app, MCP server, or HTTP group does not automatically grant access to a digital human. Brain separates two decisions:

  1. Access: which digital humans are assigned to the connection.
  2. Tool policy: whether a tool is allowed, requires approval, or is blocked. This policy is global for your account.

Access per digital human

In the Connected digital humans section, each digital human has a toggle: turn it on to grant access to all of that connection's tools at once, turn it off to revoke it. You can search by name and page through the list if you have many digital humans.

Assignment is individual: granting access to one digital human does not grant it to the others.

Global policy for each tool

In Tool permissions, each tool (grouped into read-only and write) has one of these policies — set individually or in bulk per section:

PolicyWhat it does
Allow alwaysThe digital human uses it without asking you anything.
Require approvalBefore running it, the digital human shows you a card to approve or decline the action.
BlockThe digital human can't use it, even if it has access to the connection.

By default, read-only tools are set to allow and write tools require approval — a sensible starting point you can adjust at any time.

[!IMPORTANT] A policy cannot vary by digital human. Changing it updates every digital human that uses the tool, and future assignments inherit it. A bulk change is stored as one operation: either every selected policy is applied or none is.

Change permissions with the assistant

The Tool Studio Assistant can propose a policy change. Its review card states that the effect is global and requires explicit approval before anything is saved.

When a tool is missing mid-chat

If a digital human needs a tool it doesn't have while you're chatting, Brain surfaces it right there: it offers to connect it (if no one has connected it yet) or grant access (if it's already connected but not assigned to that digital human) — without leaving the conversation.

When you add another digital human, you still need to grant it access to the connection. You do not need to configure every tool policy again: it uses the global policy you already chose.