Nous Research's Hermes Agent: The Open-Source, Model-Agnostic Orchestrator Lands $90M and an Enterprise Play
On October 7, 2026, Nous Research confirmed a $90M Series B at a reported $1.5B valuation for the team behind the open-source Hermes Agent — the best-funded direct comparable yet to Hey, Lefty's lane of the market map (Market Map & Positioning: Where Does Hey, Lefty Fit?): a model-agnostic, user-controlled research/work orchestrator. Robot Ventures led; Nvidia, Microsoft's M12, Samsung, Union Square Ventures, Y Combinator, Menlo Ventures, and 1789 Capital participated, bringing total funding to $158M (TechCrunch).
What the product is (features / core tech): Hermes Agent was released in February 2026 under an MIT license. It is a desktop agent that "can work with files and programs, retrieve information through a browser, and run code in an isolated Docker environment," keeps memory across sessions, writes reusable skills as it works, and handles scheduled tasks. Critically for the model-agnostic thesis: "It can also use models from other providers, including Claude, alongside Nous's own models. A built-in router is described as weighing cost, speed, and quality when choosing where to send a prompt" (Superpower Daily, citing SiliconANGLE).
Business model: open core plus paid hosting and administration. "Users can run it themselves or pay for hosted plans costing $20 to $200 a month" (TechFundingNews). Three paid editions: Hermes Cloud (consumer, no infra setup), and Hermes Business / Hermes Enterprise, which add team-level token-use limits and shared agent customizations. The enterprise pitch is model choice "by price and goal," private deployment, and ownership of institutional knowledge — i.e., selling administration and governance around software anyone can download.
Traction and claims (handle with care): Nous says Hermes has been cloned 24M+ times and drives ~2.5% of global AI token usage — both self-reported (the WSJ counted 22M+ downloads). The WSJ reported ~$36M annualized revenue by mid-September 2026, with a forecast to pass $100M by year-end — a run rate, not collected revenue, and unproven (TechCrunch/Superpower Daily).
What it means for the map: this is simultaneously validation and threat for Hey, Lefty's positioning. Validation: model-agnostic orchestration with user-owned knowledge is now a $1.5B-valuation thesis backed by Nvidia, M12, and Samsung1. Threat: open-source distribution (24M clones) plus a funded enterprise go-to-market aimed at exactly the "choose your models, keep your data" buyer. The open question is the one TechFundingNews puts plainly: "The harder test is whether firms that can run Hermes for free will pay Nous to run it for them."2 The lane is also crowding on the open-source side — Hermes "arrived weeks after OpenClaw, the open-source agent TechCrunch calls its direct rival"; OpenAI hired OpenClaw's creator in February, and Nvidia (a Nous investor) has built an enterprise platform on OpenClaw. For market-sizing context, Grand View Research estimates AI agents at $7.6B (2025) growing to $182.9B by 2033 (49.6% CAGR), with enterprise agentic AI $5.3B (2026) → $24.5B (2030).
The CEO framing is the positioning statement of the whole lane: "One principle has stayed constant since the first Hermes release: the people using an AI system should control it" — Dillon Rolnick, CEO, Nous Research, in the fundraise note.
-
An instance of When models and retrieval commoditize, the orchestrator seat becomes the contested prize. — Capital confirms the orchestrator seat — not the models it routes between — as the asset being fought over. ↩︎
-
An instance of Open weights are the free sample; the hosted meter is the product. — Open-source distribution is monetized through hosted administration and governance — the free sample carrying the paid meter. ↩︎