Every change is a proposed action
A budget shift, a paused ad, a new campaign, a contact enrolled in a sequence: each one starts as a proposed action with a reason, the evidence behind it, the expected value, a confidence score and a risk level. Nothing in Demander changes an ad account, a mailbox or a social profile any other way.
The gate reads that proposal and returns one of three verdicts. Auto-execute means it is inside what you delegated. Needs approval puts it in your inbox. Blocked means it breaks a hard limit and will not run, whoever approves it. The reasons come back with every verdict, including the automatic ones, so the audit log shows why a machine was allowed to act.
What each autonomy level runs on its own
You set the level per workspace. A proposer cannot claim a lower level than its action type and risk require; the gate works that out itself.
| Level | Runs without you | Still asks you |
|---|---|---|
| L0 | Nothing. Recommendations only. | Everything |
| L1 | Nothing. Changes are prepared for one-click approval. | Everything |
| L2 | Low-risk, reversible optimisation: pausing an ad, shifting budget within the total, bid changes, persona weights, account-list updates | New budgets, new creatives, launches, enrolments |
| L3 | Also setting budgets and regenerating creatives that have worn out | Launches, enrolments, social publishing |
| L4 | Also launching campaigns and enrolling contacts in sequences | Irreversible actions |
| L5 | Everything your policies allow, end to end | Whatever your policies reserve |
Plans set the highest level available: Starter L1, Growth L2, Scale L3, Enterprise up to L5 with custom policies. See pricing.
The hard limits, at every level
These are checked before anything else. Approving a blocked action does not get it past them.
-
01
Kill switch
One switch for an emergency. It pauses your running campaigns on the ad platforms at once and holds outreach e-mails, scheduled posts and broadcasts until you release it; new chat messages go to a person instead of the bot. Launches and budget increases are blocked; pauses and budget cuts still go through.
-
02
Daily and monthly caps
Spend caps per workspace. Recurring daily cost is projected to the end of the month, and today’s spend counts the committed run rate of active campaigns, not only what has been billed so far.
-
03
Change limit per day
A budget that would move by more than your daily change limit (20 percent by default) goes to approval instead of running.
-
04
No spend increase on broken tracking
If a platform has an open critical tracking failure, no action that raises spend on it can run. Warnings are shown but do not stop spend.
-
05
Country allow-list
Targeting outside the countries you allow is blocked, for ads and for geofenced campaigns alike.
-
06
Containment after anomalies
A critical anomaly, such as overspend against the plan, delivery that has stopped or clicks with no conversions recorded, switches on containment: nothing that raises spend auto-executes until it clears.
Undo, one change or everything since a point in time
-
A snapshot before every change
Before an executor touches anything it stores the entity as it was and a revert payload. Budgets, statuses, bids and audiences can all be put back.
-
Reverts pass the gate too
Undoing a pause or a budget cut raises spend, so the revert has to clear the kill switch, the caps and tracking health first.
-
Undo everything since X
If an agent or a teammate made a run of changes you do not want, revert everything they did in the last hours in one step.
Approvals that cannot run twice
Each action moves through fixed states (proposed, approved, executing, executed) with compare-and-set transitions, so a double click or a network retry cannot execute it twice. Before a human approval runs, the action is re-priced against the current state: a budget set to €400 is re-costed from today’s live budget, not the one from when it was proposed.
Proposals expire after 72 hours. Blocked actions are looked at again after 24 hours, in case the condition that blocked them has cleared.
Questions people ask
Can I let Demander change budgets but not launch campaigns?
Yes. Level 3 sets budgets and publishes creatives on its own while launches, enrolments and social posts still wait for approval. You can also list specific action types that always need approval, whatever the level.
What does an AI agent connected through MCP get to do?
Exactly what the gate allows. An agent proposes and approves through the same API as the dashboard, so at autonomy level 1 it still waits for a person. See API and MCP.
Does the kill switch pause my campaigns?
Yes. Turning it on pauses every running campaign Demander manages, right away, and tells you which ones it paused. Releasing it does not start them again: you resume them in one step, and that resume passes the spend caps and the tracking check like any increase, or you keep them paused. Campaigns on an ad platform your Demander server does not run live are named so you can pause them there.
Where do I see what was blocked and why?
In the action inbox and the audit log. Each action shows its reason, evidence, the gate verdict with every rule that applied, and the execution result.