The control layer for
software that pays.

On one side, AI agents are genuinely capable. They research, negotiate, compare, and decide. As of 2026 they can also pay: x402 went live on Algorand, giving autonomous software a way to settle real value over HTTP, instantly, for fractions of a cent. The payment rail is no longer the blocker.

On the other side, that same capability is a liability. An agent that can spend is an agent that can overspend, be impersonated, or act beyond its mandate. Most stacks have no way to express which agent, acting for whom, is allowed to do what, let alone prove after the fact what actually happened.

The result is a structural gap: the payment rail exists, but the trust layer around it doesn’t. Walletless onboarding still matters (the people and organisations behind these agents won’t manage seed phrases), but the harder problem is now governance: identity, authorization, budgets, and an auditable record for every autonomous action.

Agent-to-agent payments have specific infrastructure requirements that most L1s don’t meet. When an agent executes a transaction, it needs to know, not probabilistically but with certainty, that the payment is final. On chains with probabilistic finality, an agent would need to wait for confirmations, handle potential reorgs, and build retry logic for failed settlements. That complexity defeats the purpose of autonomous action.

Algorand’s Pure Proof of Stake provides deterministic finality in roughly 2.8 seconds with no possibility of forking. A payment is settled the moment it’s confirmed, with no reorgs and no retry logic. Transaction costs sit below $0.001, and x402 is now natively supported on Algorand (the spec merged with Coinbase, the facilitator is live), with atomic transaction grouping so a payment and the action it pays for settle together or not at all. These aren’t marketing features; they’re the technical requirements for payment rails where software transacts at software speed.

Existing account abstraction providers (Dynamic, Privy, Web3Auth) have made progress on consumer wallet onboarding. They solve the “sign in without MetaMask” problem. But they’re building for humans interacting with dApps, not for software acting on behalf of humans. None of them offer policy-constrained wallets, agent-to-agent settlement protocols, or purpose-bound fund management.

Kirke Labs operates one layer deeper. We don’t compete with wallet providers. We provide the infrastructure that makes a wallet useful for autonomous commerce. Now that the settlement rail itself is commoditised (x402 is live and open), the durable advantage is governance: the walletless onboarding is the entry point, and the policy engine and Open Agent Access are the product.

Our initial vertical is charitable giving: a sector where fund traceability is a regulatory requirement, donor trust is the primary growth constraint, and the existing payment infrastructure provides zero transparency between donation and impact. Policy wallets solve this structurally: every pound is cryptographically bound to a stated purpose, and every expenditure is auditable on-chain. From there, the same infrastructure extends to enterprise procurement, energy markets, healthcare logistics, and any domain where autonomous agents need to move money within defined rules.