1. Setting Up Hermes Agent on a Virtual Private Server (VPS) [0]

thumbnail Hermes is a powerful AI agent that can automate tasks 24/7, but it requires a specific setup on a Virtual Private Server (VPS) to run continuously.

1.1. Understanding Hermes Agent and the Need for a VPS [0]

  1. Hermes agent is a powerful AI tool that can automate various tasks, but it can be challenging for beginners to set up due to its complex dashboard and numerous settings [0].
    1. Hermes can organize emails, conduct research, monitor competitors, write posts, and handle outreach [1].
    2. It can perform these tasks continuously, even when the user is asleep [2].
  2. The desktop app version of Hermes is limited because it stops working when the computer is turned off or the laptop is closed [14].
    1. Any ongoing tasks or scheduled jobs will pause until the machine is awake again [17].
    2. An agent that only functions while being actively watched is not truly an autonomous agent [18].
  3. A Virtual Private Server (VPS) is recommended for running Hermes 24/7, as it acts as a "computer in the cloud" that is never turned off [19].
    1. Deploying Hermes on a VPS allows the agent to continue working on tasks even if the user closes their laptop or goes to sleep [24].

1.2. Deploying Hermes on Hostinger VPS [26]

  1. To set up Hermes on a VPS, users can go to hostinger.com/zhoamates and use the code "zinhoa10" for a 10% discount [26].
  2. The recommended VPS plan is KVM2, which provides two cores, 8 GB of RAM, and 100 GB of storage, offering the right balance for a fully connected agent and a team of agents [29].
    1. Longer-term plans, such as 24 months, reduce the monthly rate, though 12 months is also a viable option [31].
  3. The installation process is simplified through the Docker manager in the Hostinger panel [33].
    1. Users open the Docker manager, go to "compose," search the catalog for "Hermes," and select "Hermes agent" [34].
  4. Users must create a strong admin username and password for Hermes, as this password secures the entire agent [36].
    1. This password should be saved securely and not in an easily accessible location [40].
    2. After setting the credentials, users click "deploy" to complete the setup [41].
  5. Upon deployment, users can log in to the Hermes dashboard using their created username and password [45].
    1. The dashboard is the central hub for managing almost all agent activities [49].
  6. The Hermes dashboard features several key components [50]:
    1. Models serve as the agent's "brain," providing the intelligence from whichever AI model is plugged in [51].
    2. Each agent has a personality (its tone and writing style) and a memory (what it knows about the user) [54].
    3. Skills are step-by-step instructions for completing specific jobs, with 27 built-in skills and the ability for Hermes to write its own [58].
    4. Kuran is for scheduling jobs that the agent performs at set times, such as daily email checks [64].
    5. Channels connect Hermes to external applications like WhatsApp [67].
    6. Keys store passwords for other services [68].
    7. Profiles are used to create additional agents [69].
    8. Plugins include Kanban, which facilitates teamwork among agents [70].
    9. The gateway status on the bottom left of the dashboard indicates the agent's "heartbeat" and should always show "running" [73].

2. Building and Managing a Team of AI Agents [76]

thumbnail

This section explains how to configure Hermes with an AI model, create a personal assistant agent, connect it to WhatsApp and Gmail, and then expand it into a team of specialized agents that collaborate on tasks.

2.1. Configuring the First Agent and Connecting Services [76]

  1. To give the agent a "brain," users must integrate an AI model [77].
    1. The video uses Claude API keys, but other providers like Newsportal can also be used [78].
    2. Users navigate to "keys," then "providers," select "Anthropic API key," and paste their key, ensuring it is kept private [80].
    3. In "models," the main model is changed to "Claude Opus" [84].
  2. The first agent is set up as a personal assistant to check emails and perform research [88].
  3. To connect the agent to a phone, users go to "channels," find WhatsApp, and scan the displayed QR code with their phone, similar to WhatsApp Web [93].
    1. QR codes expire quickly, so a new one can be generated if missed [97].
    2. The user's phone number must be added to the allow list to permit communication with the agent [99].
    3. After adding the number, "restart gateway" must be clicked in the bottom left for new connections to take effect [101].
  4. To connect email, Hermes has a built-in email skill [103].
    1. Users type "set up my Gmail" in the chat, and Hermes guides them through the process [105].
    2. A Gmail app password is required, which is a special password generated in Google account security settings after enabling two-step verification [107].
  5. Once set up, the agent can be texted commands, such as "What emails need my attention today?" [110].
    1. The agent will read the inbox, filter out noise, and present the most important emails with their key information [113].
    2. The agent can also perform research and draft replies, for example, "Research the company in that second email and draft me a reply" [116].
    3. Replies are only drafted and not sent until the user gives explicit permission [122].

2.2. Expanding to a Team of Specialized Agents [125]

  1. To enhance workflow, additional agents can be created to form a team [125].
    1. For example, a second agent can be built specifically for writing content like LinkedIn posts or newsletters [127].
  2. In Hermes, each agent is called a profile, and each profile has its own memory, brain, and job [130].
    1. To create a new agent, users go to "profiles" in the dashboard and create a new one, such as "writer" [132].
    2. It is important to create the new profile as a copy of the main agent to transfer keys and settings, ensuring it works immediately [133].
    3. A specific description should be given to the new agent, such as "write posts and newsletters in my voice," as Hermes uses these descriptions to assign tasks when agents work together [136].
  3. The "writer" agent's model can be set to Claude Sonnet [139].
    1. The agent is then taught the user's writing style, for example, "short punchy sentences, a strong first line, and no corporate words," which it will remember for all future writing [140].
  4. The team is completed by creating two more agents with specific roles [144]:
    1. A researcher with the description: "find the latest news and trends in my industry" [146].
    2. An editor with the description: "fact checks content before it reaches me" [148].
  5. When assigning AI models to the team, the smartest model (e.g., Claude Opus) should be assigned to the "boss" or manager agent, which handles task distribution [153].
    1. Cheaper models (e.g., Haiku) are assigned to worker agents like the researcher and editor, as they perform most of the heavy lifting [154].
    2. The writer agent retains a quality-focused model like Sonnet because writing quality is crucial [156].

2.3. Orchestrating Teamwork with Kanban and Autopilot [157]

  1. The main agent (manager) needs its Kanban tools turned on to enable it to hand out work to the team [157].
    1. The team now consists of one manager and three workers [159].
  2. Kanban serves as the team's "office" or workspace, where tasks are represented as cards that move across columns (e.g., "waiting," "being worked on," "done") as work progresses [160].
    1. In Hermes, agents automatically move these cards [164].
    2. Users can access Kanban under "plugins" in the sidebar [165].
  3. To initiate a team task, the user texts the manager agent a command, such as "Get the team to write me a LinkedIn post about the biggest AI news this week" [169].
    1. A new card appears on the Kanban board, and the manager automatically splits it into sub-tasks [171].
    2. Research tasks go to the researcher, writing tasks to the writer, and fact-checking tasks to the editor [172].
    3. Tasks are linked, meaning writing cannot begin until research is complete [173].
  4. The board checks for new work every 60 seconds, but users can click "nudge dispatcher" to prompt immediate action [177].
    1. The research card moves to "running," and once research is done, the writer picks up the task with the research already integrated [180].
    2. The handoff between agents is automatic, resulting in a finished, fact-checked post delivered to WhatsApp in the user's voice [183].
  5. To make the team work on autopilot, users can schedule recurring tasks [187].
    1. For example, texting the manager: "Every morning at 7, check my inbox and send me what needs my attention and get the team to write me a post about the biggest news in my industry" [188].
    2. The agent confirms the schedule, and the job appears on the "Kuran" page in the dashboard, showing its next runtime [191].
    3. From the Kuran page, users can edit, pause, or run the job immediately [193].
    4. When run, the inbox summary is sent to the phone, a new card appears on the Kanban board, and the team begins working automatically [195].
    5. This process repeats every morning at the scheduled time, regardless of whether the user is awake [197].

3. Troubleshooting and Advanced Capabilities [199]

thumbnail

This section covers common mistakes to avoid when setting up an AI agent team, highlights Hermes's ability to write its own skills, and provides information on where to find resources for setting up the system.

3.1. Common Mistakes and Solutions [199]

  1. Team not picking up work: Check the "gateway status" in the bottom left of the dashboard [200].
    1. If it doesn't say "running," it means no one is available to hand out work [201].
    2. The solution is to click "restart gateway" [203].
  2. Wrong agent getting the work: This indicates that the profile descriptions are too vague [204].
    1. For example, "helper" tells Hermes nothing, but "fact checks content" provides clear instructions [205].
  3. Task jumps to "blocked" state immediately: This usually means the task is assigned to an agent name that does not exist, often due to a typo [206].
    1. Hermes attempts the task twice before flagging it as blocked [208].
    2. The solution is to fix the agent name and then unblock the task [209].

3.2. Hermes's Self-Writing Skills and Resource Availability [210]

  1. Hermes can write its own skills [210].
    1. While the team was working, Hermes created a new skill for researching and writing posts without being explicitly asked [212].
    2. This means that the next time a similar post is requested, Hermes will find and use that self-generated skill [219].
    3. This capability differentiates Hermes from a chatbot, as the team improves with every task it performs [220].
  2. This setup is highly adaptable and can be used for various tasks by swapping agents, such as an outreach email writer or a competitor researcher [221].
  3. To set up Hermes, users can visit hostinger.com/zenautomates and use the code "Zenhill Automates 10" for a 10% discount [223].
    1. The link directs to the VPS page [225].
    2. All prompts, agent descriptions, and the exact board setup are available for free in the provided school community [226].