Video Summary
Hermes Agent reached 100,000+ GitHub stars unusually fast due to rapid releases and high PR volume.
The browser harness gives Hermes hands-on web control (see, click, type) so it can automate complex browser tasks.
Setup recommended on a VPS (Hostinger demo) with OpenRouter model selection and an API key for continuous operation.
Demos include jailbreaking a model, generating a full Mandarin video, scraping Hacker News and YouTube, and creating animations.
Hermes self-improves by creating new skills, updating docs, and adapting to site structure changes to avoid repeated failures.
Hermes grew rapidly because of an extremely fast release cadence and high contributor activity — five major releases in 20 days and hundreds of merged PRs (~741), which amplified visibility and adoption.
The browser harness lets Hermes 'see, click, and type' on websites, effectively acting as the agent's hand so it can automate complex browser tasks, create new helper functions, and self-heal when pages change.
Run Hermes on a VPS for 24/7 availability; the video demonstrates Hostinger's pre-built deployment page. You also configure an OpenRouter API key and choose an appropriate model for performance.
Hermes can jailbreak models (demonstrated with Gemma 4), generate full videos in other languages (Mandarin demo), scrape and compile web data (Hacker News, YouTube), and produce high-quality creative assets.
After running tasks Hermes identifies skill gaps, generates new functions/skills, documents procedures for future runs, and can add trusted commands to allow lists so it performs better on repeat tasks.
"This is becoming the fastest growing and most popular AI agent in the world."
Hermes agent is recognized as the fastest-growing and most popular AI agent globally, evidenced by its impressive increase in GitHub stars.
In comparison with OpenClaw, Hermes agent is showing a rising trend while OpenClaw has seen a decline in popularity.
The data from Google Trends indicates a strong interest in Hermes agent over the past three months, suggesting it may outpace OpenClaw.
"Hermes agent is the fastest project in all of GitHub history to reach 100,000 stars."
Hermes agent achieved a milestone of 100,000 GitHub stars in record time, largely due to its rapid update frequency, which includes five major releases within just 20 days.
The project has seen over 741 merge pull requests, averaging approximately 37 pull requests per day, showcasing the active development and support behind it.
"Hermes agent figured out how to jailbreak this model by itself."
One groundbreaking use case involves a user named Ply, who utilized Hermes to jailbreak a model called Gemma 4, demonstrating the agent's capability to autonomously solve complex problems with minimal human input.
Another use case showcased by Adam involves using Hermes agent to create a complete video in Mandarin, which included generating HTML files, Chinese text-to-speech, and rendering the video, emphasizing its potential in education and marketing.
"It created animations and GIFs which have a certain vibe, a certain feel."
Hermes agent was used to create submissions for a creative hackathon, producing high-quality animations and graphics that surpassed typical AI-generated content, which often lacks character and customization.
This demonstrates Hermes agent's ability to produce visually rich and engaging outputs that meet higher standards of creativity.
"Browser use is the hand. So, here's how it works."
To maximize Hermes agent’s capabilities, David explains the integration of a new browser harness tool allowing the agent to perform browsing tasks just like a human.
This setup involves acquiring a VPS (Virtual Private Server) and using Hostinger as an affordable option to run Hermes agent continuously.
"You want your AI engine running forever."
The setup process is straightforward; users must select a VPS plan on Hostinger, ideally for a long duration to get the best deal.
Configuring the server involves selecting Hermes agent as the application and completing the initial setup to deploy the VPS efficiently.
An important step in the configuration is obtaining and entering an API key from OpenRouter, reinforcing the importance of secure management of access keys.
"The default model should be as strong as possible."
Users are advised to select the latest and most powerful AI model available during the setup process to ensure optimal performance of Hermes agent.
The distinctions between different models available for deployment are highlighted, with recommendations made based on the user's budget and needs for performance.
"Inside of OpenRouter, you can use any model. They have open-source models, OpenAI models, entropic models, Google models. All of the models that exist are on OpenRouter."
The video discusses how to set up the Hermes agent using different AI models available on OpenRouter, emphasizing the versatility in selecting from various options.
The presenter illustrates the process of entering the model name for the use of Opus 4.7 and explains the importance of setting up messaging services to connect through platforms like Discord, Telegram, and Slack.
However, the primary focus in this segment is to install the browser harness for Hermes, which showcases the capabilities of a self-improving AI agent.
"You can use it to solve any problem in your life or business. It's not just for coding."
The narrator highlights the utility of AI in troubleshooting and problem-solving, particularly in configuring environments like UV on a virtual private server (VPS).
A step-by-step guide is provided for connecting to the VPS via SSH, including changing the SSH password to maintain security.
The narrator encourages viewers not to feel intimidated by the technical terms and assures them that they can manage their own VPS for various applications, including running AI agents and hosting websites.
"I'm going to put everything into a single place called a GitHub Gist, so you have all of the terminal commands I’m showing you here."
To streamline the process, the narrator plans to compile all terminal commands into a GitHub Gist for viewer convenience, complete with explanations of each command.
After successfully installing the browser harness, the narrator presents two options: launching a headless browser on the VPS or utilizing the BrowserUse cloud service for ease of visualization.
The discussion also covers potential roadblocks encountered while using the Docker container setup and provides directions for installing Hermes globally on the VPS for broader accessibility.
"This single command installs all the dependencies you need."
The installation command for the Hermes agent simplifies the process by checking the necessary dependencies like Linux version, Python, git, and Node.js.
By running this command initially, users can skip the step of manually checking the environment for the required software packages.
During the installation, a headless version of Chrome is downloaded automatically, allowing the agent to operate without needing an external browser interface.
Using a Virtual Private Server (VPS) provides an effective environment for running the agent, enabling tasks to be executed privately within the cloud infrastructure.
"This is a complex task—this is scraping, clicking on multiple posts, visiting them, and saving it inside a JSON file."
In a demo task, the agent is instructed to visit Hacker News and scrape information about the top 15 posts including titles, scores, and URLs.
The operation involves not only accessing the website but also performing multiple interactions like navigating through posts and compiling the data in a structured JSON format.
By leveraging Hermes’ browser harness, the agent can self-improve and adapt its handling of websites based on previous experiences and evolving knowledge of structure and layout changes.
Tasks that would conventionally take a significant amount of time, like manually gathering and organizing information, can now be completed in a fraction of the time using the agent's capabilities.
"This is literally proactive knowledge building."
Hermes agent has the ability to identify gaps in its existing skills and automate the creation of new functions, making it more capable with each interaction it has with various websites.
After executing tasks, the agent updates its skills and creates documentation on how to navigate different sites, which allows it to streamline future tasks in a similar context.
This self-learning mechanism ensures that the agent not only executes a task once but improves its efficiency for future iterations, making it faster and more effective in its operations.
By diagnosing unexpected issues or changes in a website's structure, Hermes can adjust its approach and contribute enhancements to its skill set for others.
"Imagine the average knowledge worker is super slow on their computer. They don't even know how to navigate YouTube properly."
The agent is tasked with retrieving the 12 most recent videos from the creator's YouTube channel and saving the results as a PNG file displayed in a grid format.
This complex operation highlights the advanced capabilities of the Hermes agent, as it goes beyond simple navigation and requires an understanding of both web scraping and image generation.
Many users would struggle with the task of efficiently gathering this information due to unfamiliarity with digital tools, thereby demonstrating the unique advantage of using an AI agent for such technological processes.
"Hermes agent will not fail it. So, first it looks at skill view, browser harness, terminal."
"We should be able to see how the agent is operating."
"Sometimes browser use is overkill for a lot of tasks; it's not needed."
"Sometimes it flags the command as a dangerous command, and all you need to do is just say add to the permanent allow list."
"It's beautiful watching it solve problems."
"So if you live in Czech Republic, Poland, or Slovakia and you want to join my team, make sure to apply."
"It's like self-improving, so every time you use it, it's better and faster."
"This is the beauty of browser harness and Hermes agent."
"Hermes agent is already more powerful than the average person you could employ in your company."
"If you go through these three weeks, which are broken down into granular step-by-step modules, you will be able to build any piece of software with AI."
