Nous Research Funding Puts Open Agents Against Enterprise Control

Nous Research funding reached $90 million as the Hermes Agent developer launched a direct push into business accounts at a $1.5 billion valuation.

That combination matters more than either announcement alone. Nous built its following around an open agent that developers could inspect, modify, and operate with considerable freedom. It must now make that flexibility acceptable to companies that prioritize access controls, predictable spending, auditability, and support.

The contest is therefore not simply Nous Research against another AI vendor. It is open agent autonomy against enterprise control. Microsoft, OpenAI, Anthropic, and other established providers already sell managed AI services through familiar procurement channels. Nous is betting that companies also want ownership over the skills, memories, and workflows their employees create.

The Nous Research Funding Round Comes With a Business Launch

The round gives Nous enough capital to test whether grassroots adoption can become a credible enterprise business.

Nous Research disclosed a $90 million Series B in October 2026. Robot Ventures led the round, according to the company and subsequent reporting. Participating investors included Nvidia, Microsoft’s M12, Samsung, Union Square Ventures, Menlo Ventures, Y Combinator, and 1789 Capital.

The financing values Nous at $1.5 billion. It also brings the three-year-old company’s reported total funding to $158 million. Both the financing and valuation were detailed in the initial funding confirmation.

The final amount exceeded the figure circulating several months earlier. In July, Nous was reportedly preparing to raise at least $75 million at the same valuation. The completed round shows that the earlier valuation held while the company expanded the deal.

An independent deal summary also identified Robot Ventures as the lead investor. It listed Nvidia, M12, Samsung, USV, and 1789 Capital among the participants.

The investor mix gives Nous more than financial runway. Nvidia provides a connection to the compute layer, while M12 and Samsung bring relationships across enterprise software and hardware. USV and Menlo add experience supporting software platforms that need durable distribution.

However, strategic names do not guarantee enterprise adoption. Large companies select vendors through security reviews, legal negotiations, data-governance assessments, and internal pilots. Nous still needs to turn investor confidence into repeatable customer deployments.

The company is tying that effort to Hermes Business, a shared deployment of Hermes Agent for teams. Hermes Agent is an open-source software harness that connects AI models with tools, persistent memory, messaging services, and multi-step tasks.

A harness differs from the underlying language model. The model generates and interprets language, while the harness manages tools, instructions, memory, permissions, and task execution. That distinction lets Hermes work with different models instead of locking every workflow to one model provider.

The new business offering adds centralized administration around that agent. Team members can receive hosted agents, while administrators manage membership, spending limits, and access. Skills created by one employee can also be shared across the organization.

Nous says larger customers can run an enterprise version on infrastructure they control. The company also advertises private-cloud and on-premises deployment, single sign-on, service commitments, and tailored onboarding.

Those features explain the timing of the financing. Nous is not raising money only to train another model or attract more developers. It is financing the operational layer required to support companies with different infrastructure, security, and governance requirements.

The round also arrives after unusually fast adoption claims. Nous estimates that Hermes Agent has been cloned more than 24 million times and drives roughly 2.5% of global AI token usage. Those figures have not been independently audited, so they should be treated as company estimates.

Even with that qualification, the direction is clear. Hermes gained enough developer attention to give Nous a plausible path into corporate accounts. The funding now lets the startup test whether that attention reflects lasting demand.

Why Hermes Business Changes the Open-Agent Bet

Hermes Business turns an individual developer tool into a shared organizational system, which raises both its potential value and its risk.

A personal agent can be useful without becoming organizational infrastructure. One person can connect an inbox, create a recurring task, or teach the agent a reusable procedure. If the setup fails, that person can revise it or stop the process.

A company deployment has different consequences. A shared agent might draft customer communications, organize research, review engineering issues, or prepare recurring status reports. Its stored memories and learned skills can influence work across a team.

Nous describes Hermes Business as a shared deployment with one central balance and an agent for each team member. Its published business controls include administrative roles, member management, per-user limits, shared skills, and centralized visibility into consumption.

The most distinctive promise concerns organizational learning. When someone develops a useful skill, the system can distribute that skill across the team. A skill is a reusable set of instructions or procedures that helps an agent complete a recurring task more consistently.

That mechanism could reduce repeated setup. A product manager might create a method for summarizing customer interviews. An engineer could develop a procedure for reviewing incident reports. A sales operations employee might define how the agent prepares account briefs.

Shared procedures can accumulate into a company-specific operating layer. The value would not come only from the model answering a prompt. It would come from the organization preserving how its employees use models, tools, and internal information together.

This is where Hermes Business overlaps with the broader knowledge-management problem. Companies already struggle to preserve decisions and practices across documents, chats, meetings, and employee transitions. An agent can act only as well as the context and permissions available to it.

A searchable AI knowledge base addresses part of that challenge by making information retrievable. An agent adds an execution layer, but it also introduces more ways for outdated context or excessive permissions to produce a bad result.

The company’s open architecture is central to its pitch. Nous argues that customers should control their agents, data, memories, and accumulated capabilities. It positions openness as an alternative to relying completely on a closed application or a single model vendor.

Model flexibility can help buyers manage changing performance and availability. A team might use one model for coding and another for long-context document work. It can update the model layer without rebuilding every agent procedure.

That flexibility also creates operational complexity. Different models interpret instructions differently, support different tool formats, and produce different error patterns. Switching models is easier when a harness abstracts those differences, but it is not automatically risk-free.

Hermes Business must therefore do two jobs at once. It needs to preserve the adaptability that attracted developers while making varied models and workflows manageable for administrators.

That is harder than adding a team dashboard. An enterprise agent system needs clear permission boundaries, versioned procedures, monitoring, recovery mechanisms, and reliable records of actions. It must also help administrators understand why an agent behaved differently after a model or skill changed.

Nous says enterprise customers can deploy Hermes on controlled infrastructure. That can improve data custody, but infrastructure ownership does not solve every governance problem. A company still needs to decide which systems an agent can access and what actions require human approval.

The launch changes the open-agent bet because it shifts the measure of success. Downloads, clones, and developer enthusiasm established distribution. Enterprise buyers will judge Hermes through reliability, governance, support, and measurable workflow outcomes.

Open Agent Autonomy Meets Enterprise Control

The central tension is whether Nous can preserve user-owned agents while giving employers enough control to accept their actions.

Nous formed in 2023 around opposition to centralized AI and closed systems. In its fundraise note, the company said the new capital would support open and auditable AI rather than gate-kept access.

CEO Dillon Rolnick framed control as a continuing principle behind Hermes. The company’s position is that people and organizations using an AI system should own the resulting intelligence, including the system’s accumulated skills and memory.

That argument resonates with technical teams concerned about dependency. If a workflow exists only inside one vendor’s product, moving it can require substantial rebuilding. Open code and portable procedures can provide more negotiating room.

Enterprise control, however, has two meanings. Nous emphasizes customer control over technology and data. Corporate security teams also need control over employees, connected systems, credentials, and automated actions.

Those goals can align, but they can also conflict. An agent designed to learn from use might improve a procedure without an administrator noticing the operational change. A shared skill might spread an error as efficiently as it spreads a useful technique.

Consider a routine customer-support workflow. One employee teaches the agent how to classify incoming requests and draft responses. Sharing that skill could save work across the team.

Yet the procedure may contain assumptions that apply only to one product, region, or customer segment. If the system distributes it too broadly, the organization can produce consistent but incorrect responses. The governance challenge lies in deciding who can publish, review, update, or withdraw the skill.

The same tension appears in memory. Persistent memory lets an agent carry context between sessions. That can reduce repeated explanations and make recurring work more efficient.

For a business, memory also becomes governed information. It can contain customer details, internal decisions, speculative notes, or conclusions that later become outdated. Administrators need retention rules, access boundaries, correction mechanisms, and dependable deletion.

Tool access raises the stakes again. A read-only agent that summarizes documents presents one risk profile. An agent that sends messages, changes records, runs code, or initiates financial processes presents another.

Organizations usually manage those risks through least-privilege access, which gives a system only the permissions required for a specific task. They also use approval checkpoints for consequential actions. Hermes must support those practices without making every workflow too cumbersome to use.

This is why Nous is not merely packaging an open-source project for corporate procurement. It is trying to create a controlled distribution system for autonomy.

Established enterprise vendors approach the problem from the opposite direction. They begin with managed identities, cloud controls, compliance programs, and existing customer relationships. They then add more agent capabilities inside those boundaries.

Nous begins with an adaptable agent that developers chose for themselves. It is now adding the boundaries. That starting point can create a more flexible product, but it also leaves the company with more enterprise infrastructure to prove.

OpenClaw provides a useful market reference. Both projects let users run persistent agents, connect tools, and automate recurring work. A published agent workflow comparison described OpenClaw as easier to start, while characterizing Hermes as more focused on learning reusable skills from repeated work.

That comparison is not a definitive benchmark. It illustrates the choices buyers face. Breadth and immediate setup can attract one group, while deeper customization and accumulated procedures can attract another.

For business customers, neither route wins through autonomy alone. Companies need to know how the agent fails, which person authorized an action, and whether administrators can reverse a change.

Nous can retain its open-source identity while serving those needs. Openness can help security teams inspect software, run it in controlled environments, and customize enforcement. Still, inspectable code is not the same as an audited deployment.

The outcome will depend on product mechanics. If Hermes gives administrators strong controls without neutralizing employee experimentation, Nous can offer a meaningful alternative to closed agent platforms. If governance feels bolted on, enterprise buyers may prefer more established vendors.

The Valuation Assumes Developer Adoption Can Convert

A $1.5 billion valuation places pressure on Nous to turn community usage into durable corporate revenue.

The startup enters that test with reported momentum. The Wall Street Journal said Nous had reached roughly $36 million in annualized revenue by mid-September 2026. It also reported that the company expected to pass $100 million before year-end.

Annualized revenue is a snapshot that extends a recent revenue rate across a full year. It is not the same as revenue already collected over twelve months. Rapidly changing usage can make the figure rise or fall quickly.

The Journal’s enterprise expansion report also said Hermes had been downloaded more than 22 million times since its February launch. TechCrunch later cited the company’s estimate of more than 24 million clones.

The difference likely reflects reporting dates or measurement updates, but neither figure reveals customer quality. A clone can represent a serious production deployment, an experiment, an automated build, or repeated activity by the same user.

Token usage also needs context. Nous estimates that Hermes accounts for about 2.5% of global AI token consumption. Without an independently published methodology, readers cannot determine which providers, models, workloads, or time period the calculation includes.

These caveats do not make the metrics meaningless. They show why developer distribution and enterprise adoption must be measured differently.

A developer project can grow through public repositories, community recommendations, and experimentation. Enterprise revenue usually requires contracts, deployment services, support, integrations, security reviews, and predictable renewal behavior.

The business launch gives Nous a mechanism for that conversion. A company can begin with individual Hermes users, then centralize administration and shared resources. Larger organizations can move toward private-cloud or on-premises deployments.

This bottom-up path has worked for other developer-oriented software companies. Individual users introduce a tool, teams standardize it, and central administrators eventually purchase controls. However, AI agents create broader operational risks than many conventional collaboration tools.

An agent can interact with many systems through one interface. That expands its usefulness, but it also concentrates access. A compromised account, flawed instruction, or poorly reviewed skill can affect several connected services.

The cost of supporting enterprise deployments can also challenge a startup’s margins. Customers may need help with identity systems, infrastructure, model selection, data policies, and custom tools. Each variation makes a deployment harder to standardize.

Nous claims one harness can provide visibility across different AI models. That can become a valuable control point if customers use several providers. It can also make Nous responsible for explaining problems that originate elsewhere in the stack.

Suppose a model update changes how a workflow handles a tool call. The buyer may experience the failure through Hermes even when Nous did not change the harness. Enterprise support must diagnose the complete chain rather than redirecting customers between vendors.

Competition will compound that pressure. Large AI providers can bundle agents with model access, cloud infrastructure, and existing enterprise agreements. Productivity-software companies can place agents inside applications employees already use.

Open-source alternatives can also compete without carrying Nous’s valuation expectations. Companies with strong internal engineering teams may assemble their own agent systems from public components. Managed hosting providers can package open tools for narrower markets.

Nous must therefore sell more than access to code. It needs to make Hermes easier to govern, operate, and improve across a company than a self-built alternative.

Its funding provides time and resources for that work. It can expand sales, support, security, infrastructure, and product management. It can also subsidize the transition from individual adoption to larger managed deployments.

The valuation indicates that investors expect this transition to happen quickly. If enterprise customers adopt Hermes Business and expand usage, the developer community becomes an efficient distribution channel. If adoption stalls, Nous will remain popular while supporting a business model that does not match its capital base.

What the Adoption Numbers Do Not Prove

Nous has evidence of attention, but the public data does not yet establish enterprise reliability or retention.

The largest uncertainty concerns measurement. Nous has published or supplied impressive estimates for clones and token usage. It has not publicly provided a detailed breakdown of active organizations, production workflows, retention, or renewal rates.

Those missing figures matter because agent experimentation is easy to start. Sustained use is harder. A workflow must remain useful after the novelty fades, source systems change, and employees encounter edge cases.

Revenue growth offers a stronger signal than repository activity, but the reported figures still need interpretation. Annualized revenue can respond quickly to consumption. It does not show how much revenue comes from recurring subscriptions, model usage, enterprise contracts, or temporary demand.

The company’s year-end expectation is also a forecast, not a completed result. Reaching that figure would support the idea that demand is accelerating. Missing it would raise questions about conversion, usage stability, or the assumptions behind the projection.

Security is another unresolved area. Nous promotes private deployments and auditability. Those properties can help organizations control data and inspect components.

They do not automatically establish that a particular configuration is secure. Customers need evidence covering identity management, secrets, tool permissions, network boundaries, logging, software updates, and incident response.

Self-improving behavior requires special scrutiny. If an agent turns successful work into reusable skills, administrators need a record of what changed and why. They also need a way to test the new procedure before it reaches other employees.

A useful governance design might separate skill creation from organization-wide publication. Employees could propose a skill, while designated reviewers examine its instructions, permissions, test results, and intended audience.

Versioning would let teams trace failures to a specific change. Rollback would restore an earlier procedure. Scope controls would prevent a sales workflow from appearing automatically in engineering or finance.

Nous’s public business page highlights shared skills and centralized controls, but it does not answer every implementation question. Buyers should examine those details during pilots rather than treating broad feature labels as proof.

Deployment ownership presents a similar tradeoff. Running software on controlled infrastructure can reduce dependence on a hosted vendor. It also transfers more operational responsibility to the customer.

A company must patch the system, protect credentials, monitor activity, and maintain connected services. Private infrastructure can improve control only when the organization has the staff and processes to manage it.

The competitive comparison remains unsettled as well. OpenClaw has its own adoption and ecosystem advantages. Closed platforms can offer deeper integration with their model and cloud stacks. Smaller specialists may target individual functions with tighter boundaries.

Hermes Business does not need to defeat every alternative. It needs a clear segment where openness, model flexibility, shared learning, and controlled deployment outweigh setup and governance costs.

Technical organizations are the most obvious early candidates. They already understand open-source deployment and may value the ability to inspect or modify the agent. They are also better equipped to evaluate model behavior and tool permissions.

Less technical teams may prefer a narrower managed product. They often value predictable setup and predefined workflows more than architecture choice. For Nous, serving both groups without creating product complexity will be difficult.

The financing announcement confirms that Nous can attract prominent investors. The business launch confirms that it intends to sell beyond individual developers. Neither result confirms that enterprises will trust autonomous, learning agents with consequential work.

That proof must come from deployments, measurable outcomes, and evidence that the controls hold when users and workflows multiply.

Three Signals Will Decide the Hermes Business Push

The next phase should be judged through enterprise adoption, governance evidence, and revenue quality rather than download milestones alone.

The first signal is the shape of early enterprise deployments. Named customers would help, but deployment depth matters more than logos. A limited research assistant and an agent acting across production systems represent very different commitments.

Buyers should watch for disclosed workflows, active team sizes, expansion across departments, and the length of pilots. Movement from read-only analysis into controlled action would indicate growing confidence.

Repeated expansion inside existing customers would strengthen the case for Nous. A long list of experiments without broader deployment would weaken it.

The second signal is concrete governance documentation. Nous should show how organizations approve shared skills, restrict tools, isolate data, inspect actions, manage memory, and reverse changes.

Independent security assessments or detailed technical documentation would make the company’s auditability argument more credible. Published incident handling and transparent software-update practices would provide further evidence.

A mature control system would support both central policy and local experimentation. Excessive centralization would undermine the autonomy that differentiates Hermes. Insufficient control would keep security teams from approving meaningful use.

The third signal is revenue quality. Nous reportedly expected annualized revenue to rise sharply before the end of 2026. The important follow-up is whether that growth comes from sustained organizational adoption.

Recurring enterprise contracts, customer expansion, and stable usage would support the valuation. Revenue driven mainly by temporary individual demand or subsidized consumption would provide weaker evidence.

Competitor reactions will offer additional context, although they should not replace these three measures. Larger vendors will keep adding agent controls, while open projects will continue improving usability and governance.

Nous Research funding gives the company enough room to build its enterprise layer, but it also raises the standard by which the market will judge Hermes Business. The decisive question is no longer whether developers will try an open agent.

It is whether companies will entrust shared knowledge, tools, and workflows to one while retaining meaningful oversight.

For enterprise buyers, the sensible next step is a bounded pilot. Choose one recurring workflow, limit permissions, record every action, and define success before deployment. If Hermes Business can improve that workflow without weakening control, Nous’s open-agent thesis gains real support. If governance consumes more effort than the automation saves, the valuation story remains ahead of the product evidence.