Open-source agents got their enterprise bet: $90M on Hermes

On Wednesday, Nous Research closed a $90 million Series B at a $1.5 billion valuation, led by Robot Ventures with Nvidia, Microsoft's M12, Samsung, Union Square Ventures, and Menlo Ventures participating. The money funds the company's push into enterprise with "Hermes for Businesses": customized AI agents that handle multi-step workflows while, the company says, keeping customer data private and secure. Total funding is now about $158 million for a three-year-old company.

Hermes is the open-source agent that developers would not stop cloning. TechCrunch puts the number above 24 million clones, which the company says drives roughly 2.5% of global AI token usage. Both figures are the startup's own estimates. The Wall Street Journal reports 22.7 million downloads since February, and July figures had the GitHub repo at roughly 214,000 stars and 40,000 forks. Those numbers measure visible developer enthusiasm. They are not customers.

The enterprise pitch, per the company's own materials as traced by RuntimeWire: the agent itself stays MIT-licensed and self-hostable on your own machines; Nous sells what companies actually buy, which is deployment, administration, and governance. Team-level controls, shared skills, a choice of underlying models, private or on-premises deployment, single sign-on, and SLAs. The hosted Nous Portal already lists individual plans at $20, $100, and $200 a month, separate from the enterprise tier now being built. This is the Red Hat model applied to agents: the software is free, the operations are not.

Read the financing the way a buyer should. The WSJ reports Nous was at roughly $36 million in annualized revenue by mid-September and expects to pass $100 million before year-end. That second number is a company forecast, made while raising money, for a business line that just launched. Treat it as a fundraising slide, not a fact. The announcement carried no customer count and no enterprise revenue figure. What the round actually proves is that serious investors (Nvidia, Microsoft's venture arm, Samsung) believe an open-source agent can carry a paid enterprise business. Whether developer love converts to corporate purchase orders is the experiment the $90 million funds.

The bet has a real shape behind it. CEO Dillon Rolnick's framing, quoted in the WSJ, is that businesses should own their own destinies: inspectable code, their own infrastructure, less dependence on any single vendor. For operators with sovereignty requirements or data that cannot leave the building, that is a genuine procurement alternative to the closed-ecosystem plays like Anthropic's startup credits. The pitch is not that Hermes is the smartest agent. It is that the agent is yours.

Two questions to ask before any pilot. First, the boring enterprise ones: what does the SLA actually cover, how does SSO integrate with your identity provider, and who answers the phone when the agent misbehaves on your infrastructure. Second, the interesting one: who owns the context the agent accumulates. Hermes's documentation advertises persistent memory and reusable skills. If the agent learns your team's workflows for a year and you stop paying Nous, where does that learned context live, and how portable is it. Open source answers who can read the code. It does not by itself answer who owns what the agent learned on your data. Get that answer in writing before the pilot, not after.

The funding proves openness is a distribution strategy with a $1.5 billion receipt. It does not prove openness is a business yet. The next twelve months will, or will not.