PARIS — AI agents are starting to do more than search, analyse or recommend. They can now execute tasks, access paid services and, increasingly, initiate transactions.
This raises a fairly simple question for companies: if an agent can spend money, how much autonomy should it have and who remains in control?
Most existing payment tools were designed around a human initiating or approving a transaction. That model breaks down when an agent needs to make frequent, sometimes low-value payments as part of its work. Keyban built Agent Wallet for this situation — and from today, any company can try it free.
The solution combines a policy engine with a proprietary, European-developed self-custody wallet infrastructure. Before a transaction is signed, the Policy Engine checks it against rules defined by the company: spending limits, authorised beneficiaries, currencies, types of purchases or other business-specific conditions.
If the transaction complies with those rules, the wallet executes it autonomously. If it does not, it is blocked. Agents get room to operate without unrestricted access to company funds. Every payment is verified before being signed.
The wallet itself is based on MPC/TSS technology developed by Keyban. It is self-custodial and designed to let companies remain sovereign over their assets and transaction infrastructure, rather than handing custody to an intermediary. MPC/TSS also removes the dependency on a single private key while making automated transaction signing possible.
The first use cases are straightforward. An AI agent may need to buy API calls, access a paid data source, purchase additional compute or AI services, activate a software feature or pay for another digital resource required to complete a task. Others extend to procurement and supplier payments within predefined limits, and eventually to agent-to-agent or machine-to-machine transactions.
Requiring human approval for every €1, €10 or €100 transaction would largely defeat the purpose of giving these agents autonomy. Giving them unrestricted payment credentials would create the opposite problem.
The same principle applies whether the agent is purchasing an API for a few cents, additional computing capacity for a few euros or placing an order with an approved supplier for several thousand euros. The company defines the conditions under which the agent is allowed to act.
This also separates the responsibilities involved in an autonomous transaction. The AI decides what it needs. Keyban's Policy Engine determines whether the transaction is permitted. The Keyban wallet signs and executes it.
The company defines the conditions. The agent operates inside them.
From VivaTech to general availability
Agent Wallet was first presented on stage in Paris last June as part of the VivaTech startup challenges, where Keyban showcased its approach to trusted and controlled agentic transactions. First concrete enterprise use cases include API purchasing, access to paid digital services, procurement and autonomous payments. It provides companies with the infrastructure to define appropriate deployment patterns, governance models and transaction flows within their own environments.
Agent Wallet is now open to every company, free to try. The trial covers the full product — wallet creation, policy engine, budgets, validations and spend monitoring — with no time limit, no commitment and no card required.
Charles Kremer, founder of Keyban, states: “Companies will not give AI agents unlimited access to their bank accounts or corporate cards. And asking a human to validate every payment doesn't scale either. We built Agent Wallet to create the layer in between: the agent gets enough autonomy to do its job, while the company keeps control of the money, the rules and the wallet. The trial is free and open-ended, so teams can prove that on their own workflows before they commit to anything.”
As AI agents become part of procurement, IT, commerce and operational workflows, Keyban expects this financial permission layer to become an important part of enterprise agent infrastructure. Agent Wallet is designed to remain independent from any specific AI model or agentic framework and to support different transaction and payment environments.
The agent gains autonomy. Finance keeps control.
About Keyban
Founded in 2024, Keyban develops blockchain infrastructure for digital assets, trusted data and autonomous transactions, and more globally to enhance the performance of agentic commerce for brands and retailers. Its technologies combine wallet infrastructure, cryptographic proofs and programmable governance to help companies deploy new digital and agentic use cases while retaining control of their data, assets and transactions.
Press contact: Nenad Cetkovic — C.O.O. — nenad@keyban.io / contact@keyban.io