Hermes Agent from Nous Research is the hottest open-source AI agent of 2026. 135,000 GitHub stars in three months. Runs on any Linux server, keeps memory across sessions, learns new skills on its own, drives your computer through the terminal, generates images, searches the web, and answers you on Telegram while you sleep. All of it for $5.50/month for the VPS plus your existing ChatGPT Plus subscription at $20/month. No API tokens, no surprise traffic bills, no vendor lock-in.
This guide walks you through standing up Hermes Agent on a GMhost VPS from zero to a working Telegram bot in 30 minutes. Step-by-step with real commands — if you can copy and paste into a terminal, you have everything you need.
What Hermes Agent is and why everyone's talking about it
Hermes is an autonomous agent that lives on your server 24/7 and gets things done that you used to do by hand. In 2026 this is no longer a demo with ChatGPT — it's a real system with memory, skills, and access to the actual internet.
What makes Hermes different from a regular chatbot:
- Memory across sessions — it remembers your projects, contacts, preferences. Tell it "use format X for reports" — a week later, it still uses it.
- Self-improving skills — it learns from every task you give it and adds new skills to its inventory. The more you use it, the better it understands you.
- 40+ built-in tools — web search, image generation, TTS, cloud browser, terminal execution, file operations and more.
- Multi-gateway — connects to Telegram, Discord, Slack, WhatsApp, Email at the same time. One agent answers you from everywhere.
- 300+ models via a single OAuth — Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, Grok. Connect once, use them all.
The most important piece: you can plug your existing ChatGPT Plus subscription at $20/month via OAuth — no extra API bills. Hermes spins up a local OpenAI-compatible proxy and uses your subscription the same way the ChatGPT web interface does.
What your agent will be able to do once it's running
Real tasks our customers already automate:
- Research — "find the top 10 competitors in my niche, analyze their pricing and put it in a Google Sheet"
- Document work — "read this PDF, pull out the key points, write a short report in Notion"
- Content — "generate 5 variants of an image for a post about X in style Y", "draft a reply to this customer email"
- Monitoring — "check this competitor site every 2 hours and ping me on Telegram when a new price shows up"
- PC control — via cloud browser or terminal execution: "log in to my admin panel, download yesterday's report, forward it to accounting"
- Development — "clone this repo, read the README, suggest three improvements to the architecture"
- Personal stuff — "remind me at 18:00 every Friday to take out the trash", "book a table at [restaurant] for Saturday through OpenTable"
- Data analysis — "take these CSV files, build a pivot table and send the charts to our Slack channel"
You hand all of this off as plain messages in Telegram. The agent breaks each one into sub-tasks, picks the right tools, runs them, sends back the result.
Why a VPS instead of running it locally
You can run Hermes on your laptop and it works. But there are four reasons everyone serious switches to a VPS:
- The agent has to live 24/7 — close the laptop, head to the office, get on a plane, and the agent keeps working. On a laptop it sleeps with you.
- A stable connection — web search, image gen, the OAuth proxy all need a reliable internet link. A coffee shop hotspot won't cut it.
- Telegram access from anywhere — you message the bot from your phone on the metro, the agent on the VPS picks up the task and runs with it while you're on the move.
- A separate machine — Hermes can chew through several gigabytes of RAM under parallel load. On a laptop that eats battery and competes with your day job.
Which GMhost VPS to pick for Hermes
The minimum for Hermes is 2 GB RAM, 20 GB disk, Linux. Comfortable work starts at 4 GB RAM and 40+ GB NVMe.
I'd recommend one of three plans depending on your budget and load:
- PT-NVME-VPS-4 — $5.50/month (currently -50% off). 2 vCPU, 4 GB RAM, 40 GB NVMe, located in Lisbon. The cheapest starting point, perfect for a single user.
- UA-NVME-VPS-4 — $10.99/month. 2 vCPU, 4 GB RAM, 40 GB NVMe, located in Ukraine. Pick this if you care about low latency to Kyiv (2-5 ms) and Ukrainian jurisdiction.
- PT-NVME-VPS-8 — $9.49/month (discounted). 4 vCPU, 8 GB RAM, 60 GB NVMe. Pick this if you're planning parallel tasks, heavy image generation, or connecting several messenger gateways at once.
All plans come with NVMe disks, unmetered traffic, and a choice of Ubuntu/Debian/Rocky at order time. This guide uses Ubuntu 24.04 LTS — Hermes installs fastest on it.
Step 1. Order the server (5 minutes)
- Open the VPS catalog at gmhost.ua, pick a plan (I recommend PT-NVME-VPS-4 or UA-NVME-VPS-4).
- At checkout, choose Ubuntu 24.04 LTS in the OS field.
- Pay — the server is activated in 5-10 minutes and the root credentials (IP, password) land in your inbox.
- Connect over SSH:
bash ssh root@your-server-ip
If this is your first VPS — set up SSH keys instead of a password (ssh-copy-id) right away and change the default password. Basic hygiene, but plenty of people skip it.
Step 2. Prep the system (5 minutes)
Hermes needs Python 3.11+. On Ubuntu 24.04 it's already there — you just need to update packages and install a few dependencies:
I'd also recommend creating a separate user instead of root (for safety):
From here on, all commands run as the hermes user.
Step 3. Install Hermes Agent (3 minutes)
The official installer puts everything in place with one line:
Check the install worked:
It should show v0.14.0 (or newer — Nous Research ships updates every 2-3 weeks).
Step 4. Connect ChatGPT Plus (5 minutes)
This is the part that brings a lot of people over to Hermes. Instead of paying separately for the OpenAI API (where tokens drain and the bill is unpredictable), Hermes can use your existing ChatGPT Plus subscription at $20/month.
It works over OAuth — no API keys, no extra billing.
Run the one-command setup:
The script opens a browser (or shows a QR code for mobile) with an OAuth login link. Pick "Login with OpenAI", sign in with your ChatGPT Plus account — done. The token is stored locally on the server and the agent starts using your subscription.
If it all worked, a quick test:
The interactive CLI opens. Type Hi, what model are you and do you have internet access? — you'll get an answer from GPT-4 (or whichever model your subscription gives you right now).
Exit the CLI with Ctrl+D or /exit.
Step 5. The Telegram gateway (10 minutes)
Now the most useful part — hooking up Telegram so you can message the agent from your phone.
First, create a bot through @BotFather:
- Find
@BotFatherin Telegram - Send
/newbot - Pick a name and a username (must end in
_bot) - Save the token BotFather hands you — it looks like
123456789:ABCdef...
Now over in Hermes:
The script asks which gateway you want — choose Telegram. Paste the token you saved. Done.
Start the gateway:
Open your bot in Telegram and send /start — Hermes should reply. Try simple things first: "What's the dollar exchange rate right now?", "Find me the top 5 coffee shops in Kyiv near Khreshchatyk metro."
To keep the gateway running 24/7 (so it doesn't stop when you log out of SSH), run it under systemd:
Check the status:
The agent now answers you on Telegram even after you close the SSH session. You can reboot the server and the gateway comes back up on its own.
What skills (tools) you already have out of the box
After hermes setup --portal you already have 40+ tools installed. The basics:
- web_search — searches Google/Bing and summarizes the results
- web_browse — opens a page and pulls out the content (cloud browser, works with JS-heavy sites too)
- image_generation — generates images (DALL-E 3 via the OpenAI OAuth, or Flux/SDXL through providers)
- text_to_speech — voice synthesis for replies
- terminal_execute — runs bash commands on the server (with your confirmation, for safety)
- file_read / file_write — works with files on the server
- code_executor — runs Python code in a sandbox
- github_tools — creates issues, PRs, reads repositories
- calendar_tools — Google Calendar integration
- email_tools — reads and sends mail over IMAP/SMTP
- notion_tools — works with your Notion workspace
- scheduler — sets up cron-style jobs for the agent ("do this every morning at 9:00")
Full list of what's available:
Or from Telegram: /skills
How to make the agent genuinely useful: 5 working scenarios
A handful of real use cases you can set up in 10-15 minutes each:
1. Morning briefing. Tell the agent: "Every morning at 8:30, find the top 5 stories from my industry RSS feeds, put them in a short brief and send to my Telegram." It schedules the cron and you get a digest with breakfast.
2. Email auto-replies. Hook up your mailbox with email_tools. Instruction: "For client emails with common shipping questions — reply with the template from our FAQ; for trickier ones — forward to my Telegram with a short summary."
3. Competitor monitoring. "Every day at 9:00 visit my competitors' sites (list below), diff the prices against the last check, ping me when something changes."
4. Content pipeline. "Once a week pick a topic from my Notion backlog, write a 1000-1500 word blog post draft, generate a cover image, save as draft. I review and publish."
5. Decision-making researcher. "I'm choosing between SaaS services X and Y. Find comparisons, Reddit/G2 reviews, build a comparison table, give me a recommendation with reasoning."
A few questions that usually come up
How much traffic does Hermes eat? Under average use — 5-15 GB/month. Our VPS comes with unmetered traffic, no overage worries.
Is it safe to give the agent root access? By default Hermes doesn't run destructive commands without your confirmation. Still, run it under a separate user (hermes in our guide), not root.
What happens if my ChatGPT session logs out? Hermes detects it and prompts you to log back in. Normally the OAuth token lasts weeks, not hours.
Can I plug in Claude instead of ChatGPT? Yes. hermes model lets you switch providers. Supports Claude Pro/Max via OAuth (just like ChatGPT).
Can I run several agents on the same VPS? Yes. The default config has one profile, but you can create hermes profile create work and hermes profile create personal. Each gets its own memory and skills.
How do I cancel the VPS? Anytime, from your account at gmhost.ua. No penalties, no hidden conditions.
Wrap-up and CTA
In 30 minutes and $5.50-10.99 a month, you have your own AI agent — living on your server, replying on Telegram from anywhere, getting tasks done while you sleep, and using your existing ChatGPT Plus subscription with no extra API bills.
This isn't the future, it's just what 2026 looks like day-to-day. Hermes Agent is already being used by 135,000 developers worldwide, and the number is doubling every quarter.
The stack:
- GMhost VPS from $5.50/month (Ubuntu 24.04, NVMe, unmetered traffic)
- Hermes Agent (free, open-source, Apache 2.0)
- ChatGPT Plus $20/month (your existing subscription via OAuth)
- Telegram (free)
Order a VPS for Hermes:
- 🇵🇹 PT-NVME-VPS-4 for $5.50/month (Lisbon)
- 🇺🇦 UA-NVME-VPS-4 for $10.99/month (Ukraine)
- Cancel anytime. 24/7 support in English on Telegram: @gmhost_support_bot
If you hit any snags setting up Hermes on our VPS — message the support team. An engineer will help you get from zero to a working Telegram bot in 30 minutes, free of charge. It's part of the standard onboarding for AI users.
