Add a capability

Adding is always scoped to one agent, and always checked with the provider before it is charged.

The sequence

Add ─▶ availability check ─▶ payment (if priced) ─▶ delivery ─▶ visible on the agent
  1. Add. Pick the capability and the agent.
  2. Availability check. CitizenAI asks the provider whether it can fulfil this exact request, using the strongest evidence that provider supports without reserving anything. Email can be checked properly, because CitizenAI creates the mailbox itself. A rented phone number is best-effort: configuration, reusable inventory, and vendor credit can be checked, but exact availability is only known once the number is rented.
  3. Payment. Priced capabilities draw on the internal wallet first, then a USDC rail for any remainder. See Wallet and funding.
  4. Delivery. The capability is provisioned and attached to the agent.

Failed checks charge nothing

If the check fails you see Provider unavailable. No money moves, and no purchase record is created — a failed availability check is not a purchase and does not appear in payment history.

One at a time

Only one unresolved purchase can exist for the same agent and capability. A duplicate attempt is blocked before it can charge or provision anything. If a settlement result is unclear, the duplicate stays blocked until it is reconciled — deliberately, so you cannot be billed twice for one thing.

Confirmed is not complete

A confirmed payment is a milestone. The purchase completes only when the capability is delivered. If payment settled and delivery did not, that is a refund case, not a silent loss. See Refunds and failed delivery.

Capabilities that need a signup

Adding a social or application account starts a live signup rather than an instant provisioning. Those usually need an email address or phone number already on the agent.