How to Run Hermes Agent 24/7 on a $5 VPS (Anonymous Setup)

Hermes Agent is the first open-source AI agent that actually gets smarter the longer it runs. It builds skills from experience, maintains memory across sessions, and reaches you through Telegram, Discord, Slack, or Signal from a single gateway process.

But it only works if it's running. Here's how to deploy it on a cheap VPS, keep it alive 24/7, and do it without handing over your identity.

Hermes Agent is an open-source self-improving AI agent built by Nous Research and released in February 2026. It crossed 95,000 GitHub stars in its first seven weeks. The reason for the traction is simple: unlike chatbot wrappers that reset to zero every session, Hermes maintains persistent memory, creates reusable skills from solved problems, and connects to over 200 LLM providers without lock-in. It's the agent framework that actually compounds over time.

Running it on your laptop works for testing. Running it on a VPS is what makes it useful. The agent's learning loop, messaging gateways, and scheduled automations need a host that stays online when you close your MacBook. This guide covers how to get Hermes running 24/7 on a VPS for as little as a few dollars a month, with an anonymous setup that keeps your identity out of the billing records.

Why Hermes Agent Needs a VPS

The pitch for Hermes Agent is "an agent that grows with you." Growth requires continuity. Continuity requires uptime. A VPS provides both.

Here's what changes when you move Hermes from your laptop to a VPS:

Persistent memory actually persists. Hermes stores learned skills, conversation history, and user modeling in a local SQLite database. On your laptop, that database is only accessible when your laptop is open. On a VPS, it's accessible 24/7. The agent's FTS5 full-text search and LLM summarization across past sessions only become useful when there are weeks of sessions to search through.

Messaging gateways stay online. Hermes can connect to Telegram, Discord, Slack, WhatsApp, Signal, Matrix, email, and more from a single gateway process. These integrations need a stable, internet-facing host with consistent uptime. Your laptop going to sleep kills the bot. A VPS doesn't sleep.

Scheduled automations run on time. Hermes has a built-in cron scheduler that can generate reports, monitor systems, scrape data, or send summaries on any schedule. Cron jobs need a machine that's actually running at the scheduled time. That's a VPS.

Your conversations don't live on a shared device. If you're using Hermes for anything sensitive — business automation, research, financial monitoring, development workflows — the conversations and learned context shouldn't be sitting on your daily driver laptop. A dedicated VPS isolates the agent's data on infrastructure you control.

What Hermes Agent Actually Does (60-Second Overview)

If you haven't looked at Hermes yet, here's the short version of what makes it different from every other agent framework:

Hermes Agent vs. Typical AI Chatbots:

Typical chatbot:
  → Stateless. Every conversation starts from zero.
  → Single provider. Locked to one LLM.
  → Chat only. Copy-paste is your "integration."
  → No persistence. Close the tab, lose everything.

Hermes Agent:
  → Persistent memory. Remembers your projects, preferences,
    and context across every session. Forever.
  → 200+ models. OpenRouter, Nous Portal, OpenAI, Anthropic,
    self-hosted endpoints. Switch with one command.
  → Multi-platform. Telegram, Discord, Slack, WhatsApp,
    Signal, email, CLI. One gateway, all channels.
  → Self-improving. Solves a hard problem, writes a reusable
    skill document, uses it next time automatically.
  → Autonomous. Cron scheduler, subagents for parallel work,
    browser automation, terminal execution.
  → Your infrastructure. MIT licensed. No telemetry.
    No data leaves your machine.

The skill system is the feature that separates Hermes from everything else. When the agent solves a complex, multi-step problem, it can synthesize the approach into a reusable skill document stored locally. Next time it encounters something similar, it queries its own skill library instead of starting from scratch. Skills are searchable, shareable, and compatible with the agentskills.io open standard. As of April 2026, Hermes ships with 118 built-in skills.

The Setup: Hermes Agent on a VPS in Under 10 Minutes

Here's the actual deployment process. No fluff, just commands.

Step 1: Get a VPS

Hermes Agent is lightweight. The actual LLM inference happens on the provider's side (OpenRouter, Anthropic, OpenAI), not on your VPS. You're running a Python process, a messaging gateway, and a SQLite database. Minimum requirements are 2 GB RAM and 10 GB disk.

On Servury, the O-50 plan at $3.99/month gets you 2 GB RAM, 20 GB NVMe, and unmetered bandwidth on owned hardware in Montreal. That's more than enough for personal use. For multi-channel setups with heavy tool usage, the 4 GB plans starting at $10.99/month give you headroom.

No email, no name, no KYC. Generate a credential, pay with crypto if you want, and your server is live in 30 seconds. Pick Ubuntu 24 or Debian 12.

Step 2: SSH In and Install

Once your server is deployed, SSH in and run:

# Update the system
apt update && apt install -y python3-pip python3-venv git tmux

# Install Hermes Agent via the official one-liner
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

The installer handles everything: Python dependencies, virtual environment setup, and symlinking the hermes command to your PATH. It works on Ubuntu, Debian, and most Linux distributions.

Alternatively, if you prefer Docker:

docker run -d \
  --restart unless-stopped \
  --name hermes \
  -v $HOME/.hermes:/root/.hermes \
  ghcr.io/nousresearch/hermes-agent:latest serve

Docker gives you cleaner isolation and automatic restarts. The persistent volume mount (-v) ensures your skills, memories, and configuration survive container restarts and upgrades.

Step 3: Configure Your LLM Provider

Run the setup wizard:

hermes setup

This walks you through choosing your LLM provider and model. The most common setup:

• OpenRouter — Access to 200+ models with a single API key. Route different models to different tasks (premium for complex work, budget for simple queries). Best flexibility.
• Nous Portal — Direct access to Nous Research's own Hermes models. If you want to run the Hermes model family specifically, this is the cleanest path.
• OpenAI / Anthropic — Direct API access. Works if you already have keys.
• Self-hosted — Point Hermes at your own Ollama, vLLM, or SGLang endpoint. Zero external API costs. Requires a beefier VPS (16+ GB RAM) or a separate GPU server.

You can switch providers later with hermes model — no code changes, no migration. This is one of Hermes' strongest design decisions.

Step 4: Connect Your Messaging Platforms

This is where a VPS deployment starts making obvious sense:

# Start the messaging gateway setup
hermes gateway setup

# Configure platforms (Telegram, Discord, Slack, etc.)
# You'll need your bot tokens / API keys for each platform

# Start the gateway
hermes gateway

# Install as a system service so it survives reboots
hermes gateway install

The gateway process connects to all your configured platforms simultaneously from a single process. You can message your agent from Telegram on your phone while it executes tasks on the VPS. That's the workflow that makes Hermes feel like a real assistant rather than a terminal toy.

Step 5: Keep It Running

Three options, from simplest to most robust:

tmux (quick and dirty):

tmux new -s hermes
hermes serve
# Ctrl+B then D to detach. The agent keeps running.

systemd (production):

# Create a systemd service
cat > /etc/systemd/system/hermes.service << 'EOF'
[Unit]
Description=Hermes Agent
After=network.target

[Service]
Type=simple
User=root
ExecStart=/root/.local/bin/hermes serve
Restart=always
RestartSec=5

[Install]
WantedBy=multi-user.target
EOF

systemctl enable hermes
systemctl start hermes

Docker with auto-restart (cleanest): Already covered in Step 2 — the --restart unless-stopped flag handles reboots automatically.

Why Anonymous Hosting Matters for AI Agents

Here's the angle most Hermes deployment guides skip entirely: the privacy implications of running a persistent AI agent on identifiable infrastructure.

Hermes Agent stores everything locally. Your conversations, learned skills, user modeling data, project context, and browsing history all live on the VPS. If you signed up for that VPS with your real name, email, and credit card, all of that data is linked to your legal identity through the hosting provider's billing records.

This matters more for an AI agent than for a static website. A website serves content you chose to publish. An AI agent accumulates content from every interaction you have with it. Over weeks and months, your Hermes instance builds a detailed model of how you think, what you work on, who you communicate with, and what decisions you make. That's a high-value target for anyone with access to the hosting records.

Running Hermes on an anonymous VPS breaks the link between your identity and that data. No email at signup means no correlation point. Crypto payment means no financial paper trail. No logging means the provider can't reconstruct your activity even if compelled.

The data on the VPS is still there — but nobody knows it's yours. That's the difference between privacy and anonymity, and it's the difference that matters when your AI agent knows more about your work than your colleagues do.

Hermes vs. OpenClaw: Why People Are Switching

OpenClaw was the previous darling of the self-hosted agent world. Hermes Agent launched as a direct alternative, and the migration has been fast. The hermes claw migrate command exists specifically because Nous Research expected the switch.

The differences that matter for a VPS deployment:

Self-improving skills vs. static plugins. OpenClaw has 52+ built-in skills. Hermes ships with 118+ and creates new ones autonomously. After a week on a VPS, your Hermes instance has skills that don't exist in any other installation because it wrote them based on your specific workflows.

Multi-platform messaging. Hermes supports 16 messaging platforms from a single gateway. Telegram, Discord, Slack, WhatsApp, Signal, Matrix, email, SMS, and more. OpenClaw's platform support is narrower.

200+ model support. Hermes connects to any OpenAI-compatible endpoint. Switch providers with a single command. OpenClaw is more tightly coupled to specific providers.

Zero telemetry. Nothing phones home. No analytics, no tracking, no data leaves your VPS unless you configure an external service. For anonymous VPS deployments, this is essential — if the agent itself is sending telemetry, anonymous hosting is pointless.

If you're already running OpenClaw on a VPS, migrating to Hermes preserves your SOUL.md persona, memories, skills, API keys, and messaging settings:

hermes claw migrate          # Interactive migration
hermes claw migrate --dry-run # Preview what gets imported

Cost Breakdown: What This Actually Costs Per Month

The "$5 VPS" in the headline isn't marketing. Here's the real math:

Monthly Cost Breakdown:

VPS hosting (Servury O-50, 2GB/20GB):       $3.99
LLM API costs (depends on usage):
  - Light use (few queries/day):            $1-5
  - Moderate (Telegram bot, daily use):     $5-15
  - Heavy (multi-channel, automations):     $15-50+
  - Self-hosted LLM (Ollama):              $0 (but need bigger VPS)

Total for light personal use:               ~$5-9/month
Total for active daily driver:              ~$19-30/month

The VPS is the cheap part. LLM API costs scale with usage. OpenRouter lets you route expensive models (Claude, GPT-4) to complex tasks and budget models (Hermes 4, Llama) to simple ones, keeping costs manageable.

Compare this to managed AI assistant services that charge $20-30/month for a single model with no self-hosting, no privacy, and no persistent memory. A self-hosted Hermes setup on an anonymous VPS gives you more capability for less money, with full data ownership.

Common Mistakes (And How to Avoid Them)

Not using a persistent process manager. If you just run hermes serve in a regular SSH session, it dies when you disconnect. Use tmux, systemd, or Docker. This is the number one "why did my bot stop working?" issue.

Skipping loginctl enable-linger. If you're running Hermes as a non-root user with systemd user services, you need this command or systemd kills your services when you log out.

Forgetting to secure SSH. Your Hermes VPS has root access to an agent with your API keys and conversation history. Disable password authentication, use SSH keys only, and restrict access:

# /etc/ssh/sshd_config
PasswordAuthentication no
PermitRootLogin prohibit-password

Exposing the web UI without a tunnel. If you're running Hermes WebUI, don't expose it on a public port. Use an SSH tunnel or Tailscale instead. The WebUI binds to localhost by default for good reason.

Using a KYC exchange to pay for your "anonymous" VPS. If your goal is anonymous hosting, paying from Coinbase creates a trail back to your identity. Use an intermediate wallet. Or better, use Monero. We covered this in detail in our guide to buying VPS with Bitcoin or Monero.

The Bottom Line

Hermes Agent is the most capable open-source AI agent available in 2026. But an agent that only runs when your laptop is open isn't an agent. It's a fancy terminal app.

A VPS turns Hermes into what it's designed to be: a persistent, always-on assistant that learns from every interaction, reaches you on any platform, and runs automations while you sleep. The setup takes under 10 minutes. The cost starts at $3.99/month for hosting plus whatever you spend on LLM API calls.

And if you care about who can link your AI agent's accumulated knowledge back to your legal identity — and you should — doing this on anonymous infrastructure isn't paranoia. It's operational hygiene.

Deploy a Hermes Agent VPS on Servury. No email, no KYC, crypto accepted. Your agent is live in 30 seconds. What it learns from there is between you and it.

---

Frequently Asked Questions

How much RAM does Hermes Agent need on a VPS?

The official minimum is 2 GB RAM with 10 GB disk. For daily use with messaging gateways and a growing skill library, 4 GB is more comfortable. The agent itself is lightweight — the heavy lifting (LLM inference) happens on the provider's side. You only need 16+ GB if you're running a local model on the same server.

Can I run Hermes Agent for free on a VPS?

The VPS itself always costs something (cheapest plans start around $3-5/month). LLM API costs can be zero if you self-host a model with Ollama, but that requires a beefier server. The most cost-effective path is a cheap VPS with a budget model through OpenRouter or Nous Portal.

Which messaging platforms does Hermes Agent support?

As of v0.9.0: Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Mattermost, Email, SMS, DingTalk, Feishu, WeCom, BlueBubbles, Home Assistant, and CLI. All from a single gateway process. Microsoft Teams is the notable absence for enterprise users.

Is it safe to run Hermes Agent on a shared VPS?

Yes, with caveats. Hermes stores API keys and conversation data locally. On a shared (non-dedicated) VPS, the hypervisor operator technically has access to your VM's memory. For most threat models this is fine. If you're handling highly sensitive data, consider a dedicated server or encrypted VPS with LUKS2 where even the provider can't read your disk.

How do I update Hermes Agent on my VPS?

Run hermes update from inside your Hermes environment. If you're using Docker, pull the latest image and restart the container. Your skills, memories, and configuration are stored in ~/.hermes/ and persist across updates.

Can I run multiple Hermes agents on one VPS?

Yes. Use Docker containers with separate volumes, or separate user accounts with their own ~/.hermes/ directories. Each instance needs its own messaging bot tokens and LLM API keys. A 4 GB VPS can comfortably run 2-3 lightweight instances.