The pitch from Nous Research, as reported alongside its new funding, boils down to something refreshingly plain: companies should be able to deploy customized AI agents that handle multi-step workflows without handing over the keys to their data. That’s it. No talk of machine consciousness, no promises about reinventing work itself. Just agents that can do several things in a row, on your terms.
I’ll admit my first reaction was a small sigh of relief. If you’ve spent the past couple of years reading AI announcements, you know how rare it is for a company to describe its product in words your operations manager would actually understand.
What actually happened
Nous Research confirmed it closed a $90 million Series B at a $1.5 billion valuation. The round was led by Robot Ventures, with backing that includes Nvidia, Union Square Ventures, and Samsung. That brings total funding to $158 million.
Alongside the money came a product: “Hermes for Businesses,” which lets companies deploy customized AI agents for multi-step workflows while keeping data under their own control.
The revenue picture, per reporting on the round: roughly $36 million in annualized revenue by mid-September 2026, with the company expecting to pass $100 million by year-end.
The number that tells the real story
Forget the valuation for a second. The figure I keep coming back to is this one: the open-source Hermes agent has been cloned more than 24 million times, and it drives about 2.5% of global AI token usage.
Let me unpack both of those, because they’re the kind of stats that get quoted without being explained.
“Cloned 24 million times”
Cloning, in the open-source world, means someone copied the project onto their own machine to run it, study it, or modify it. It’s not a download in the app-store sense and it’s not 24 million separate companies. Developers clone things repeatedly, and automated systems clone constantly. Still, that volume means a very large number of people have put their hands on this software directly rather than just reading about it.
“2.5% of global AI token usage”
Tokens are the chunks of text AI models read and write. Think of them as roughly word-sized pieces. Token usage is the closest thing the industry has to a meter reading of how much AI work is actually being done worldwide.
So 2.5% of global token usage means that out of all the AI processing happening on the planet, about one in forty pieces of that work traces back to Hermes. For a three-year-old company with $158 million raised, sitting next to firms that have raised tens of billions, that’s a genuinely odd and interesting position to hold.
Why the “keeping data” part matters to you
If you work at a company that has thought about AI agents and then quietly backed away, there’s a decent chance the sticking point was data. Someone in legal asked where customer records would end up, nobody had a clean answer, and the project went into a folder.
Open-source models sidestep a chunk of that problem. Because the model’s inner workings are published rather than locked away, a company can run it on infrastructure it controls instead of shipping sensitive information to an outside service. That’s not a security guarantee by itself, and it doesn’t remove the need for proper review. But it changes the conversation from “trust this vendor” to “configure this ourselves.”
Nous Research is now selling exactly that distinction to businesses.
The multi-step thing is the whole point
“Multi-step workflows” is the phrase doing the heavy lifting in this announcement, and it’s the difference between a chatbot and an agent.
- A chatbot answers your question. One input, one output, done.
- An agent takes a goal and works through a chain of actions to reach it: look up the record, compare it against the policy, flag the mismatch, draft the follow-up email, log what it did.
Each link in that chain is a place where things can go wrong, which is why most agent deployments still need a human checkpoint somewhere. Anyone planning one should budget for that supervision rather than assume it away.
What I’d watch
The open-source community gave Hermes its reach. Enterprise customers will now fund its future. Those two groups want different things, and plenty of companies have strained the relationship trying to serve both.
For non-technical readers, the practical takeaway is simpler. There’s now a serious, well-capitalized option for teams that want agents without surrendering their data, and the company behind it has the usage numbers to back up the claim that people actually build on its software. Whether the revenue projection lands is next quarter’s question. The 24 million clones already happened.
đź•’ Published: