Hermes Agent on Free VPS + Free LLM:The Complete $0/Month Setup

The Architecture

Before you start clicking through Oracle Cloud's setup wizard, it helps to understand how the four components fit together. The diagram below shows where each piece lives and how they communicate.

Notice that the heavy lifting — the actual LLM inference — happens on OpenRouter's servers, not on your VPS. Your free VPS only needs to run the lightweight Hermes Agent process (~500MB RAM at idle), a small SQLite database for memory, and any tools the agent invokes (browser, terminal, file ops). That's why a 24GB Oracle Cloud instance feels like overkill — and why this setup actually works on much smaller VPSes too.

About Qwen 3.6 Plus (Reality Check)

If you came here looking specifically for qwen/qwen3.6-plus:free, here's the honest situation as of April 10, 2026:

For the rest of this guide, we'll use Qwen3 Coder 480B. Everything else (VPS setup, Hermes installation, configuration) stays identical regardless of which model you choose — so if a better free model appears next week, you can swap with one command: hermes model.

The Best Free Models on OpenRouter (April 2026)

OpenRouter currently lists 27 free models. Most are not suitable for AI agents because they lack tool/function calling support. Below is the shortlist of free models that actually work with Hermes Agent — ranked by context window.

Detailed Comparison (Tool Calling Confirmed)

Source: OpenRouter free models collection, verified April 10, 2026. Free models share capacity with paid tiers and may experience throttling during peak hours.

Step 1: Get Your Free VPS

We're using Oracle Cloud's Always Free tier. It's the only major cloud that gives you genuinely free compute with no time limit and no credit card expiration. You get up to 4 ARM OCPUs (Ampere A1), 24GB RAM, 200GB block storage, and 10TB outbound data transfer per month.

1.1 Sign up for Oracle Cloud

1. Go to oracle.com/cloud/free and click "Start for free"
2. Choose your home region carefully — pick one with ARM capacity (US East Ashburn, Frankfurt, and London tend to have the best availability)
3. Complete email verification and credit card validation
4. Wait 5-10 minutes for account provisioning

1.2 Create an Always Free ARM instance

1. From the Oracle Cloud dashboard, click Create a VM instance
2. Name it hermes-agent
3. Under Image and shape, click "Edit" and select Canonical Ubuntu 22.04
4. Click "Change shape" → select Ampere → VM.Standard.A1.Flex
5. Set OCPUs to 4 and Memory to 24 GB (the maximum free allocation)
6. Add an SSH key (generate one with ssh-keygen if you don't have one)
7. Click Create

For a deeper VPS comparison and Oracle Cloud setup tips, see our Best VPS for AI Agents guide.

Step 2: Initial Server Setup

Once your instance is running, grab the public IP from the Oracle Cloud dashboard and SSH in. Run these commands in order — each one is essential.

2.1 SSH into your VPS

# Replace YOUR_IP with the public IP from Oracle Cloud
ssh ubuntu@YOUR_IP

# If you get a permissions error on your private key:
chmod 600 ~/.ssh/id_rsa

2.2 Update the system & install Git

sudo apt update && sudo apt upgrade -y
sudo apt install -y git curl ca-certificates
sudo apt autoremove -y

2.3 Open the firewall (Oracle-specific)

Oracle Cloud uses iptables by default and blocks most ports. You need to allow outbound HTTPS so Hermes can talk to OpenRouter:

# Allow outbound HTTPS (already allowed by default but verify)
sudo iptables -L OUTPUT -n | head -5

# If you plan to use Hermes via Telegram/Discord (no inbound port needed)
# you can skip the rest. If you want to expose a webhook or web UI later,
# open port 443 in BOTH places:
#   1. Oracle Cloud Console: Networking → VCN → Security List → Ingress Rules
#   2. The Ubuntu firewall:
sudo iptables -I INPUT -p tcp --dport 443 -j ACCEPT
sudo netfilter-persistent save

2.4 (Optional but recommended) Create a non-root user

# Create user, grant sudo, copy SSH keys
sudo adduser hermes
sudo usermod -aG sudo hermes
sudo rsync --archive --chown=hermes:hermes ~/.ssh /home/hermes
sudo su - hermes

Step 3: Get Your Free OpenRouter API Key

OpenRouter is a unified API gateway for 200+ LLMs. It's OpenAI-compatible, which means anything that speaks the OpenAI API format (including Hermes Agent) works out of the box. Free tier signup takes 30 seconds.

1. Go to openrouter.ai and sign in (Google, GitHub, or email)
2. Navigate to Keys in the sidebar
3. Click Create Key, name it hermes-agent
4. Copy the key (starts with sk-or-v1-...)
5. You don't need to add credits to use free models — but be aware of the rate limits

Step 4: Install Hermes Agent

Hermes Agent has the cleanest install of any AI agent framework I've tested. One command does everything — installs Python 3.11, Node.js, ripgrep, ffmpeg, uv (the Python package manager), and Hermes itself. The installer auto-detects your OS.

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

The installer takes 3-5 minutes on a fresh Oracle Cloud ARM instance. When it finishes, reload your shell:

source ~/.bashrc

# Verify the install
hermes --version

Step 5: Configure With the Free Model

Now we wire Hermes to OpenRouter and tell it to use Qwen3 Coder 480B (free). There are two pieces: the API key (stored as a secret) and the model selection (stored in config).

5.1 Set the OpenRouter API key

hermes config set OPENROUTER_API_KEY sk-or-v1-YOUR_KEY_HERE

5.2 Set Qwen3 Coder 480B as the active model

hermes model openrouter/qwen/qwen3-coder:free

Or if you prefer to edit the config file directly:

model:
  provider: openrouter
  model: qwen/qwen3-coder:free

# Optional: use a different model for vision tasks
auxiliary:
  vision:
    provider: openrouter
    model: google/gemma-4-31b:free

hermes model openrouter/qwen/qwen3.6-plus-preview:free

Step 6: Your First Agent Run

Time to see if everything works. Start with a simple chat to verify the model connection:

hermes chat -q "Hello! What model are you running on?"

You should see a response within a few seconds. If you get a 401 error, your API key is wrong. If you get a 429, you've hit the rate limit (wait a minute). If you get a 404 with "model not found," the free tier may have ended — try a different model from the table above.

Test the agentic capabilities

Now let's test something that actually exercises Hermes's tool-calling and learning loop:

hermes chat -q "Check disk usage on this server, then write a summary to ~/disk-report.md"

Hermes will: invoke the terminal tool to run df -h, parse the output, format a summary, write it to a file, and confirm completion. The first time it does this, it'll create a skill document in ~/.hermes/skills/. Next time you ask something similar, it'll reference that skill and run faster.

Step 7: Run Hermes as a Service (Optional)

If you want Hermes to run continuously — for scheduled tasks, Telegram/Discord bot mode, or always-on memory — set it up as a systemd service so it survives reboots.

[Unit]
Description=Hermes Agent
After=network.target

[Service]
Type=simple
User=hermes
WorkingDirectory=/home/hermes
ExecStart=/home/hermes/.local/bin/hermes serve
Restart=on-failure
RestartSec=10
Environment="HOME=/home/hermes"

[Install]
WantedBy=multi-user.target

sudo systemctl daemon-reload
sudo systemctl enable hermes
sudo systemctl start hermes

# Check it's running
sudo systemctl status hermes

# View logs
sudo journalctl -u hermes -f

Cost Reality Check

"$0/month" sounds too good to be true, so let's break down what you're actually getting for free vs. what the equivalent paid setup would cost.

You're essentially getting ~$1,400-2,600 of value per year for free. The catch is rate limits and the inherent fragility of relying on free tiers — both providers can change terms with little notice.

Rate Limits & Gotchas

Free tiers come with constraints. Understanding them upfront prevents frustration when your agent suddenly stops working at 3 AM.

Skip the Setup: Cognio Professional Deployment

The free setup is great for personal use and learning. For business-critical deployments where you need reliability, security hardening, and someone to call when things break — Cognio engineers deploy a production-ready Hermes Agent (or OpenClaw) instance on your infrastructure in 24-48 hours.

Frequently Asked Questions