Nous Research announced that its Hermes agent now supports a full suite of Stripe skills, built in partnership with Stripe, that let the agent transact with money of its own rather than just draft a purchase for a human to approve. There are three installable skills, added through a command of the form hermes skills install official/payments/, and they map to three escalating capabilities: stripe-link-cli lets Hermes buy things on the open web, mpp-agent lets it pay for pay-per-call APIs, and stripe-projects lets it provision its own SaaS. Each comes with configurable safety limits on every action.
The safety design is the part worth slowing down on, because it mirrors where the rest of the field has been heading. Primary credentials never enter the agent's transcript, human-approval gates are configurable per skill, and one-time credentials are cleaned up after use. That is the same principle behind this month's other agent-infrastructure launches, keep the real secret out of the model's reach and hand it a scoped, disposable token instead, which is exactly the pattern NewCore's split-key identities and Databricks Omnigent's egress proxy were built around.
What makes this notable is where it points. The agent-with-a-wallet idea has mostly meant checkout so far, an agent completing a purchase a person lined up. Hermes paying for metered APIs and standing up its own paid software is a step past that, toward an agent that behaves like an economic actor: it can incur recurring costs, run services, and spend against a budget without a human in the loop for each transaction. That is genuinely useful for automation and genuinely the thing that makes governance people nervous.
The honest framing is that this is a capability announcement, a skills catalog, not a track record. Provision its own SaaS sounds larger than it is in practice, which is spinning up Stripe-billed projects inside configured guardrails, and how tight those guardrails hold under real workloads is the thing to watch rather than take on faith. It lands in a busy week for exactly this question: NewCore raised $66 million to give agents managed identities, and the broader argument over what an autonomous agent should be allowed to buy, sign, and spin up is only getting louder. An agent that can pay is more useful and more accountable to design for than one that cannot, and Nous just shipped the version that can pay.