Kong's New Tools Let Companies Charge AI Costs Back to Teams and Agents

Kong, whose gateways manage the traffic between applications and the services they call, has released its AI cost tools in its AI gateway, after limiting them in September to early-access design partners.

A model provider’s invoice shows what each model cost, not which team or agent ran up the charge. Kong prices each call as it crosses the gateway, which turns that invoice into a ledger by team and by agent, and a hard budget can stop an overrun before the bill arrives. A finance team can track whose agent is spending as the tokens are used, rather than waiting for the month-end bill.

What does a buyer get?

The tools start from a price list. Kong keeps each provider’s token prices in one catalog, including a company’s negotiated rates, and assigns the spend to a team, user, application, agent or cost center. Any of them can carry a budget, and a hard budget stops spending before it overruns. Kong also forecasts consumption and flags spikes as they happen.

None of it costs extra: the cost tools come with Kong’s AI gateway on Konnect, its management platform, at no separate charge. The bundle also carries Token Vault, a new way of giving agents access, and that is where each agent’s identity comes in.

What do the agents get?

Every interaction an agent has “has to be discovered, authorized, governed, observed and paid for,” said Augusto Marietti, Kong’s CEO and co-founder. Kong starts with a name: Kong Identity identifies each agent as it does people and applications, including when an agent acts for a user.

Token Vault builds on that name. The agent signs in to Kong, which checks the systems it may use. When the agent calls one, Kong supplies the key or token that system requires, so the agent never holds it.

The same identity carries the bill. Kong reads which agent, team and application sent each request as it passes, so the agent that signs in at the gateway is the agent the cost lands on.

What is new, and what is not?

Kong’s gateway could already cut the bill. Since September 2024 it has reused answers to prompts that mean the same thing, a feature called semantic caching, and sent each request to a suitable model through semantic routing.

What is new is the ledger, and it is not finished. A buyer cannot yet see what caching and routing saved; Kong lists that measure as coming soon. Nor did Kong say how many customers run the cost tools or what any has saved.

The audit trail comes next. Advanced AI Observability, due in late November or early December, follows an agent through every model call and tool so a team can explain what it did and what each step cost.

Can the company’s finance team name the team, and the agent, behind last month’s largest model bill?