AI agents · Always on · Fixed IP

AI Agent VPS A server that keeps your agents online day and night

A Linux VPS for AI agents that have to stay reachable: chat assistants, coding agents and automation workflows. You get a private KVM server with full root, reserved memory and one fixed IP. The 4 GB plan that suits most agents starts at $14.99₹1,389.27 a month, billed monthly.

AI agent VPS plans

Most agents call a model over the internet, so the server needs memory for the agent and its tools, not a GPU. 4 GB is the safe pick for one agent. 2 GB is enough for a light chat agent, and 8 GB suits several agents or a headless browser. Billed monthly, with no long contract.

Compute Region

Choose the deployment location
Currently out of stock

AI agent VPS plans · Miami

Miami, Florida · AMD EPYC KVM VPS with full root access and a choice of 15 Linux builds
Plan vCPU Memory Storage Bandwidth Setup Price
4GB Plan Recommended 4 vCPU 4 GB 80 GB SSD 5 TB Instant $14.99/mo₹1,389.27/mo Buy Now
6GB Plan 4 vCPU 6 GB 100 GB SSD 5 TB Instant $19.99/mo₹1,852.67/mo Buy Now
8GB Plan 6 vCPU 8 GB 100 GB SSD 5 TB Instant $29.99/mo₹2,779.47/mo Buy Now
16GB Plan 8 vCPU 16 GB 100 GB SSD 5 TB Instant $59.99/mo₹5,559.87/mo Buy Now
Choose Ubuntu 24.04 on the order form. The setup steps on this page are written for it.

Which agent are you running?

Six agents people commonly put on a VPS, the plan we suggest for each, and a page with the exact install steps.

Only Claude Code and Codex publish a memory figure. The other sizes are our own suggestion, as of November 2026.
AgentWhat it isPlan we suggestFull guide
OpenClawA self-hosted assistant gateway from the OpenClaw Foundation. It runs on your server and you talk to it in your chats.4 GB. 2 GB for a light chat agent.OpenClaw VPS
Hermes AgentA self-improving AI agent built by Nous Research, MIT licensed, with a messaging gateway for Telegram, Discord, Slack, WhatsApp, Signal and email.2 GB. 4 GB if it drives a browser.Hermes Agent VPS
Claude CodeAnthropic's agentic coding tool. It works in the terminal and can be steered from the Claude app with Remote Control.4 GB, the requirement Anthropic lists.Claude Code VPS
CodexOpenAI's terminal coding agent.4 GB minimum, 8 GB recommended, per the install notes in the Codex repository.Codex VPS
OpenCodeAn open source coding agent that works with the model provider you choose.4 GB. 2 GB is workable for light use.OpenCode VPS
n8nWorkflow automation you host yourself with Docker. The Community edition is the free edition.2 GB to start. 4 GB for busy workflows.n8n VPS

What an agent gets from its own server

An agent is only useful while it is running and reachable. That is the whole case for a server.

Always on

A laptop sleeps, restarts for updates and leaves the house with you. A server does none of that, so a message sent to your agent at 3 a.m. still gets an answer.

A fixed IP for webhooks and bot gateways

Each server has its own IPv4 address that does not change. Services that call your agent back, and allow-lists that ask where your agent calls from, can use the same address every day.

Kept apart from your personal machine

An agent can read files and run commands. On a VPS it only sees what you copied there, not your photos, browser logins or the other devices on your home network.

Root, so nothing is off limits

Install any runtime, any package and Docker. Each VPS is a real KVM virtual machine, so you are not waiting on a host to support the tool you picked.

Reserved RAM

The memory in your plan is set aside for your server. An agent with a browser and a few tools open is not squeezed because a neighbour got busy.

Rebuild any time

If an agent makes a mess of the system, reinstall the operating system from the client area and start again from clean. No ticket needed.

Set up any AI agent on a VPS in 6 steps

These steps are the same whichever agent you choose. The installer commands in step 4 are the official ones from each maker.

  1. 1

    Order a plan with Ubuntu 24.04

    Pick a plan in the table above and select Ubuntu 24.04 when the order form asks for the operating system. Once the payment clears, the server builds itself and an email brings the IP address and the root password.

  2. 2

    Create a normal user and log in as that user

    Agents should not run as root. Log in as root once, create a user, log out, and come back in as the new user. Replace YOUR_SERVER_IP with the address from the email.

    ssh root@YOUR_SERVER_IP
    adduser agent
    usermod -aG sudo agent
    exit
    ssh agent@YOUR_SERVER_IP

    Logging in directly, and not with su, matters later: Linux starts a user's own background services only for a real login.

  3. 3

    Turn on the firewall with only SSH open

    Allow SSH first, then switch the firewall on. Everything else coming in from the internet is now refused, while the agent can still call out to model APIs and chat services.

    sudo apt update && sudo apt install -y ufw
    sudo ufw allow 22/tcp
    sudo ufw enable
    sudo ufw status

    One catch if you use Docker: ports published by a container are not filtered by ufw. Publish them to 127.0.0.1 only.

  4. 4

    Install your agent with its official installer

    Install the basic tools first, then run one of the installers as the normal user. Each one is the line published by the maker of that agent.

    sudo apt install -y git curl tar tmux
    
    # OpenClaw
    curl -fsSL https://openclaw.ai/install.sh | bash
    
    # Hermes Agent
    curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
    
    # Claude Code
    curl -fsSL https://claude.ai/install.sh | bash
    
    # Codex
    curl -fsSL https://chatgpt.com/codex/install.sh | sh
    
    # OpenCode
    curl -fsSL https://opencode.ai/v2/install | bash
  5. 5

    Set the agent up and run it as a service

    A service starts at boot and keeps running after you log out. Agents with a gateway bring their own command for this. The last line lets your user's services stay up when you are not logged in.

    # OpenClaw: onboarding installs its gateway service
    openclaw onboard --install-daemon
    openclaw gateway status
    
    # Hermes Agent
    hermes setup
    hermes gateway setup
    hermes gateway install
    
    # keep user services running after logout
    sudo loginctl enable-linger $USER

    Coding agents such as Claude Code are interactive, so they are usually kept alive inside tmux instead. For your own script, see the systemd example further down.

  6. 6

    Reach the gateway through an SSH tunnel, not an open port

    If your agent has a gateway or web page, leave it listening on the server itself (127.0.0.1) and do not open its port. Run a tunnel on your own computer instead. This example uses port 18789, the OpenClaw gateway port. Use your agent's port in its place.

    ssh -N -L 18789:127.0.0.1:18789 agent@YOUR_SERVER_IP

    While that command runs, port 18789 on your own computer leads to the gateway on the server. Nobody without your SSH login can reach it.

VPS vs your own hardware

A spare PC or a Mac mini at home can run an agent too. Here is where each one is the better choice.

Own hardware is the better choice for local models and for experiments that never need to be reached from outside.
Spare PC or Mac mini at homeVPSLake VPS
How you payBought once, plus the electricity to leave it onMonthly, from $14.99₹1,389.27 for 2 GB
Same public IP every dayUsually not on a home line
Reachable by webhooks without port forwarding
Keeps running through a home power or internet cut
On the same network as your personal devicesYesNo
Can run a large local modelYes, if it has the GPU or memory for itNo. There is no GPU
Start again from a clean systemReinstall by handReinstall from the client area

What an AI agent VPS is

An AI agent is a program that takes a goal, asks a language model what to do next, and then does it: runs a command, edits a file, sends a message, calls another service. An AI agent VPS is simply the computer that program lives on. It is a Linux server in a datacenter that you reach by SSH, with full root, reserved memory, SSD storage and its own IPv4 address.

It suits agents that must be reachable when you are not at your desk: a chat assistant that answers on a messaging app, a coding agent working through a long task, or a workflow that fires when a webhook arrives. If you only talk to an agent now and then while sitting at your own computer, run it there. You do not need a server for that.

No GPU here: how the agent gets its model

These servers have no GPU. That is fine for nearly every agent, because the agent does not do the thinking itself. It sends each request to a model over the internet, using an API key or a subscription you already have, and the server carries only the agent, its tools and its files. This is why a few gigabytes of memory are enough.

Running a large model on the server itself is a different job. It needs a GPU machine, and these plans are not that. If your goal is a fully local model, a VPS without a GPU is the wrong purchase, and we would rather say so here than after you have paid.

What we do and do not set up for you

What you rent is the machine: a clean Linux install, the root login and an address that does not change. The agent is yours to choose and install, with no one-click installer and nothing preinstalled, so you start on the version its maker ships today. Model subscriptions and API credits come from the model provider, not from us. If the server itself needs attention, such as a lost SSH login, a reboot or a fresh reinstall, our support team answers by ticket, WhatsApp and Telegram every day of the week.

Running your own agent script as a service

If you wrote the agent yourself, in Python or Node.js for example, let systemd look after it. Save a unit file like this as /etc/systemd/system/my-agent.service, with your own paths. Keep API keys in the .env file, readable only by the agent user, and not in the code.

[Unit]
Description=My AI agent
After=network-online.target

[Service]
User=agent
WorkingDirectory=/home/agent/my-agent
EnvironmentFile=/home/agent/my-agent/.env
ExecStart=/home/agent/my-agent/venv/bin/python main.py
Restart=always

[Install]
WantedBy=multi-user.target

Then start it and have it start at every boot:

sudo systemctl daemon-reload
sudo systemctl enable --now my-agent
sudo systemctl status my-agent

Restart=always brings the agent back if it crashes. If your agent serves tools to other AI clients, the same server can do MCP server hosting.

Where the servers are, and what else fits

You choose the location in the plan table: Miami, Florida, Frankfurt, Germany or Amsterdam, The Netherlands. For an agent, pick the place closest to the services it talks to or to the people who use what it builds. Your own distance matters little, because only text travels to you. If an agent needs a browser you can watch, a graphical Ubuntu RDP desktop is easier than a terminal. If you plan to host the app your agent writes on the same machine, see vibe coding VPS.

OpenClaw, Hermes Agent, Claude Code, Codex, OpenCode and n8n are products of their respective makers. VPSLake is an independent hosting provider and is not affiliated with any of them.

Pay the way you already pay

  • USDT, BTC & 20+ coinsvia Cryptomus
  • UPI / PhonePe / PaytmINR via Razorpay
  • Visa / Mastercardvia Paddle
  • Net banking & walletsIndia

Frequently Asked Questions

It is a rented virtual machine in a datacenter that stays switched on, used as the home for agent software such as OpenClaw, Hermes Agent or a coding agent. The agent, its files and its logins live on the server, so it keeps answering messages and finishing tasks while your own computer is off.
The one sized for what your agent does. A chat agent that calls a model over the internet is light, and 2 GB is our suggestion for it. Claude Code and Codex both list 4 GB as their minimum, and several agents or a headless browser are more comfortable on 8 GB. Apart from memory, look for full root, a fixed IP and monthly billing, which every plan on this page has.
Yes. If it runs on Linux, it runs here: a Python or Node.js script, a Docker container, or a ready-made agent installed with its official installer. You have root over SSH, so you install what the agent needs and run it as a service that starts again after a reboot. The setup steps on this page hold for any agent.
Not for the agent itself. Most agents send each request to a model over the internet, through an API key or a subscription, and the server only runs the agent program, its tools and its files. These servers have no GPU. Running a large model locally is a different job that needs a GPU machine, and we do not sell one.
It depends on the job. A VPS has a fixed public IP, sits outside your home network and does not depend on your home power or internet. Your own hardware has no monthly fee and can run local models if it has the GPU or memory for them. The comparison table on this page shows both sides.
No. You get a clean Ubuntu or Debian server with root access and you run the official installer of the agent you chose. That way you are on the version its maker ships today, not an old image. The installer commands for five common agents are in step 4 on this page.
No. The VPS is the computer the agent runs on. The model behind it comes from your own account: a Claude or ChatGPT plan, or an API key from a provider such as Anthropic, OpenAI or OpenRouter. We do not sell subscriptions or credits.
Each tool has its own way. Claude Code prints a login link: press c to copy it, open it on your own device and paste the code back. Codex has codex login --device-auth (beta), and OpenCode uses opencode auth login. An API key from your model provider is the other route and needs no browser at all.
Keep it closed. An agent gateway can read files and run commands, so let it listen on localhost only and allow nothing but SSH in the firewall. You then reach it through an SSH tunnel, which needs your SSH login. OpenClaw's own server guide says the same: keep the Gateway on loopback and access it via SSH tunnel or Tailscale Serve.
Yes. Switch the price to INR with the currency button in the menu and pay by UPI. Crypto and card are accepted too. The servers are in the USA and Europe, which is fine for an agent, because it talks to model APIs and chat services over the internet.

Give your agent a home that stays switched on

Order a VPS, run the official installer of your agent and leave it working. Billed monthly, cancel when you like.

Related solutions