Hermes Agent's $90M Funding: Decoupling Models from Execution for Enterprise AI

Funding and Market Context

On October 8, 2026, The Wall Street Journal reported that Nous Research closed a $90 million Series B round, bringing total funding to $158 million at a $1.5 billion valuation. Investors include Robot Ventures, NVIDIA, Microsoft's M12, and Samsung. The company claims Hermes has been downloaded approximately 22.7 million times since February 2026 and cloned over 24 million times, with internal estimates suggesting Hermes accounts for roughly 2.5% of global token usage — figures that lack independent verification but indicate significant open-source traction.

CEO Dillon Rolnick simultaneously announced Hermes for Businesses , repositioning the MIT-licensed agent from individual developers to enterprise customers. The core pitch: enterprises should own and improve their AI systems, choosing models, controlling deployment environments, and retaining the knowledge and capabilities agents accumulate during work.

Why Hermes Moves from Personal Assistant to Enterprise

Nous Research, founded in 2023, initially focused on open-model post-training and tool-use research (e.g., YaRN long-context work). In February 2026 they released the Hermes Agent under MIT license to support internal research. For individual users, Hermes offered flexibility: tasks via terminal, Telegram, Discord, Slack; tool calls for files, browser, code execution; model provider choice; local or remote deployment.

For enterprises, that flexibility translates into architectural leverage. The article illustrates a scenario: a dev team builds a code-review agent integrated with internal Git repos, test environments, and bug trackers. Six months later they want to swap the underlying model. If tools, state, and execution logic are tightly bound to the original model platform, migration may require rewriting the entire workflow. Hermes' design aims to avoid that by making the model a configurable component while keeping tools and runtime independent.

Architecture: Decoupling Model, Tools, and Execution

According to official developer docs, Hermes comprises:

Multiple interaction entry points (CLI, messaging gateway, API)

A shared AI Agent core

Provider Resolution layer for model access (provider, API schema, auth, endpoint)

Independent tool registration system (terminal, filesystem, browser, external MCP services)

Session storage

A hypothetical Slack request flow: Messaging Gateway receives message, verifies auth, finds session; AI Agent reads context, builds prompt, sends to configured model; model decides tool calls; Tool System executes operations ; results return to model; session saved; response sent via Slack.

This separation expands model choice. Example: an internal data-analysis agent with database query, Python execution, and report-generation tools can be tested against multiple models — evaluating task complexity, latency, token cost, tool-call success rate — without locking the whole pipeline to one vendor. However, the article cautions that model swap does not guarantee identical behavior; multi-step tool understanding, failure recovery, and context handling differ, so re-evaluation is mandatory .

Execution backends are also configurable: local, Docker, SSH, Daytona, Modal. This matters for enterprise workloads needing network isolation, compute resources, data access, or running untrusted code. Yet data residency isn't automatic: if Hermes calls external model APIs, context may leave the organization. Actual data boundaries depend on model deployment location, tool connections, credentials, and log configuration.

Memory and Skills: Turning Business Experience into Maintainable Assets

Hermes addresses long-term knowledge retention via three mechanisms:

Persistent memory in MEMORY.md (project conventions, environment, ~2200 chars) and USER.md (user preferences, ~1375 chars), loaded at session start.

Full session history stored in SQLite with FTS5 full-text search , enabling on-demand retrieval via session_search.

Skills — reusable, executable procedures managed by skill_manage. When an agent completes a complex task or refines a method, it can write the workflow to a Skill file for future reuse.

Illustrative scenario: a Data Agent builds an analysis flow for a key metric — query prescribed tables, run validation SQL, check business constraints, emit fixed-format report. The method becomes a Skill; environment details go into Memory. The article emphasizes this is a hypothetical example, not a published customer case, but shows the intended division: Memory stores facts and context; Skills store executable methods.

File-based knowledge is easier to inspect, version, and integrate into existing engineering practices than ephemeral context. However, the article notes Hermes' "continuous learning" relies on external memory and Skill updates — not automatic model-weight training — and more experience doesn't guarantee higher accuracy. Risks include codifying accidental successes or stale information.

To mitigate, Hermes provides write-approval gates : enabling skills.write_approval or memory.write_approval puts changes in a pending state for human review before application. This moves experience accumulation into a governed workflow, letting organizations maintain critical steps rather than relying solely on a vendor's memory feature. Still, versioning, collaboration, permission inheritance, and rule updates require additional design; the current mechanisms are a foundation, not a complete governance system.

Permissions and Safety: Engineering Guardrails for Real Business Integration

Hermes connects to external tools via MCP (Model Context Protocol) , exposing databases, filesystems, GitHub, internal APIs. Developers control which tools the model sees, narrowing the agent's operational scope. Official security docs also list: messaging-platform user allowlists, DM pairing, high-risk command approval, file-write restrictions, container isolation, MCP subprocess env-variable filtering.

Boundaries are explicit: file-tool path restrictions don't cover shell access if the agent has host terminal rights; limiting visible tools doesn't prevent misuse if a tool itself holds excessive business permissions. The article stresses that enterprises must balance capability with risk using tool permissions, human approval, runtime isolation, and execution logs matched to actual threat models. Open-source runtimes allow inspection and custom execution policies, but don't automatically yield a full enterprise security posture — identity management, auditing, disaster recovery still need concrete implementation.

Business Model and Enterprise Fit

Nous Research's commercialization logic: open-source version attracts developers building apps and workflows; enterprise product monetizes deployment, security, org management, and ongoing services. Details remain limited; the article warns not to treat potential commercial capabilities as already delivered.

Enterprise selection criteria:

Managed assistants suit teams wanting quick, low-maintenance general-purpose help.

Open Agent Runtimes like Hermes fit organizations with extensive internal APIs, data systems, complex workflows, desire for model flexibility, and in-house deployment/maintenance capacity.

Greater control brings greater responsibility: self-managed deployment, monitoring, failover, security auditing, upgrades, Skill governance. Model inference costs may drop, but total engineering cost may not decrease proportionally.

No detailed, independently verifiable enterprise production case studies exist yet. Download/clone/token metrics don't substitute for task success rates, long-term stability, or total cost of ownership. The article recommends a practical validation: pick one real internal workflow, run it under identical permissions and task conditions, and compare completion rate, failure recovery, model-switch cost, and maintenance effort against the current solution.

Outlook

The $90M gives Nous Research runway to execute. Their enterprise value proposition is straightforward: let organizations continuously own their agent runtime, preserving models, tools, and hard-won experience through technology shifts. The next proof points aren't download counts but how many enterprises trust Hermes with real business processes long-term, and whether those deployments prove reliable, controllable, and economical.

References: 1. The Wall Street Journal, "Nous Research Scores $90 Million to Bring Open-Source AI Assistant to Enterprises", Oct 2026. 2. Nous Research / Dillon Rolnick, "A Note on Our Fundraise", Oct 2026. 3. TechCrunch / Marina Temkin, "Nous Research confirms it hit $1.5B valuation, launches AI agents for business users", Oct 7, 2026. 4. Nous Research, Hermes Agent — Official GitHub Repository.