How approval gates work in Kantum.

Most AI tools answer questions. Kantum does the work. Doing the work means writing to your tools: sending a reply, updating a CRM deal, booking an appointment, pushing a code fix. Anything that changes the outside world is a write. Kantum proposes changes and waits for your confirmation before it sends, creates, edits or deletes in the main workflows.

A write is proposed, not performed

When Kantum decides an action is warranted, it doesn't just do it. It prepares the action: the email draft, the CRM change, the branch and diff. It proposes it for confirmation. The proposal shows you what will happen and the context behind it, so the decision is informed rather than blind.

You decide which actions are gated. Low-stakes steps can run automatically; anything that touches money, customers, or production can require a person to approve first. The line is yours to draw.

A person approves before it lands

Kantum proposes changes and waits for your confirmation before it sends, creates, edits or deletes in the main workflows. A teammate with the right role reviews the proposal and approves or holds it. Approvals can reach the right person where they already are: a message, or a call if it's urgent.

• Kantum proposes changes and waits for your confirmation before it sends, creates, edits or deletes in the main workflows.
• You set which actions are gated and which run automatically.
• Role-based access controls who can approve what.

Agent actions are logged

Every action your AI agents take is recorded in the audit log.

Why it's built this way

Autonomy without control is a liability. Kantum's premise is governed autonomy: it can run real operations because the controls are part of the product, not an afterthought. Approval gates, an audit log, role-based access and editable memory are how you stay in control while Kantum acts on your behalf.