Author: 0xjacobzhao

Opening the public application ranking of OpenRouter, Hermes Agent ranks first across the platform with a token usage of 30.5 trillion, while also ranking first in the categories of Productivity, Coding Agents, Personal Agents, and CLI Agents, significantly ahead of well-known agents like OpenClaw and Claude Code.

Figure 1 · Historical data snapshot of Hermes Agent on OpenRouter (captured on August 4, 2026, dynamic page data may change over time)

Although OpenRouter's statistical scope cannot cover the entire industry's token consumption from directly connecting to official APIs (such as native subscriptions for Claude or Codex), as the current largest AI large model routing and aggregation platform globally, its rankings have strong "barometer" significance. While in high-end professional tasks, the core business workflows of many users—complex code generation, architecture design, high-value data analysis—still flow towards Claude Code and ChatGPT, Hermes maintains an advantage in scenarios such as backend automation, message entry response, long-term online monitoring, and lightweight task scheduling. As an agent product developed by a Web3 team, Hermes has achieved far beyond expected success in dissemination, community, and usage intensity, prompting us to pay attention to:

• Why has Hermes managed to surpass in OpenRouter inference calls?

• What is the real distinction between it and OpenClaw?

• How does Hermes maintain "differentiated coexistence" rather than "direct competition" with Claude Code and Codex?

I. From Development Framework to Personal AI System—The Path of OpenClaw

Why early agent frameworks did not produce consumer products

Before the emergence of OpenClaw, although the agent field had mature infrastructure, it had fundamental limitations: its unit of adoption was "developer project enterprise workflows," rather than "individual users." The common feature of early frameworks was that they were developer-oriented, outputting code or configurations—they built the infrastructure for agents but did not deliver the agents themselves. The high engineering threshold kept them stuck in the "developer tool" stage, lacking a productization closed loop that could transform technology into "personal exclusive assets," leaving the "personal agent product layer" aimed directly at end users almost blank.

Figure 1 · Six-layer structure of the agent technology stack (Model Layer → Protocol Layer → SDK Development Framework Layer → Orchestration Runtime Layer → Execution Infrastructure Layer → Deployment Governance Layer)

What has OpenClaw truly changed

OpenClaw did not reinvent the agent loop or task scheduling technology at the underlying level; its core contribution lies in the systematic packaging at the product level. LangChain addresses "how to build an agent," while OpenClaw addresses "how to own an agent." It skips the intermediate layers of the tech stack, integrating scattered framework capabilities into a complete product that individuals can directly configure and use long-term, achieving a fundamental shift in the unit of adoption from "development projects" to "individuals," specifically reflected in six dimensions of product innovation:

• Identity personalization: Giving agents continuous names and identities, breaking the tool-like feeling of stateless API calls.

• Entry normalization: Using high-frequency communication software like Telegram/WhatsApp as the interaction interface, replacing complex command lines or IDEs.

• State permanence: Running as a background process online for a long time, achieving a leap from passive "standby" to active "presence."

• Permission materialization: Deeply incorporating users' file systems, browsers, terminals, and real-world action capabilities into the operational boundaries of the agent.

• Capability scalability: Through Skills, Memory, and community plugins, sedimenting processes into reusable capabilities and expanding action boundaries.

• Mental ownership: The most core transformation—users shift from "using an AI tool" to "owning a dedicated digital partner."

Why the lobster craze did not form a second mindset

The popularity of OpenClaw has spawned numerous imitations. These products address real user problems: cumbersome installation processes, difficult environment configurations, the absence of channels like WeChat and Feishu, compatibility with domestic models, rapid deployment of cloud hosts, enterprise permission management, automatic updates, and security isolation, among others. They each have their user bases and reasonable business logic. However, almost none have formed an independent brand mindset—the reason being that they answer the question of "how to use OpenClaw more easily," rather than "where should personal agents evolve after OpenClaw." The narrative challengers are extremely scarce in the entire personal agent market.

II. Why Hermes Ultimately Surpassed

Nous's model, community, and crypto-native background

Nous Research originated from a Discord open-source AI research community in 2022 and officially completed its corporate operations in 2023. The core founding team includes Jeffrey Quesnelle, Karan Malhotra, Teknium, and Shivani Mitra, with business covering:

• Hermes model series: The most representative open-source model brand of Nous, focusing long-term on model post-training, instruction fine-tuning, and agent capabilities, establishing a large developer adoption base on Hugging Face.

• DisTrO (Distributed Training Over-the-Internet): Significantly reducing cross-node communication overhead in distributed training, making it more feasible for cross-regional, heterogeneous hardware to participate in collaborative training under internet bandwidth conditions.

• Psyche decentralized training network: Further networking DisTrO, coordinating global distributed computing nodes through Solana, allowing GPUs from different networks and hardware environments to jointly participate in large model training.

• Hermes Agent: A personal agent product launched by Nous for end users, integrating the Hermes model, tool calls, Memory, Skills, messaging channels, and long-term operational capabilities into a resident agent.

In April 2025, Nous Research completed a $50 million Series A financing led by Paradigm, with a post-investment token valuation reaching $1 billion. Prior to this round of financing, the company had cumulatively completed about $20 million in early financing, with investors including well-known institutions such as Distributed Global, North Island Ventures, and Delphi Digital.

Nous has built a technical closed loop of "Hermes (model capabilities), DisTrO (distributed training), Psyche (decentralized computing network), and Hermes Agent (personal terminal product)." The release of Hermes Agent is not a temporary fork chasing hype, but a strategic extension initiated by Nous to the demand side (real users, tasks, workflows) after long-term sedimentation on the supply side (data, models, training, open weights)—this provides a deeper starting point for establishing differentiation than ordinary imitations.

The operational pain points of OpenClaw have become the growth engine for Hermes

The underlying encapsulation of Hermes and OpenClaw (model + tools + Memory + scheduling) is not significantly different. Its phenomenal explosion does not rely on technological generational differences but precisely closes a systematic growth causal chain: by seamlessly migrating tools to directly take over users who have been educated and tormented by operational pain points from OpenClaw, forming the early core growth engine.

Product power leap: Establishing "Delegation Trust"

The core product hypothesis of Hermes is to solve "the transfer of operational responsibility," promising "errors will be absorbed and repaired by the system internally":

• Reliability trust: Ensuring continuous task advancement and failure recovery (persistent Kanban, /goal mode, tool self-healing).

• Security trust: Preventing overreach, accidental deletions, or data leaks (Approvals approval flow, sandbox, strict permission boundaries).

• Verifiable trust: Proving that tasks are truly completed (Completion Contract and Grounded Citations).

Conceptual distinction: "Self-evolution" (narrative advantage) vs "Autonomous recovery" (experience difference)

In Hermes's product narrative, there is a significant difference in product value between "self-evolution" and "autonomous recovery":

• Self-evolution: Essentially process adaptation based on Memory and Skills. Given that competitors have similar infrastructure, differentiation lies more in being the first to integrate into a default system with lifecycle management, occupying a narrative advantage of "will grow," rather than a proven, insurmountable technological barrier.

• Autonomous recovery: This is the most worth validating experience difference currently. Thanks to structured error returns and Provider automatic fallback, Hermes can digest faults within the system. This "infrequent disturbance to users" system-level stability is a more direct, perceivable product power difference.

Architectural dividends: Delegation and supervision capabilities for professional agents

The core value of Hermes lies not in personally executing all professional tasks but in serving as the control layer (Orchestrator) responsible for demand completion, task decomposition, routing monitoring, and final acceptance. By delegating specific work to external CLIs like Claude Code/Codex through built-in Skills, the community has sedimented a practice paradigm of "Hermes control + external CLI as Worker" (such as the /goal mechanism and oh-my-hermes collaborative tools), reflecting its architectural advantage in raising the upper limit of task complexity through scheduling professional agents.

From Crypto-native to Crypto-invisible: The Web3 backend operating system of Hermes

Attributing Hermes's success simply to a "Web3 background" is an oversimplification. Web3 provides Nous with an "organizational operating system" that is difficult for other AI startup teams to obtain simultaneously, allowing it to enter the mainstream market with a smooth experience of standard AI products:

• Patience of venture capital: Crypto-native capital supports long-term, high uncertainty, and multi-route parallel investments, enabling Nous to layout models, training, Runtime, and Cloud simultaneously without prematurely converging on a single revenue verification.

• Ready user market: Providing a familiar user base of Crypto AI users who understand Telegram, servers, APIs, and self-hosting, significantly reducing cold start education costs and fostering high-intensity usage, tutorial dissemination, and Skills contributions.

• User sovereignty values: Upholding self-hosting, openness, portability, and anti-platform lock-in orientations, directly implemented as MIT License, multi-Provider support, BYOK, and Memory/Skills portable underlying architecture.

• Community R&D and verticalization: Relying on global remote collaboration and open-source culture, users spontaneously become Contributors, Skill authors, and product designers for vertical scenarios.

Hermes has almost entirely concealed Crypto from the user interface. Using its Agent, Memory, Skills, and automation capabilities does not require connecting wallets, purchasing tokens, or understanding Solana. Meanwhile, Paradigm capital, Psyche, distributed training, and the Crypto AI community still exist in the product backend. This forms a product form that can be summarized as "Crypto-native in organization, crypto-invisible in product"—retaining the most valuable parts of Crypto at the organizational level (capital, global community, user sovereignty, and coordination capabilities) while removing the parts that most easily hinder mainstream adoption at the product level (wallets, tokens, speculative narratives, and on-chain operational friction).

Why OpenClaw rejects Crypto, and why Hermes hides Crypto

The apparent opposition between OpenClaw and Hermes on the Crypto issue is not an ideological struggle of "rejection" versus "embrace"; from the product results, both reflect orientations of open-source, user control, and reducing platform lock-in; the difference lies in how Nous further applies the crypto-economic mechanism for distributed training coordination, while OpenClaw mainly achieves user sovereignty through a local-first architecture:

• OpenClaw (Local-first Sovereignty): Resisting financial speculation, defending "local-first" sovereignty. Due to early encounters with counterfeit scams, it adopts a "zero tolerance" approach to Crypto. By being purely open-source and running locally, it achieves a non-blockchain-style defense of user sovereignty, firmly rejecting financialization at the product level.

• Hermes/Nous (Cryptoeconomic Sovereignty): Engineering-oriented, Crypto serves only as a foundational coordination tool. Introducing blockchain is a pragmatic choice to address engineering challenges (such as the Psyche network utilizing Solana to coordinate heterogeneous computing power), rather than constructing a financial narrative aimed at end users.

III. The Advanced Model of Hermes—From Personal Agent to Task General Manager

This section aims to answer a more fundamental question: When Claude Code and Codex can already complete most professional execution tasks with high quality, what is the reason for the existence of Hermes as an independent product?

• Model A: Direct collaboration type (limited gain): Users are accustomed to manually generating prompts in LLM and handing over execution, manually transporting results and reviewing. Although the quality of a single output is high, they must bear all project management and multi-agent coordination work. For such hands-on users, Hermes's automation is seen as "an intermediate layer that increases opacity," failing to effectively reduce their burden.

• Model B: Delegated management type (obvious gain): Users treat Hermes as a resident control layer, only issuing final goals. Hermes is responsible for task decomposition, delegating sub-tasks, tracking GitHub/CI status, and automatically triggering rework. Community practices (such as oh-my-hermes) show that the core value of Hermes is precisely to replace cumbersome cross-agent coordination and project management work.

First Layer · User Goals and Governance

Users define final goals, budgets, risk boundaries, and approval authority

↓

Second Layer · Hermes Task Control Layer

Demand completion / Planning decomposition / Cross-session Memory / Agent selection and delegation / Progress supervision / Result acceptance / Initiating rework

↓

Third Layer · Professional Execution and Infrastructure

Claude Code / Codex / Other vertical CLIs · GitHub / VM / Browser / Database / Cloud

In this framework, Hermes and Claude Code/Codex are not in a substitutive relationship but in a layered relationship: the latter provides execution quality at the third layer, while the former provides continuity, cross-session state, and cross-agent coordination at the second layer. The value of Hermes is not evenly distributed among all users but may be highly concentrated among advanced user groups engaged in cross-agent, cross-system, long-term asynchronous tasks. This judgment is more precise than the vague assertion that "the second mindset of personal agents has formed" and is more suitable for guiding commercialization and product priorities.

Figure 2 · Overview of Hermes Agent's technical architecture (User Entry → Gateway → Control Core → Provider Layer → Execution Layer → Orchestration Layer → State Layer → Governance Layer)

Based on official documentation and community research, the panoramic framework of Hermes Agent's technical architecture covers the entire link from user interaction to learning governance:

• System-level support for autonomous recovery: The "Control Core" clearly includes Context compression, Provider Fallback, and interruption state preservation, providing the technical foundation for fault recovery and system self-healing capabilities in case of task failure.

• "Delegation rather than substitution" execution logic: The "Tool and Professional Execution Layer" places external CLIs like Claude Code and Codex alongside Hermes native tools (Terminal, Browser, etc.), confirming its positioning as a scheduling hub.

• "Self-evolution" governance attributes: The "Learning, Maintenance, and Governance Layer" includes nodes like Curator and Skill/Command Approval, indicating that its experience sedimentation is a governance process with human intervention mechanisms, rather than a fully automated black box.

IV. Business Model—Who Pays for "Hermes"?

If we only compare its token costs with direct subscriptions to Claude Code/Codex, we would reach misleading conclusions. This algorithm ignores the core value of Hermes: replacing users' hands-on project management, context transportation, and cross-agent coordination work.

User value formula: Hermes user value = Saved manual coordination time + Asynchronous and unattended value + Cross-system automation benefits − Token and tool costs − Manual intervention costs − Failure and security risks

Therefore, the economics of Hermes is not absolute but highly dependent on the user's "delegation depth":

• High delegation depth (economics established): If Hermes can convert tasks that originally required hours of manual monitoring into truly unattended execution, even if the token cost is slightly higher, its overall time cost and efficiency gains remain positive.

• Low delegation depth (economics collapse): If users still need to frequently intervene to correct errors and put out fires, Hermes becomes merely a token consumer and fault magnifier.

This mechanism precisely explains why different user groups have starkly contrasting evaluations of Hermes's economics and suggests that validating its business logic hinges on quantifying "unattended completion rates" and "manual intervention counts per task," rather than simply comparing the unit price of model APIs.

Commercialization foundation: Nous Portal and Hermes Cloud

Hermes Agent is open-sourced under the MIT license, positioned as an ecological growth engine. The true commercialization closed loop focuses on Nous Portal, whose core value proposition is "one subscription, integrating multiple types of API keys," covering three major modules:

• Model routing: Aggregating 252 models (providing inference through OpenRouter and direct connection to providers).

• Tool Gateway: Built-in high-frequency tools such as Firecrawl (web search), FAL (image generation), Browser Use (cloud browser), Modal (sandbox execution), and OpenAI Audio (TTS).

• Hosting services: Out-of-the-box Hermes Cloud instances (charged daily for operation, excluding inference and tool call fees).

Nous's actual revenue highly depends on user usage paths, currently showing significant structural differentiation:

Open-source and commercialization, will Hermes become the "Linux of the Agent world"?

Hermes's MIT open-source strategy drives explosive growth while also constituting a structural constraint on commercialization. The self-hosted free model requires its paid version to provide irreplaceable additional value, but a clear differentiated monetization path has not yet formed. A deeper risk lies in "value capture": if Hermes continues to be widely integrated as an optional runtime by cloud vendors, it may replay the classic dilemma of Linux or K8s, where the core commercial value is captured by cloud vendors providing computing power and hosting. The MIT license, while facilitating ecological prosperity, also means relinquishing absolute control over distribution channels. As long as users can freely choose "self-hosting + own API" or "third-party cloud deployment," the vast usage cannot be forcibly converted into direct income, putting Nous in a severe test of "ecological position enhancement" versus "actual commercial returns mismatch."

V. Agent Ecological Position—The Triangular Pattern of Personal General Manager, Professional Tools, and Big Company Claw

OpenClaw, Hermes, Claude Code, Codex, and large company hosting products have significant differences in target users and core propositions, belonging to different sub-tracks. To clarify the current market landscape, the core competitive matrix of AI agents is as follows:

Hermes does not pursue a broad mass market but precisely targets four types of high-density power users, forming the cornerstone of its phenomenal dissemination:

• Self-hosting and infrastructure players: Familiar with VPS/Docker/SSH, viewing Hermes as a natural control layer for existing infrastructure.

• Multi-model arbitrageurs: Rejecting single vendor lock-in, accustomed to dynamically scheduling cutting-edge or local models based on tasks.

• Multi-agent coordinators: Urgently needing to automate the orchestration of complex workflows across platforms and tools.

• Open-source and Crypto AI community: Highly resonating with user sovereignty and decentralization concepts, deeply aligning with Nous's organizational culture.

Although the base of these groups is small, they possess extremely high token consumption, code contribution, and technical evangelism capabilities, serving as the core engine driving early word-of-mouth dissemination.

Claude Code/Codex: Both a supplier and a threat

Short-term symbiosis: Raising execution limits

In actual workflows, Hermes serves as the control layer, calling Codex (code implementation) and Claude Code (architecture and review) through delegation mechanisms. The stronger the underlying professional agents, the higher the task complexity limit Hermes can deliver, forming a symbiotic relationship where "Hermes is responsible for routing and acceptance, while professional agents are responsible for execution."

Long-term potential to absorb Hermes's independent value:

Model vendors are accelerating their penetration into the control layer, posing a threat closer than expected. Anthropic's Claude Managed Agents now support multi-agent parallel orchestration; OpenAI has explicitly positioned Codex App as the "command center for agents," supporting multi-agent parallelism, automation, and long-cycle background operation. This means Codex's multi-agent control capabilities within the software engineering boundary are relatively mature, even surpassing Hermes in some areas, no longer merely a "bottom-level executor."

Hermes currently holds an advantage in personal control planes across channels, models, and projects; however, Codex already possesses strong task ownership and multi-agent management capabilities within the software engineering boundary, which may be stronger than Hermes in that boundary. The core competitive question is: Can Hermes, ahead of model vendors, sediment users' project states, approval rules, Skills, Memory, and cross-agent workflows at its layer, forming assets that users are reluctant to migrate? Or will it ultimately be absorbed as a standard feature by model-native products?

Internet giants' choices on the agent route

Discussing how large companies should respond to the personal agent wave requires first clarifying their product boundaries: the positioning of resident agent hosting aimed at individuals (such as Tencent's QClaw, Byte's ArkClaw) is distinctly different from general work agents aimed at office/enterprise (such as WorkBuddy, Trae):

• Large company Claw route: Lowering the threshold through one-click deployment, preset templates, and local ecosystem access. However, the deeper gap lies in the untrustworthiness of platform incentives: regardless of how many external models are supported, users inherently believe that the ultimate goal is to funnel them into their own cloud and model systems.

• Hermes Runtime integration: Byte's ArkClaw and Tencent Cloud have officially integrated Hermes Agent as an optional plugin or exclusive template into their cloud control consoles, establishing a clear multi-runtime strategy: large companies retain their cloud hosting, billing, security, and enterprise-level control foundations while viewing Hermes as a pluggable advanced component, achieving complementary coexistence between the open-source ecosystem and commercial cloud platforms.

• General office agent transformation: Currently, large companies are shifting core resources from Claw to general office agent platforms with clear demand, easy acceptance, and direct monetization potential (such as WorkBuddy). Such tasks can be deeply bound to their own ecosystems like WeChat, DingTalk, and Feishu, converting into revenue.

VI. Insights from Hermes on Crypto AI

Web3 did not directly make Hermes a smarter agent but provided Nous with a capital structure, organizational approach, seed user base, and value source different from traditional AI startups. Hermes at least proposes a more mature Crypto AI path: making Crypto an organization and infrastructure, rather than a product interface that users must face.

Hermes has completed the transition from a Crypto AI research brand to a global open-source agent product, establishing large-scale attributable inference activities and a clear second mindset—but this mindset is currently still concentrated in the OpenRouter ecosystem and the global developer circle, not yet translating into GitHub Stars or a comprehensive surpassing of OpenClaw in overall community scale. It does not possess exclusive technologies that OpenClaw cannot replicate but has completed a noteworthy challenger product iteration by precisely engaging high-intensity users and establishing "delegation" and "self-evolution."

• Insight One: Crypto can serve as an "organizational operating system," rather than a product function: The true value of Web3 can be reflected in capital structure, early high-intensity user pools, and value base, without being forcibly exposed as wallet or token interactions. Achieving "Crypto-native at the organizational level, crypto-invisible at the product level" is an effective strategy balancing innovation drive and user experience.

• Insight Two: Decentralized infrastructure must anchor demand-side entry to form a closed loop: Purely supply-side distributed training networks (like DisTrO, Psyche) without real user entry and execution data support struggle to self-validate commercial value. Hermes Agent is precisely the key validation for Nous's leap from foundational computing infrastructure to real demand-side.

• Insight Three: Moats can be built on "delegation trust" rather than merely "model capability": The differentiation of personal agents may not stem from stronger single execution capabilities but from whether "users dare to entrust long-term responsibilities to it." This soft trust asset is a dimension often overlooked but highly barrier-creating in Crypto AI projects.

• Insight Four: The relationship with cloud giants is not a zero-sum game but ecological complementarity: Large companies integrating Hermes as an optional runtime proves that open-source runtimes and large company control planes can coexist. For entrepreneurs, "being integrated" is a viable commercialization path, but they must be wary of the risk of core value being captured by cloud vendors' hosting layers.

• Insight Five: The ultimate competition will shift from "single execution capability" to "task ownership and trust accumulation": The most valuable in the future may not be the strongest model at the execution layer but rather the "upper control system" that can receive final goals, maintain long-term context, intelligently schedule professional executors, and allow users to confidently entrust responsibilities.

OpenClaw has made "personal ownership of agents" a clear product category; Hermes, through persistent state, task recovery, evidence acceptance, multi-model supply, and professional agent delegation, has advanced "long-term entrusted agents" into a more systematic product direction. The real test is: when Claude Code and Codex's control capabilities within the software engineering boundary continue to strengthen, and large company cloud platforms make multi-runtime integration smoother, will users still be willing to entrust their final goals and long-term trust to this open runtime from a Web3 background—and continue to pay for it?

Disclaimer: This article utilized AI tools such as Claude Opus 5, ChatGPT-5.5, and Qwen 3.7 during its creation. The author has made efforts to proofread and ensure the information is true and accurate, but there may still be omissions, for which we ask for your understanding. It should be particularly noted that the content of this article is intended for information integration and academic/research communication, does not constitute any investment advice, and should not be viewed as a recommendation for buying or selling any tokens.