IntelXview AI Control Plane
Give AI agents the freedom to act. Keep control of what they do.
AI Control Plane checks each action an agent sends through it before it runs, and records the decision.
An initial conversation about one workflow, to see whether ACP could help. No charge, no pressure. Booking or attending the call does not commit you to buying anything; any paid work needs a separate written scope and price that you choose to accept. To change or cancel, reply to your calendar invitation.
AI Control Plane checks each action an agent sends through it before it runs, and records the decision.
On the agent actions you route through ACP:
Autonomy
Put more AI agents to work.
Prevention
Stop unauthorised actions before they happen.
Revocation
Withdraw an agent's authority when needed.
A paid pilot can test whether withdrawing an agent's authority prevents its subsequent actions on the workflow connected to ACP.
Accountability
Know who authorised what, and keep the evidence.
Actions that bypass ACP are outside these controls.
A proof run, with its limits visible
ACP public evidence screen · 4 October 2026
In plain terms: an automated test run of ACP's controls covered all 8 critical steps and passed, on 4 October 2026. It shows the product working, not the outcome of your workflow.
What holds agents back
Agents stay in pilots when nobody can say what they may do.
Before an agent takes actions that matter, someone has to answer three things: what it may do, who approves the risky actions, and what record exists when someone asks later.
The examples here are themes we heard in recent conversations with practitioners and delivery partners. They are not survey results.
Access that outlives a role
Someone changes role, but the agent acting for them keeps the old permissions.
Approvals that turn into a queue
Sensitive actions wait for sign-off, and people find a way round the approval.
Rules nobody can explain later
After a handover, no one can say why the agent may do what it does.
How ACP does it
Decide, approve and prove, for each agent action.
Decide
Allowed or refused, before it runs
You set what each agent may do. Requests outside that permission are refused on routes connected to ACP.
Approve
A named approver for chosen changes
Changes you choose can be held until a named approver decides, and the decision is recorded.
Prove
A record of every decision
Each allow, refusal and approval is recorded so it can be reviewed later.
Scope. ACP controls only the actions sent through it. Each tool or environment needs an integration. Anything that bypasses ACP is outside these controls.
See how it worksOne example
One action, three possible outcomes.
- An agent asks to send a payment to a supplier.
- ACP checks the request against the permission you set.
- Within the permission, it is allowed. Outside it, it is refused. Above an approval threshold, it waits for a named person.
- Each outcome is written to a decision record.
See a published record
From a demonstration tenant with synthetic data. It is not a customer result.
Scope. This applies only to actions routed through ACP. Actions that bypass ACP are outside these controls.
1 · Agent request
“Send a payment to a supplier”
2 · ACP check
Against the permission you set
3 · One of three outcomes
Allowed
Within the permission
Refused
Outside the permission
Needs human approval
Held until a named person decides
4 · Decision record
Each outcome is recorded for review
Where ACP fits
Workflows where an agent takes a consequential action.
For example, an agent that sends a payment, changes a production schedule or updates a patient record. What matters is the action, someone who owns it, and a route we can send through ACP.
These are examples of actions, not customer deployments.
What the first call covers
- 1
We talk through one workflow, its owner and the result you want. No charge.
- 2
If ACP looks useful, we send a written scope for a paid pilot: the actions, tests, acceptance criteria, dependencies, deliverables and price.
- 3
You decide. A paid pilot starts only after you accept that written scope and price.
Before a paid pilot. A paid pilot may start with a controlled demonstration on synthetic data. Installing ACP in your environment, cloud permissions and integration are scoped and priced as separate stages. A demonstration on synthetic data does not show that a live workflow is protected.
Talk through one workflow.
Bring a short description of the workflow and who owns it.
An initial conversation about one workflow, to see whether ACP could help. No charge, no pressure. Booking or attending the call does not commit you to buying anything; any paid work needs a separate written scope and price that you choose to accept. To change or cancel, reply to your calendar invitation.