Docs

ArkClaw Enterprise

Hermes Agent

Using Hermes Agent

To enhance the intelligent processing capabilities of your ArkClaw in complex task scenarios, Volcano Engine integrates Hermes Agent into ArkClaw as a runtime capability. Hermes Agent can run directly within ArkClaw and share conversation context and memory with ArkClaw. This allows Hermes Agent to more accurately understand your needs and maintain consistency across multi-turn interactions. While ArkClaw is designed for general assistant scenarios in daily work and life, Hermes Agent is better suited for handling highly repetitive and structured complex tasks. Hermes Agent can also accumulate reusable skills from completed tasks, improving efficiency for similar tasks in the future. The collaboration between ArkClaw and Hermes Agent enables ArkClaw to continuously optimize its service performance during usage. This offers a more professional and tailored intelligent experience for you.

warning

This feature is in Beta. Welcome to explore. If you experience any problems, exceptions, or have any suggestions during use, you may inform us via the feedback feature.

Background

Hermes Agent is an open-source, autonomous agent framework developed by the renowned AI research lab Nous Research, with the core philosophy of "Grows With You". It is not a simple chatbot, but a personal AI assistant designed for long-term user companionship, with continuous learning and evolution throughout the process. Hermes has powerful self-evolution capabilities that include learning from conversations, automatically generating and optimizing reusable skills, and maintaining persistent cross-session memory. These capabilities precisely compensate for the limitations of ArkClaw in personalization and long-term knowledge accumulation. Learn more.

A core feature of Hermes is its built-in Closed Learning Loop mechanism.

Memory and learning

Hermes is equipped with a persistent memory system that can retain user preferences, work habits, and details of previous tasks across multiple conversations. Rather than only recording information passively, it actively extracts and summarizes experience from complex task execution to form reusable methodologies.

Skill evolution

Upon completion of a complex task, Hermes automatically abstracts and solidifies problem-solving steps and methods into a reusable skill. This skill can be invoked directly when similar problems arise in the future, and are continuously refined and improved through subsequent usage to achieve genuine self-evolution.

Powerful tool and model ecosystem

Hermes incorporates more than 47 built-in native tools that cover file operations, web browsing, code execution, Docker, and other scenarios. Meanwhile, it provides strong tool expansion capabilities through the MCP standard protocol. You can always select the most suitable tools and models for specific tasks.

Automation and collaboration

With its built-in Cron scheduled task system, Hermes can independently execute periodic tasks such as daily report generation and regular inspection. It also supports splitting complex tasks and assigning them to multiple parallel sub-agents for processing, which greatly improves efficiency and capability in handling complicated problems.

Application scenarios

Execution of long-horizon tasks

Conventional chat models often lose context or stall when handling multi-step, long-duration tasks. Hermes is purpose-built for complex task orchestration, with strong task decomposition, state memory, and goal management capabilities. It maintains high focus during tasks lasting several hours until the final goal is achieved.

Handling of repetitive and structured complex tasks

If you frequently perform similar workflows with minor variations, such as compiling weekly data reports, regularly inspecting service status, or generating code from templates, the self-evolving skill capability of Hermes delivers exceptional value. It learns and solidifies your complete first-time workflow as a reusable skill. For subsequent tasks, you only need to issue a simple natural language instruction. Hermes will automatically invoke the skill to complete most work, greatly improving efficiency.

Usage notes

The Hermes feature is supported only when collaboration mode is disabled.

This feature is only available for ArkClaw instances running version ark-26.4.23 or later. You can check your ArkClaw version and perform upgrades on the "Status and usage" tab of the ArkClaw details page. Learn more.

When Hermes Agent is enabled for the first time, the system automatically uses the current model configuration of ArkClaw as the initial Hermes Agent configuration. When Hermes Agent is enabled, the model configurations of Hermes and ArkClaw become independent and do not affect each other. To change the model used by Hermes, see How do I replace the model used by Hermes Agent? for details.

Procedure

Switch to the target ArkClaw page.

For first-time use, install the Hermes Agent plugin by clicking "Hermes > Install plugin".

warning

The installation process takes about 5-10 minutes. Wait patiently.

If you are prompted that "The current ArkClaw version is outdated. Please upgrade first before installing the plugin.", click the "Upgrade and install plugin" button to upgrade the ArkClaw version and install the Hermes Agent plugin.

After the plugin is installed, refresh the current page or create a new session.

Enable the Hermes feature. The feature is enabled when the toggle button is in colored state. Subsequently, all conversations and tasks sent to ArkClaw will be processed and replied by Hermes.

tip

You can also use the /model hermes quick command in the ArkClaw web chat box or third-party message channels (such as Feishu or DingTalk) to quickly enable this feature.

Note that when you enable the Hermes Agent feature by using the quick command, the state of the Hermes button in the ArkClaw console does not update synchronously. The actual execution state prevails. For example, after the /model hermes command is executed, subsequent tasks will be processed by Hermes. Even if the Hermes button does not appear colored, the actual switch result is not affected.

If you temporarily do not need to use Hermes, you can disable this feature via the console or quick command. After you disable Hermes, all conversations and tasks sent to ArkClaw will be handled and replied by ArkClaw.

Disable through the console:

Disable the Hermes feature. The feature is disabled when the toggle button is in gray state.

warning

If your ArkClaw uses inference services from third-party LLMs with custom model configurations, you can currently disable this feature only via aquick command.

Disable via quick command

You can use the following quick commands in the ArkClaw web chat box or third-party message channels (such as Feishu or DingTalk) to quickly disable Hermes Agent.

/model ark: Applicable when ArkClaw uses a model from ModelArk Coding Plan, Agent Plan, or ModelArk "Foundation models".

/model custom: Applicable when ArkClaw uses third-party large models via custom model configurations.

warning

When you disable the Hermes Agent feature by using the quick command, the state of the Hermes button in the ArkClaw console does not update synchronously. The actual execution state prevails. For example, after the /model ark command is executed, subsequent tasks will be processed by ArkClaw. Even if the Hermes button does not appear gray, the actual switch result is not affected.

Usage examples

This section uses game review data analysis as an example to show how to use Hermes to complete tasks, accumulate skills, and reuse those skills for similar tasks later.

Sample prompt:

First task: Hermes, please help me write a script to scrape the latest 100 1-star and 2-star reviews for Escape from Duckov (App ID: 6743841750) from the Apple App Store. Use the model to analyze the top 3 core pain points most complained about by users and return the conclusion. After successful execution, solidify this workflow as a skill.

Subsequent task: Use the previously accumulated skill to analyze Honkai Impact 3rd (App ID: 1143402987) on the App Store.

Replay

Play
00:00 / 00:00 Live

00:00

Fullscreen

Cssfullscreen

1x

2x
1.5x
1x
0.75x
0.5x

Click and hold to drag

Last updated: 2026.05.27 20:37:25