Skip to main content
With Apostra, connecting an AI assistant grants it no authority to spend, and Apostra refuses to launch a campaign without a second, separate confirmation call tied to the exact campaign revision that was previewed. Apostra enforces that second call, not who makes it: the same agent can make both calls unless your application or host puts a person in front of the second one. Billing changes wait for an administrator to approve them on Apostra’s own site. An agent’s access to an advertiser can be read-only. Where the seller supports it, a buy can carry a daily budget cap, and a sandbox advertiser lets the assistant rehearse a whole campaign without spending.

What Apostra is, and what it is not

If you are looking for a single-platform ad optimiser that tunes one account’s bids and budgets on its own, Apostra is not that. Apostra is a cross-seller system. Your assistant or application finds sellers, requests proposals, stages buys and reads delivery through Apostra’s tools, and Apostra applies the controls below around those actions. It does not generate creative, and it does not host your agent: your application or assistant runs the loop and decides who makes the launch confirmation.

Controls and their limits

How a person stays in the loop

  1. The assistant drafts a campaign. Drafting does not launch it.
  2. It requests proposals or browses products, then stages a draft media buy.
  3. It asks for the launch preview and shows you the buys, sellers and combined budget.
  4. A person approves, if your application or host asks one to. Apostra does not check who makes the confirmation call. The confirmation names the revision that was previewed. If the campaign changes afterwards, a new preview and a new confirmation are needed.
  5. If a launch fails, Apostra does not retry what only you can fix. For causes that may clear on their own, such as an unreachable seller, it retries up to five times and tells you when.
Apostra can see API calls, server-side model calls and Apostra-hosted turns. It cannot see the token usage or model cost of an external client’s own model, so you keep that budget yourself.

When not to use Apostra, or to add something else

  • You need a spending ceiling enforced by the ad platform or your payment card. Apostra’s controls apply per buy, per launch and per grant. Set platform and card limits as well where they exist.
  • You need Apostra itself to guarantee that a person approved each launch. It requires a confirmation call, and the application you build decides who makes it, so human approval is enforced by your application or host, not by an Apostra server rule.
  • You want a hard cap on an existing buy, or on a platform-connected seller. Neither is supported today.

Next step

Connect an assistant and try a launch in a sandbox advertiser, which spends nothing. In Claude or Cowork, follow Claude and Cowork setup. To build your own approval checkpoint, follow Build an agent. To set an agent’s advertiser access, read Manage buyer agents.

Sources