Open-Source Agents Grow Up: Nous Research Raises $90M to Take Hermes Into the Enterprise

The open-source agent story hit a milestone in 2026: Nous Research has closed a $90 million Series B at a $1.5 billion valuation, officially taking its open-source AI assistant Hermes into the enterprise market. The round was led by Robot Ventures, with an impressive roster of participants — Nvidia, Microsoft's M12, Samsung, Union Square Ventures, Menlo Ventures, and 1789 Capital, where Donald Trump Jr. is a partner. Total funding now stands at roughly $158 million.

For readers who haven't met it yet, one sentence: Hermes is Nous's flagship open-source personal AI assistant that can read and write files, drive a browser, search the web, and send messages — essentially an agent that operates your computer and gets work done. Its differentiating design choices are notable: it can learn across sessions by writing down "how the user works" into instruction files, improving itself over time; and users can freely swap the underlying model without losing their agents, memory, or skills. That is a rare architectural choice among today's mainstream agent products.

Why now: the vacuum OpenClaw left behind

Hermes's breakout didn't come out of nowhere. OpenClaw was the first wave of open-source agents in early 2026 — but its founder was subsequently hired by OpenAI, which sent a shockwave through the developer community: a beloved open-source agent project had effectively become a big-tech talent acquisition. Anxiety over whether open-source agents would be "acqui-hired" one by one has lingered ever since.

Hermes caught the overflow of that traffic and sentiment. Since February, Hermes has been downloaded 22.7 million times; on OpenRouter, its usage has surpassed OpenClaw. M12's Peter Lenke, explaining the investment, said the team spotted Nous by watching the OpenRouter charts:

They caught our attention as we were watching the OpenRouter charts… viral adoption.

That remark is worth unpacking. What VCs are looking at is no longer how cool the demo is, but real usage data — viral adoption is itself proof of value. In the agent race, downloads and usage are the hardest metrics to fake and the most convincing ones.

The business model: free agent + paid Portal, selling "data sovereignty"

Nous's business model is straightforward: Hermes itself is free, alongside a paid offering called Nous Portal that bundles models, cloud hosting, and enterprise features. The financials are striking: annualized revenue of about $36 million as of mid-September, with the company expecting to cross $100 million within this year.

But the enterprise pitch deserves a closer look. CEO Dillon Rolnick put it in one line:

I think that businesses should have a level of ownership over their own destinies.

In plain terms: enterprises should own their own destiny. Hermes is open-source, inspectable, downloadable, and modifiable — meaning companies can self-host, and can swap the underlying model at any time without losing agents, memory, or skills. Your data, your workflows, your agent assets are never locked into a single vendor. That is a promise no closed-source assistant can make.

My take: "data sovereignty" will become the primary selling point for open-source agents entering the enterprise — arguably more important than features themselves. A decade of SaaS taught enterprises the cost of vendor lock-in; the agent era will reprice that lesson. Nous is betting that companies will pay for "portability" — what Portal sells is not model capability, but the convenience of hosting plus the freedom of never being locked in.

What it means for indie developers: open-source agents are becoming a new platform layer

This round's signal may matter even more to indie developers and small teams than to enterprises. Two Hermes mechanisms are worth following for anyone building agent applications:

• The skills ecosystem: Hermes organizes and extends capabilities as skills, which means third-party developers can write reusable capability packs for it — a distribution structure reminiscent of the early App Store.
• The self-improvement mechanism: writing user habits into instruction files so the agent gets smoother across sessions turns "memory" from a marketing word into a product mechanism. Whoever nails this first owns the retention playbook for agents.

One level deeper: open-source agents are becoming a new platform layer. Like Linux back then, and Kubernetes later, a good-enough, open-enough agent runtime attracts developers to build on top instead of reinventing the wheel. The endorsement from Hugging Face co-founder and chief scientist Thomas Wolf carries real weight — he has no financial relationship with Nous; HF merely hosts Nous's models and datasets:

Nous has done more for open-source AI than most labs many times its size.

A company founded in 2023 by a group of open-source developers who met on Discord, with no traditional academic pedigree and early crypto-native backers, riding a decentralization ethos all the way to a $1.5 billion valuation. That story alone is a powerful rebuttal to the "big tech monopolizes AI" narrative.

The sober side: four hurdles and a 40x PS multiple

The coming-of-age party is over; the real test begins. M12's Lenke summarized the enterprise challenge bluntly:

It's security, it's governance, it's scale, and it's trust.

Security, governance, scale, trust — each of these four hurdles can kill a consumer hit. Openness is a plus in the developer community but can be a minus in enterprise procurement: security teams ask who audited the code, compliance asks where data flows, the CIO asks who backs the SLA. Rolnick argues that "open, inspectable, downloadable, modifiable" is itself the answer to security. That logic holds technically, but in procurement politics it still needs proving, deal by deal.

The competitive landscape is unforgiving too. Closed-source assistants like Meta's Muse and Instinct are chasing the same enterprise budgets, and incumbents bring distribution, brand, and existing enterprise relationships. Nous's edge is community momentum and architectural freedom; its gaps are a sales machine and a compliance apparatus that must be built from scratch.

Finally, the valuation math: $1.5 billion against $36 million in annualized revenue implies roughly a 40x price-to-sales multiple. That number says the market is paying a premium for the "open-source agent" narrative — betting Hermes becomes the Linux of the agent era. But between narrative and execution lie exactly those four hurdles. Whether Nous delivers on its "cross $100 million in annualized revenue this year" expectation will be the first real test of this round.

Either way, the 2026 agent race has a new variable worth tracking long-term: when open-source agents start selling the freedom of "never being locked in," what will the closed-source giants offer in response?