Once an integration is enabled and your account is connected, LuumenAI can use it in any conversation. There is nothing to attach or select: mention the system, or ask for something that lives in it, and LuumenAI reaches for the right tool. Owners and Admins can set a permission on each tool, which determines whether LuumenAI runs it automatically, asks for approval first, or cannot use it at all.

For more information see integration governance or setting tool permissions.

Ask in plain language: "Pull up the ServiceNow incident for the payments API outage and summarize what has been tried", or "Check this host for pending security updates, then open a change ticket describing them". LuumenAI decides which tools it needs. Reads run as soon as your policy allows. Anything that changes the external system pauses for approval.

Every tool call is visible in the conversation as it starts, runs, and returns, for example "searching ServiceNow" with its result attached when it completes. Failed calls show the provider's error so you can tell a permission problem from an outage.

When LuumenAI reaches a tool set to Ask for approval, the conversation shows an approval card and waits. The card shows:

You answer with Approve or Reject, which declines and hands the conversation back to you to redirect. While a decision is in flight the card reads Approving or Rejecting, then Approved or Rejected. If you step away, Luumen re-verifies your authorization when you return and submits the decision rather than making you start over. An approval nobody answers expires cleanly rather than leaving the run hanging.

There is no "always allow", every governed call is its own decision, and an approved action executes exactly once even if the connection retries.

Every read and write goes to the external system under your connected account. A ticket LuumenAI creates on your behalf is created by you, with your name on it, and the system's own audit trail says so. If you have not connected an account to an integration, LuumenAI cannot use it for you, even if a teammate has connected theirs.

For more information on connecting your account to an integration.

A Skill can also use integrations: a pre-patch Skill might check the change calendar before it starts, or a triage Skill might open a ticket when it finds a problem. The integration's tool permissions apply inside the Skill exactly as they do in a conversation, so a gated write still produces an approval prompt at that step.

On Enterprise, every tool call, approval, and outcome goes to the audit log with your identity, the tool, the risk level, and the outcome. Credential values never appear in a conversation or in the log. Open Audit in the left navigation to read it.

For more information please see audit log.

If LuumenAI needs a tool that is Not enabled, or an integration your organization has not switched on, it tells you rather than working around it. Ask an admin to change the permission, or request the integration from the directory.

For more information: request an integration