IOSG Weekly Brief | The Rise of Hermes: An Advanced Journey of a Web3 Team #340

Original Title: "IOSG Weekly Brief | The Rise of Hermes: An Advanced Web3 Team's Journey #340"
Original Author: IOSG Ventures

The phenomenal growth of Hermes did not stem from exclusive technology based on the OpenClaw principle that cannot be replicated. Instead, it was due to closing a critical window of opportunity in the personal Agent category with the most precise "Challenger Growth System": inheriting the already educated user pool from OpenClaw and establishing "Delegation Trust" as a more authentic experiential difference than the "self-evolution" narrative. As professional execution Agents become stronger, users still need a long-term online, trustworthy steward.

Upon opening the Public Application Leaderboard of OpenRouter, the Hermes Agent ranked first on the platform with a Token usage of 30.5 trillion, while also leading in the Productivity, Coding Agents, Personal Agents, and CLI Agents categories, far surpassing well-known Agents such as OpenClaw and Claude Code.

▲ Figure 1: Historical Snapshot of Hermes Agent on OpenRouter (as of August 4, 2026, dynamic page data subject to change over time)

Although OpenRouter's statistical methodology cannot cover the industry-wide Token consumption from direct official APIs (such as Claude or Codex native subscriptions), as the world's largest AI large-scale model routing and aggregation platform currently, its leaderboard holds significant "guiding" significance. Despite the fact that in the realm of high-end professional tasks, the core business workflows of a large number of users—complex code generation, architectural design, high-value data analysis—still flow towards Claude Code and ChatGPT, Hermes maintains an advantage in backend automation, message entry response, long-term online monitoring, and lightweight task scheduling use cases. As an Agent product built by a Web 3 team, Hermes has achieved far greater than expected dissemination, community engagement, and usage intensity, prompting us to ask:

· How did Hermes manage to surpass others in reasoning calls on OpenRouter?

· What is the real distinction between it and OpenClaw?

· In the relationship between Claude Code and Codex, how does Hermes maintain "differentiated coexistence" rather than "head-on competition"?

From Development Framework to Personal AI System: The OpenClaw Journey

Why Early Agent Frameworks Did Not Yield Consumer Products

Before OpenClaw emerged, the Agent field already had mature infrastructure but faced a fundamental limitation: it was geared towards the "development project enterprise workflow" rather than the "individual user." Early frameworks were characterized by targeting developers, 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 the productization loop to transform the technology into "personal proprietary assets," leaving the "personal Agent product layer" directly targeting end-users nearly blank.

▲ Figure 1: Six-layer structure of Agent Technology Stack (Model Layer → Protocol Layer → SDK Development Framework Layer → Orchestration Runtime Layer → Execution Infrastructure Layer → Deployment Governance Layer)

▲ Figure 1: Hermes Agent in OpenRouter's historical data snapshot (taken on August 4, 2026, dynamic page data will change over time)

What OpenClaw Truly Changed

OpenClaw did not reinvent the Agent Loop or task scheduling technology at the core but made a systemic encapsulation at the product level. Where LangChain addressed "how to build an Agent," OpenClaw tackled "how to own an Agent." It skipped the mid-layer of the technology 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 project" to "personal." This is specifically reflected in six dimensions of product innovation:

· Identity Personification: Endow the Agent with a continuous name and identity, breaking the tool-like feel of stateless API calls.

· Entry Normalization: Using high-frequency communication tools such as Telegram/WhatsApp as the interaction interface, replacing complex command lines or IDEs.

· State Persistence: Running as a background process for a long time, achieving a transition from passive "standby" to active "presence."

· Permission Entification: Deeply integrating the user's file system, browser, terminal, and real-world actions into the Agent's operational boundary.

· Capability Scalability: By using Skills, Memory, and community plugins, solidify processes into reusable capabilities to expand the scope of actions.

· Mental Ownership: The most core transformation—the user shifts from "using an AI tool" to "owning a dedicated digital companion."

Why Lobster Hotpot did not form a second mindshare

The popularity of OpenClaw has spawned numerous imitations. These products have solved real user problems: the cumbersome installation process, the difficulty of environment configuration, the lack of channels such as WeChat and Feishu, compatibility with domestic models, rapid deployment of cloud servers, enterprise permission management, automatic updates, and security isolation, among others. They each have their own user base and reasonable business logic. However, almost none of them have formed an independent brand mindshare—the reason being that they answer the question of "how to use OpenClaw more easily" rather than "where should the personal Agent evolve after OpenClaw." The position of a narrative challenger is extremely scarce in the entire personal Agent market.

Why Hermes ultimately prevailed

The Model, Community, and Crypto-native Background of Nous

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

· Hermes Model Series: Nous's most representative open-source model brand, focusing long-term on post-training of models, instruction fine-tuning, and Agent capabilities, establishing a large developer adoption base on Hugging Face.

· DisTrO (Distributed Training Over-the-Internet): By significantly reducing the cross-node communication overhead of distributed training, it dramatically decreases the cross-node communication requirements in distributed training, making it a more feasible engineering path for cross-regional, heterogeneous hardware to participate in collaborative training over the internet.

· Psyche Decentralized Training Network: It further networks DisTrO by coordinating global distributed computing nodes through Solana, allowing GPUs from different networks and hardware environments to participate in large-scale model training.

· Hermes Agent: A personal Agent product launched by Nous for end-users, integrating the Hermes model, tool invocation, Memory, Skills, messaging channels, and long-running capabilities into a resident Agent.

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

Nous has constructed a technical loop of "Hermes (model capability), DisTrO (distributed training), Psyche (decentralized computing power network), and Hermes Agent (personal end-product)." The release of the Hermes Agent is not a temporary fork chasing hype but a strategic extension initiated by Nous on the demand side (real users, tasks, workflows) after long-term precipitation on the supply side (data, model, training, open weights) – providing a deeper starting point for differentiation compared to a common mimicry.

OpenClaw's Operational Pain Points Turned into Hermes' Growth Engine

The difference between Hermes and OpenClaw at the foundational level (model + tool + Memory + scheduling) is not significant. Its explosive growth is not dependent on technological delta but has accurately closed a systematic growth causality chain: by seamlessly migrating tools to directly address OpenClaw's already educated and operationally troubled user base, it has formed the most core growth engine in the early stages.

Productivity Leap: Establishing Delegation Trust

The core product assumption of Hermes is to address "Operational Responsibility Shift," promising "internal system absorption of errors for post-fault remediation":

· Reliability Trust: Ensuring seamless task progression and failure recovery (persistent Kanban, /goal mode, tool self-healing).

· Security Trust: Preventing unauthorized access, accidental deletion, or data leakage (Approvals workflow, sandboxing, strict permission boundaries).

· Verifiability Trust: Proving task completion authenticity (Completion Contract and Grounded Citations).

Conceptual Analysis: "Self-Improvement" (Narrative Advantage) vs "Autonomous Recovery" (Experience Discrepancy)

In Hermes' product narrative, there is a significant distinction in the product value between "Self-Improvement" and "Autonomous Recovery":

· Self-Improvement: Essentially based on Memory and Skills-driven process adaptability. Given that competitors offer similar infrastructure, the differentiation lies more in being the first to integrate into a lifecycle-managed default system, occupying a narrative advantage of "will grow" mentality rather than a proven, insurmountable technical barrier.

· Autonomous Recovery: This is currently the most compelling experience difference. Benefiting from structured error feedback and Provider automatic Fallback, Hermes can internalize faults within the system. This system-level stability of "minimal user disruption" represents a more direct, perceptible productivity difference.

Architectural Dividend: Delegation and Supervision Capabilities for Specialized Agents

The core value of Hermes lies not in personally executing all specialized tasks but in acting as the orchestrator layer to undertake requirement completion, task decomposition, route monitoring, and final acceptance. By delegating built-in Skill to external CLI like Claude Code/Codex for specific work execution, the community has established the practice paradigm of "Hermes Orchestrator + External CLI as Worker" (such as the /goal mechanism and collaborative tools like oh-my-hermes), showcasing its architectural advantage of elevating the task complexity threshold through scheduling specialized agents.

From Crypto-native to Crypto-invisible: Hermes' Web 3 Underlying Operating System

Attributing Hermes' success solely to its "Web 3 background" is an oversimplification. Web 3 has provided Nous with an "organizational operating system" that other AI startups find challenging to access simultaneously, enabling it to enter the mainstream market with a seamless AI product experience:

· Venture Capital Patience: Crypto-native funding supports long-term, high uncertainty, and multi-path parallel investment, allowing Nous to simultaneously develop models, training, Runtime, and Cloud without prematurely converging on a single revenue validation.

· Ready-made User Base: It has offered a familiar Telegram, server, API, and self-hosted Crypto AI user community, significantly reducing the cold start education cost and fostering high-intensity usage, tutorial dissemination, and Skills contribution.

· User Sovereignty Values: Upholding a self-hosted, open, migratable, and anti-platform lock-in orientation, directly translated into an MIT License, multi-provider support, BYOK, and Memory/Skills-migratable underlying architecture.

· Community-Led Development and Verticalization: Leveraging global remote collaboration and open-source culture, users spontaneously become Contributors, Skill authors, and vertical scenario product designers.

Hermes has almost shielded Crypto from the user's front end. Through its Agent, Memory, Skills, and automation capabilities, there is no need to connect a wallet, purchase tokens, or understand Solana. Meanwhile, Paradigm Capital, Psyche, distributed training, and the Crypto AI community still exist in the product's backend. This has created 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) and removing the most significant barriers to mainstream adoption at the product level (wallets, tokens, speculative narratives, and on-chain operational friction).

Why OpenClaw Shuns Crypto and Why Hermes Conceals Crypto

OpenClaw and Hermes do not have a surface-level opposition on Crypto issues, as it is not a matter of "rejection" versus "embracement" ideology. From a product perspective, both demonstrate a focus on open source, user control, and reducing platform lock-in. The difference lies in Nous further applying cryptoeconomic mechanisms to distributed training coordination, while OpenClaw mainly achieves user sovereignty through a Local-first architecture:

· OpenClaw (Local-first Sovereignty): Resists financial speculation, upholds "local-first" sovereignty. Due to early encounters with counterfeit coins, it takes a "zero-tolerance" approach to Crypto. Through pure open source and local operation, it defends user sovereignty in a non-blockchain manner, firmly rejecting the financialization at the product level.

· Hermes/Nous (Cryptoeconomic Sovereignty): Engineering-oriented, Crypto serves solely as a fundamental coordination tool. The introduction of blockchain is a pragmatic choice to address engineering challenges (e.g., the Psyche network utilizing Solana to coordinate heterogeneous computing power) rather than to construct a financial narrative for end-users.

Hermes's Advanced Mode - From Personal Agent to Task Steward

This section aims to answer a more fundamental question: when Claude Code and Codex are already capable of high-quality completion of most professional tasks, what is the rationale for Hermes as a standalone product?

· Mode A: Direct Collaboration (Limited Gain): Users are accustomed to manually generating prompts in LLM and handing over execution, manually handling result transfer and review. While the quality of a single output is high, users need to take on all project management and multi-agent coordination work. For users who are hands-on, Hermes's automation is seen as adding an "opaque intermediate layer," failing to effectively reduce the burden.

· Mode B: Delegated Management (Significant Gain): Users view Hermes as a permanent controller, only setting the final goal. Hermes is responsible for task breakdown, delegation of subtasks, tracking GitHub/CI status, and automatically triggering rework. Community practices (such as oh-my-hermes) demonstrate that Hermes's core value lies in replacing tedious cross-agent coordination and project management tasks.

Within this framework, Hermes and Claude Code/Codex do not have a substitute relationship but a hierarchical one: the latter provides the quality of execution on the third layer, while the former provides continuity on the second layer, cross-session state, and cross-Agent coordination. The value of Hermes may not be evenly distributed among all users but may be highly concentrated in a sophisticated user group that deals with cross-Agent, cross-system, long-term asynchronous tasks. This judgment is more precise than the broad "the second mind of the personal Agent has formed" and is also more suitable for guiding commercialization and product priorities.

▲ Figure 2 · Hermes Agent Technical Architecture Overview (User Entry → Gateway → Control Core → Provider Layer → Execution Layer → Orchestration Layer → State Layer → Governance Layer)

Based on official documentation and community research, the Hermes Agent technical architecture overview framework covers the full cycle from user interaction to learning governance:

· Autonomous Recovery of System-level Support: The "Control Core" clearly includes Context Compression, Provider Fallback, and interrupt state preservation, providing a technical foundation for fault recovery and system self-healing capability in the event of task failure.

· Execution Logic of "Delegation Rather Than Replacement": The "Tools and Professional Execution Layer" positions external CLIs such as Claude Code, Codex on the same level as Hermes' native tools (Terminal, Browser, etc.), confirming its role as a scheduling hub.

· Governance Attribute of "Self-evolution": The "Learning, Maintenance, and Governance Layer" includes nodes such as Curator, Skill/Command Approval, indicating that its expertise accumulation has a governance process with human intervention mechanisms rather than being fully automatic and opaque.

Business Model — Who Pays for Hermes?

If only the token cost is compared with directly subscribing to Claude Code/Codex, a misleading conclusion will be drawn. This approach overlooks the core value of Hermes: replacing the manual project management, context switching, and cross-Agent coordination.

User Value Formula Hermes User Value = Saved Human Coordination Time + Asynchronous and Unattended Value + Cross-System Automation Benefits − Token and Tool Costs − Human Intervention Costs − Failure and Security Risks

Therefore, the cost-effectiveness of Hermes is not absolute, but highly dependent on the user's "Delegation Depth":

· High Delegation Depth (Economic Viability): If Hermes can transform tasks that would originally require hours of manual monitoring into truly unattended execution, even if the Token cost is slightly higher, the overall time cost and efficiency benefit remain positive.

· Low Delegation Depth (Economic Collapse): If users still need to frequently intervene for error correction and firefighting, Hermes then becomes a pure Token consumer and fault magnifier.

This mechanism precisely explains why different user groups have completely opposite assessments of the cost-effectiveness of Hermes, and also reminds us: the key to validating its business logic lies in quantifying the "unattended completion rate" and "single-task manual intervention frequency," rather than simply comparing the price per model API.

Commercial Foundation: Nous Portal and Hermes Cloud

The Hermes Agent is open source under the MIT License and positioned as an ecosystem growth engine. The true commercial closed-loop focus is on the Nous Portal, whose core value proposition is "one subscription, integrating multiple API keys," covering three main modules:

· Model Routing: Aggregating 252 models (provided through OpenRouter and direct Provider inference).

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

· Hosted Services: Ready-to-use Hermes Cloud instances (charged a daily operational fee, excluding the cost of inference and tool usage).

Nous's actual revenue highly depends on the user's usage path, showing a clear structural differentiation at present:

Open Source and Commercialization: Will Hermes Become the "Linux of Agents"

Hermes's MIT open source strategy has driven explosive growth while also posing a structural constraint on commercialization. The self-hosted free mode requires its paid version to provide irreplaceable additional value, but a clear differentiating monetization path has not yet been established. A deeper risk lies in "value capture": if Hermes continues to be widely integrated by cloud providers as an optional runtime, it may replay the classic dilemma of Linux or K8s, where the core business value is captured by the cloud providers offering compute and hosting. The MIT License, in exchange for ecosystem prosperity, also means relinquishing absolute control over distribution channels. As long as users can freely choose between "self-hosted + proprietary API" or "third-party cloud deployment," the vast user base cannot be forcibly converted into direct income, putting Nous to the severe test of "ecosystem elevation" and "mismatched actual commercial returns."

Agent Ecosystem: Personal Chief Butler, Professional Tooling, and Big Tech Claw's Tripartite Structure

OpenClaw, Hermes, Claude Code, Codex, and Big Tech-hosted products have significant differences in target users and core propositions, belonging to different niche tracks. To clarify the current market landscape, the AI Agent panoramic core competitive matrix is as follows:

Hermes did not pursue the mass market but precisely targeted four categories of high-density Power Users, forming the cornerstone of its phenomenal dissemination:

· Self-Hosting and Infrastructure Players: Familiar with VPS/Docker/SSH, they see Hermes as a natural control layer of existing infrastructure.

· Multi-Model Arbitrageurs: They reject vendor lock-ins, preferring to dynamically schedule cutting-edge or on-premises models based on tasks.

· Multi-Agent Coordinators: They urgently need to automate the complex cross-platform, cross-tool orchestrated workflows.

· Open Source and Crypto AI Community: They highly resonate with user sovereignty and decentralized ideals, deeply aligning with Nous' organizational culture.

Although this group's base number is small, they have high token consumption, code contribution, and technical evangelism capabilities, serving as the core engine driving early word-of-mouth propagation.

Claude Code/Codex: Both a Supplier and a Threat

# Short-term Symbiosis: Elevating Execution Ceilings

In actual workflows, Hermes, as the control layer, calls Codex (code implementation) and Claude Code (architecture and review) through a delegation mechanism. The stronger the underlying professional Agents, the higher the complexity ceiling of tasks Hermes can deliver, forming a symbiotic relationship of "Hermes is responsible for routing and acceptance, while professional Agents are responsible for execution."

# Long-term Potential to Erode Hermes' Standalone Value

Model vendors are rapidly penetrating the control layer, posing a closer threat than expected. Anthropic's Claude Managed Agents already support multi-Agent parallel orchestration; OpenAI even explicitly positions Codex App as a "command center for agents," supporting multi-Agent parallelism, automation, and long-running background processes. This means that Codex's multi-Agent control capabilities within the software engineering boundary are relatively mature, even locally surpassing Hermes, no longer just being the "underlying executor."

Hermes currently holds a cross-channel, cross-model, and cross-project individual control plane advantage; however, Codex has acquired a strong task ownership and multi-agent management capability within the software engineering boundary, which may give it a stronger competitive edge than Hermes. The core competitive question is: Can Hermes, ahead of model vendors, solidify users' project status, approval rules, Skills, Memory, and cross-Agent workflows at its own layer, forming assets that users are unwilling to migrate? Or will it eventually be absorbed as standard functionality by model-native products?

Internet Giant Agent Route Selection

Exploring the strategy of major companies in response to the personal Agent wave requires first clarifying their product boundaries: the positioning of resident Agents for individuals (such as Tencent QClaw, Byte ArkClaw) and universal work Agents for office/corporate use (such as WorkBuddy, Trae) are fundamentally different:

· Giant Claw Route: Reducing barriers through one-click deployment, preset templates, and native ecosystem integration. However, the deep-seated gap lies in the platform's incentivization not being trustworthy: no matter how many external models are supported, users naturally believe that the ultimate goal is to divert traffic to their own cloud and model system.

· Hermes Runtime Integration: Byte ArkClaw and Tencent Cloud have formally integrated the Hermes Agent as an optional plugin or exclusive template into their cloud console, establishing a clear multi-runtime strategy: major companies retain their own cloud hosting, billing, security, and enterprise-level control base, while viewing Hermes as a pluggable advanced component to achieve complementary symbiosis between the open-source ecosystem and commercial cloud platforms.

· Transition to General Office Agents: Currently, major companies are shifting core resources from Claw to general office Agent platforms that have clear requirements, are easy to validate, and can be directly monetized (such as WorkBuddy). These tasks can be deeply integrated with their proprietary ecosystems like WeChat, DingTalk, and Lark and converted into revenue.

Inspiration from Hermes for Crypto AI

Web 3 did not directly make Hermes a smarter Agent, but it enabled Nous to have a capital structure, organizational structure, seed users, and source of values different from traditional AI startups. At the very least, Hermes has proposed a more mature Crypto AI path: making Crypto the organization and infrastructure, rather than the product interface users must face.

Hermes has completed the migration from a Crypto AI research brand to a global open-source Agent product, establishing a large-scale attributable reasoning activity with a clear second mind—but this mind is currently focused on the OpenRouter ecosystem and the global developer community, rather than being transformed into GitHub Stars or an overall community-scale overtake of OpenClaw. It lacks exclusive technology that OpenClaw cannot replicate, but has completed a challenger product iteration worthy of study through precise adoption of high-intensity users, establishing "delegation" and "self-evolution."

· Insight One: Crypto can serve as an "organizational operating system," not just a product feature: The true value of Web 3 can be manifested as capital structure, an early high-intensity user pool, and a values base, without the need to be forcibly exposed as wallet or token interaction. Achieving "organization-level Crypto-native, product-level Crypto-invisible" is an effective strategy that balances innovation drive and user experience.

· Insight Two: Decentralized infrastructure must anchor the demand side entry point to form a closed loop: A purely supply-side distributed training network (such as DisTrO, Psyche) without a real user entry point and execution data support is difficult to prove its commercial value. The Hermes Agent is precisely Nous's key validation of the transition from underlying computing power infrastructure to real demand-side.

· Insight Three: Moats can be built on "delegated trust" rather than just "model capability": The differentiation of individual Agents may not necessarily stem from stronger one-time execution capability, but from whether "users dare to entrust long-term responsibility to them." This soft trust asset is a dimension often overlooked yet highly barriered in Crypto AI projects.

· Insight Four: The relationship with cloud giants is not a zero-sum game, but ecosystem complementarity: The giants have adopted Hermes as an optional Runtime entry, proving that open-source Runtimes can coexist with giant control planes. For entrepreneurs, "being integrated" is a viable path to commercialization, but they need to beware of the risk of core value being intercepted by cloud providers' hosting layers.

· Insight Five: End-game competition will shift from "single-task execution capability" to "task ownership and trust accumulation": The most valuable entities in the future may not necessarily be the strongest models at the execution layer, but those that can receive the ultimate goal, maintain long-term context, intelligently schedule specialized executors, and allow users to confidently delegate responsibility to the "upper-level control system."

OpenClaw has made "individual ownership of an Agent" a clear product category; Hermes has advanced "long-term entrusted Agents" into a more systematic product direction through persistent state, task recovery, evidence acceptance, multi-model provisioning, and professional Agent delegation. The real test is: as Claude Code, Codex enhance their overall control capabilities within software engineering boundaries, and cloud platform giants make multi-runtime integration smoother, will users still be willing to entrust their ultimate goals and long-term trust to this open Runtime from a Web 3 background—and will they continue to pay for it.

Original Article Link