Skip to content
05

Broker

Human Approval

When policy resolves to a hold, the call pauses and a person is asked. The request shows exactly what will happen, in what environment, on whose behalf. Approve or deny, and the agent continues or re-plans.

The problem

The alternative to autonomy is not oversight, it is a human copying commands out of a chat window. Approval has to live inside the execution path or it becomes theatre.

Capabilities

How Human Approval works

  • 01

    The call waits

    A held call is suspended, not failed. The agent's run stays alive with its context intact and resumes on approval.

  • 02

    Reviewable request

    The approver sees the resolved arguments, the diff or query, the environment, the agent, the owner, and the policy that raised the hold.

  • 03

    Where work happens

    Slack, Microsoft Teams, email, or the API. Approvals route to a group, escalate on timeout, and expire rather than sitting open indefinitely.

  • 04

    Approvals are evidence

    Who approved what, when, and on which policy is written into the audit chain as part of the decision record.

In practice

What it looks like

Slack approval request
Cirvix  ·  approval required                         APP   Agent      deploy-bot  (agt_3H8N1P)  Owner      marcus@example.com  Action     k8s.apply  →  production/checkout  Policy     require-approval-prod-deploy  Waiting    00:42   ── manifest diff ─────────────────────────────   replicas:      3 → 5   image:  checkout:1.9.2 → checkout:1.9.3  ──────────────────────────────────────────────   [ Approve ]  [ Deny ]  [ Open in Cirvix ]   Expires in 15m · escalates to #platform-oncall

Limits

What this does not do

Stated plainly, because a control plane that claims no limits has either not been deployed or is not being described accurately.
  • Approval quality depends on the approver reading the request. Cirvix shows the resolved action rather than a summary precisely to make that reading possible, but it cannot make a person careful.
  • A held call consumes a slot for as long as it waits. Timeouts and escalation exist so a forgotten request fails closed rather than hanging a run indefinitely.

Get started

Let the agents do real work.

Start with the free local scanner — it needs no account and tells you what is already ungoverned on your machine. Move to a governed runtime when you are ready.

Cirvix AgentControl is in private beta. Pricing is published and plans are not yet chargeable — early access confirms your place and we will tell you before anything bills.