This is a complete, hands-on walkthrough by Leon van Zyl covering everything from installing Hermes Agent on your own machine to deploying it on a virtual private server so it runs around the clock. Leon notes Hermes Agent received a lot of updates since his last video, including a new desktop app, a web interface, and a more streamlined setup. The end goal he sets up front: your very own AI assistant that runs 24/7 and that you can access from anywhere, even after you power off your PC.
What: Hermes Agent is a personal assistant that can read your emails, manage your calendar, connect to third-party platforms, and even create skills to keep up with you. It uses persistent memory to learn your preferences and workflows, and it focuses on automation via recurring/scheduled jobs.
Why: Because the agent remembers context across conversations and can automate routine work (for example a daily "good morning report" that goes online, does research, and compiles information), it becomes more useful the more you use it.
How: The agent lives everywhere and can be reached through many channels. By the end of the build you will be able to switch off your PC and still message the agent from your phone via Telegram or WhatsApp and from a web interface.
What: There are two ways to set up locally: the terminal or the desktop app. In the terminal you copy the install command for Mac/Linux or Windows from the official Hermes Agent website, run it in PowerShell (on Windows), and it installs Hermes plus all dependencies.
Why: The terminal route is the only place (at time of recording) where you can configure a custom endpoint, which is what you need to run free local models via Ollama or LM Studio.
How: During setup you pick a provider from a long list. Leon demonstrates option 32 (custom endpoint), pastes the Ollama endpoint, uses the word ollama as the API key, auto-detects a local model (Qwen 3.6), chooses the backend, optionally configures messaging platforms (skipped here), then runs the command hermes to launch the CLI. The first local response was slow (roughly two minutes to load the model into GPU memory), then sped up.
What: A newer option is the Hermes Desktop app, downloaded from the same website and installed with a one-click installer. It resembles ChatGPT or Claude, with conversations on the left that can be grouped, pinned, and renamed.
Why: It is the friendlier day-to-day interface, with access to your chat models, your file system (so it can act like a coding agent on project folders), and an integrated terminal.
How: Leon points out a key limitation: at time of recording you cannot set custom endpoints directly in the desktop app, so local models still require the terminal setup up front. The model list may be empty until you connect a provider; his list already showed Gemini only because he was signed into the Gemini CLI, and a local Qwen model because of the earlier CLI process.
What: To add a model, go to Settings in the top-right, then Providers. Hermes may auto-detect other coding agents you are signed into (Leon saw it using a Claude Code subscription).
Why (important caution): Leon says he would not recommend piggybacking on a Claude Code subscription, noting there is "bad blood" between Anthropic and assistants like this, because Anthropic wants users to pay for their own usage rather than piggyback. He also advises against raw API keys with these agents because it gets too expensive.
How: He connects OpenAI via OAuth, authorizes with a pasted code, then selects GPT-5.6 Soul (clicking "refresh models" if none appear). Hovering a model reveals options: thinking mode (more deliberate answers), fast mode (faster but pricier, which he disables), and an effort level (he prefers high).
What: A standout benefit is skill management. If you teach Hermes a repeated workflow (log into a system, pull stats, add them to a spreadsheet, email the result), it can create a reusable skill. You can also enable extra tools (for example Home Assistant), add MCP servers, and use "Browse Hub" to search community-built skills, tools, and MCP servers.
Why: These extensions turn Hermes from a chat box into an automation platform wired into your real systems.
How: Under Messaging you can connect Telegram, WhatsApp, Slack, Discord, and more (multiple at once). Leon sets up Telegram: in Telegram he opens BotFather, runs the new-bot wizard to get an API key, pastes it into Hermes, and adds his allowed Telegram user ID (found via a user-info contact's start command) to lock the bot to himself, with room to add family or colleagues by comma-separating IDs. After saving, he restarts the messaging gateway, and messages sent in Telegram appear in Hermes Desktop under a Telegram channel.
What: The major limitation of local running is that the agent goes offline when your PC does. The fix is to deploy Hermes Agent to a VPS (virtual private server), a cheap rented computer that stays on.
Why: With Hermes on a VPS you can power off your own PC and still reach the agent from your phone and a web interface, with the VPS instance acting as the single source of truth.
How: Leon prefers Hostinger for a near one-click install (disclosed sponsor, with a coupon code "LeonHermes" for an extra 10% off). He chooses a KVM plan and billing period, disables the ready-to-use AI extras, picks a nearby server location, and selects the Hermes Agent template under operating system. Credentials are provided during setup and can be recovered later via Docker Manager, then Projects, then the Hermes environment variables.
What: Opening the VPS Hermes URL gives a bookmarkable public web interface, accessible from any device, that looks like the desktop app but adds views for cron jobs, skills, plugins, MCP servers, and channels.
Why: This remote instance is the always-on brain you will point everything else at.
How: In the chat, type /setup. Rather than the quick setup (which uses the Now Spark portal), Leon chooses full setup, picks OpenAI Codex, authorizes with a pasted code, and selects GPT-5.6 Soul. After hitting a snag configuring Telegram through the wizard, he shows an alternative: authenticate, skip the messaging step, then simply ask the agent in plain language ("Please help me configure Telegram") and paste the bot token and his user ID. A test message confirms Telegram now talks to the VPS agent, which can run 24/7.
What: With the VPS running, the desktop app still points at its own local instance, so the two must be synced to the remote agent.
Why: You want a single source of truth (the VPS) so Telegram, the web, and the desktop all share the same conversations, skills, and schedules.
How: First, disable Telegram on the local instance and restart the local gateway so the desktop stops listening for Telegram. Then go to Settings, then Gateway, and point to the remote gateway URL (the VPS address minus /chat), signing in with the remote Hermes username and password and using "Test Remote" to confirm. After restarting, a question asked in one place ("What is the current weather in New York?") appears in the web session history, proving sync. Leon then asks the agent to create a cron job that runs every morning at 9:00 a.m. to send Claude and OpenAI news to Telegram; the job appears in the Cron view on both the web and desktop, and he reschedules it to 7:00 a.m. simply by asking over Telegram.
The walkthrough ends with the agent fully wired up: connected to OpenAI GPT-5.6, reachable from desktop, web, and Telegram, running 24/7 on a VPS, and executing scheduled automations. Leon notes you can extend it further to Gmail, calendar, and other systems, and points viewers to his free "7-Day Builder Challenge" course teaching the RAM framework for coding agents. The through-line: with persistent memory, skills, messaging channels, and an always-on VPS, Hermes Agent becomes a genuinely personal assistant you can reach from anywhere.