Nous Research Lands 90 Million Dollars As Its Open Source Hermes Agent Heads Into The Enterprise

Hermes Agent was released under the permissive MIT licence in February 2026. Rather than answering a single prompt and stopping, the agent is built to carry out multi step tasks. It remembers projects, writes reusable skills as it works, and runs either locally or in the cloud on macOS, Windows and Linux. Because the code is open, developers and individuals can inspect it, modify it and run it on their own machines, which helps explain its spread.

The adoption figures are large, though most come from the company. The Journal reported 22.7 million downloads. Nous says the agent has been cloned more than 24 million times and, by its internal estimates, accounts for roughly 2.5 percent of global AI token usage. Tokens are the units in which AI models bill their work, so a share of that size would imply a very large amount of everyday agent activity running through Hermes, but the estimate has not been independently verified.

Revenue is where the story gets more specific. According to the Journal, Nous reached about 36 million dollars in annualised revenue by mid September 2026 and expects to pass 100 million dollars before the year ends. That would nearly triple the run rate in a few months. At the reported valuation, investors are paying roughly 40 times current annualised revenue, a multiple that depends heavily on the company hitting its own forecast.

The new capital is earmarked for Hermes for Businesses, an enterprise offering that lets companies deploy customised agents to handle multi step workflows without exposing their data. Privacy is the central selling point. Many corporate buyers are wary of sending internal documents and customer information to external services, and an agent that can run under a company's own control addresses that concern more directly than a hosted chatbot.

Nous began as an informal open source research collective before incorporating as Nous Research, Inc. It first built its reputation on its Hermes family of open models and on research such as the YaRN method for extending the context length of language models. Hermes Agent turned that credibility into a product with mass usage, and the Series B is the first large bet on converting that usage into recurring contracts.

The investor list gives some clues about how the market sees that bet. Nvidia sells the chips on which agents run, Samsung makes devices where they could live, and M12 represents a major platform company. Their participation points to hardware and platform players treating open agent software as something worth supporting, not merely watching.

The strategy carries a real tension. Open source distribution has built a wide base of users at almost no acquisition cost, but enterprise buyers have historically preferred vendors that offer clear accountability, support contracts and service guarantees. Nous has to persuade them that an agent anyone can download can also be dependable enough to run sensitive workflows, while keeping the open community that made it popular.

Competition is also building from larger rivals. Major AI labs and software vendors are promoting their own agents and bundling them into existing products, which gives them distribution that a startup cannot match. Nous is wagering that openness, local control and a large developer following will count for more than a bundled convenience.

The coming months will show whether the revenue forecast holds. If Hermes for Businesses converts even a small share of its enormous developer base into paying customers, the valuation will look conservative. If it does not, the round will stand as a high priced test of whether popularity in open source translates into enterprise sales.