With the AI agents not only tools, but also workers in the internet, a new issue arises: how to allow the software to spend the money and not lose control over it? It is hard to entrust an agent with the full wallet, but depriving them of it leaves them without any real autonomy. Kite is being constructed to eliminate this tension by ensuring that delegation is made safe by design. Kite is an agentic and EVM-compatible Layer 1 blockchain. In contrast to the traditional chains, which rely on human approval of all transactions, Kite is based on the assumption of agents acting continuously, generating lots of actions, and transactions involving relatively high frequencies and low payments. Such a change of assumptions results in radically new infrastructure decisions--speed, real-time settlement, and damage-limiting security are put in the first place.
Kite is centered around three-layer identity model: user, agent and session. The origin of authority is the user layer and it contains root keys and policies. It makes accountability real and so the responsibility always goes back to a human or an organization. The agent layer implies autonomy but with some rigid boundaries. There are fences to agents: they are given roles, wallets and limits where they can behave freely but in a fenced way. Safety is applied to the session layer. Sessions are scoped windows of execution, or missions in other words. They determine the amount of time, the amount, and what an agent is able to do. The sessions may be cancelled immediately, i.e. in case of any wrongdoing, the blast radius remains minimal. This transforms trust into a decision to trust all or to trust none.
Kite also focuses on programmable governance implemented in execution time. Spend limits, service limitations and action constraints are not guidelines but hard rules. This is because the safety is guaranteed through prevention and not guarantees. To make payments, Kite is bent towards state channels so as to provide agent friendly micropayments. Agents are able to pay per query, per second of compute/request, or an almost instantly settled request and the ultimate settlement occurs on-chain. That is how machine-speed commerce can be made possible without overloading the network.
Other than payments, Kite views identity as a source of trust and coordination. The agents are able to develop reputation, services to authenticate and interactions can be tracked. The outcome is a system not only to be autonomous, but to be safe and accountably automated. In case AI agents will be involved in actual business, Kite is hoping to be the silent, yet vital system that enables it to happen without losing control.
