Set Up Approval Workflows

How approval works

Approval is not a list of named approvers — it is a mode you set on each action the AI can take. Every capability has its own setting, so you can let the safe ones run unattended and hold the risky ones for a human.

The four modes are:

Mode	What happens
Disabled	The AI cannot run this action at all
Suggest only	The AI can suggest it but never execute
Requires approval	The AI queues it — a human must approve
Fully automatic	The AI executes immediately, within rate limits

Each action also carries a risk tier — read, low, medium or high — which is how you judge what deserves holding back. Rebooting a device defaults to high; clearing temp files defaults to low.

Requiring approval for an action

1. Go to Autonomy → Action Policies.
2. Find the action — they are grouped by category, such as Remediations.
3. Set its mode to Requires approval.

That takes effect immediately. Anything the AI wants to do with that action now waits instead of running.

Approving and rejecting

1. Go to Autonomy → Approvals to see everything queued.
2. Each item shows the action, the target, and why the AI proposed it.
3. Click Approve to let it run, or Reject to discard it.

Items stay queued until someone acts on them — nothing auto-rejects on a timer.

Who can do this

Both screens are restricted to admin and super admin roles. There is no way to nominate an approver who is not already an administrator, so approval authority follows the platform role rather than a separate list.

Every decision is recorded in the audit trail under Settings → Audit Trail.

Where to go next

• Understand the whole module → How AI Autonomy Works
• Next in this module → Configure Autonomy Policies
• How AI Triage Works
• Configure the AI Policy Engine
• See where this fits in the bigger picture → Start Here: What Plenix Is and How It Fits Together