Nous Research isn’t saying much beyond the numbers, but what the company has confirmed is interesting enough: it has hit a $1.5 billion valuation, and the money is going toward something called “Hermes for Businesses” — a way for companies to deploy customized AI agents that handle multi-step workflows while keeping their data private and secure.
That last clause is the part I want to sit with. Not the valuation. Not the investor list, which we’ll get to. The promise that a business can hand work to an AI agent without handing over its data.
What was actually announced
Here are the confirmed details, stripped of hype:
- Nous Research raised a $90 million Series B at a $1.5 billion valuation.
- The round was led by Robot Ventures.
- Participating investors include Nvidia, Union Square Ventures, Menlo Ventures, Samsung, and 1789 Capital, where Donald Trump Jr. is a partner.
- Total funding to date is $158 million.
- The capital funds a push into the enterprise sector via “Hermes for Businesses.”
So a company that most people outside AI circles have never heard of now carries a valuation north of a billion dollars, and it’s aiming squarely at ordinary businesses rather than researchers and hobbyists.
Translating “multi-step workflows” into plain English
If you’ve only used a chatbot, you’ve used something that answers one question at a time. You ask, it replies, the exchange ends. Useful, but it’s a conversation, not an employee.
A multi-step workflow is the difference between asking someone for directions and asking them to run an errand. An agent built for multi-step work is supposed to take a goal, break it into pieces, and work through those pieces in order — checking a record, pulling a number, filling a form, sending a notification, and reporting back on what it did.
Think about what that looks like in a small operations team:
- A new customer signs a contract, and someone has to create their account, file the paperwork, schedule onboarding, and alert the account manager.
- An invoice arrives, and someone has to match it against a purchase order, flag discrepancies, and route it for approval.
- A support ticket comes in, and someone has to look up the customer’s history before deciding where it goes.
None of that is intellectually difficult. All of it eats hours. That’s the work agents are being pointed at, and “customized” matters here, because your account creation process is not the same as anyone else’s.
Why the privacy line is the real selling point
For a lot of non-technical business owners I talk with, the hesitation around AI isn’t “will it work.” It’s “where does my stuff go.” If an agent needs to read your customer records to do its job, you have just given a third party a tour of your most sensitive data. For anyone in healthcare, finance, law, or any business operating under a data agreement with its own clients, that’s not a small worry. It’s the whole worry.
Nous Research is known for open-source work on its Hermes agent, and the enterprise pitch leans on keeping data private and secure. The company hasn’t published the technical specifics of how that’s enforced in “Hermes for Businesses,” so I’d treat it as a stated goal rather than a verified guarantee. Still, the fact that privacy is the headline feature of the launch — not speed, not raw capability — tells you what enterprise buyers have been asking about.
Questions to ask before you believe any version of this promise
- Where does the model actually run, and who can see the data it processes?
- Is your data used for training, and can you opt out in writing?
- What does the agent log, and who has access to those logs?
- If the agent takes a wrong action in a multi-step process, how do you find out and undo it?
That last one is underrated. An agent that acts rather than chats can make mistakes that persist — a wrong email sent, a wrong record updated. Any serious agent deployment needs an audit trail and a way to reverse things.
What I’d watch next
The investor mix is worth a second glance. Nvidia has an obvious interest in anything that increases demand for compute. Samsung brings hardware and a sprawling enterprise footprint. The venture firms bring the usual appetite for the next platform. It’s a coalition betting that agents become infrastructure rather than a feature.
For readers here, the practical takeaway is simpler. Agents built for business workflows are moving from demo to product, and the companies selling them have figured out that privacy objections are the gate they have to get through. That’s a healthy development, as long as buyers keep asking for specifics instead of accepting reassurance. A billion-dollar valuation tells you what investors believe. It doesn’t tell you whether the agent will handle your invoices correctly, and that’s the only test that matters in your office.
🕒 Published: