The moving parts
The host keeps the agent connected and its settings available.
The Gateway is the part that stays online and receives messages. It keeps the agent’s settings and conversations. The agent can use the tools and access you connect to it, so give it only what it needs. An outside AI company may still process your messages.
A different kind of gateway
The agent hub and the messaging service do different jobs.
Keeps the agent ready
It keeps the agent’s sessions and settings together. It also connects tools and outside AI models.
Carries a WhatsApp or SMS message
It carries messages between WhatsApp or SMS and the agent. OpenClaw and Hermes each have their own way to connect WhatsApp. Other WhatsApp services are separate products and do not work with every agent automatically.
Start with the job
You may not need an always-on server yet.
| Your need | A sensible place to start |
|---|---|
| You are only learning | Your own computer |
| It must work while your computer is off | An always-on server or managed app |
| You can use SSH, a secure command window, and update a server | A self-managed VPS |
| You want less server work | A managed or ready-made setup, after checking which jobs it really handles |
| You want to run an AI model on the server too | Check the hardware requirements for the exact AI model |
Read the setup guides:OpenClaw resource guidanceHermes Docker resource guidance
Four options
Each option balances work and control.
Your computer
- Good for
- A way to learn and test without another server.
- Your work
- You install and update the app. It stops when the computer is off.
- Be careful
- Avoid it when the agent must answer while your computer sleeps.
Home server
- Good for
- Keeps the agent on hardware you control at home.
- Your work
- You care for power and internet. You also own remote access, updates and backups.
- Be careful
- Avoid it when nobody can fix the home network or hardware.
Self-managed VPS
- Good for
- An always-on virtual server that lets you choose the software and how you connect.
- Your work
- You update the server and the app. You also protect access, set network rules, and make backups.
- Be careful
- Avoid it when nobody owns those server jobs.
Managed agent plan
- Good for
- Can remove some setup, update, or server work.
- Your work
- You still choose who can use it, which actions are allowed, the secret service keys, the tools, and the AI model.
- Be careful
- Avoid it when the plan never explains what the provider manages.
Two products
OpenClaw and Hermes can both keep a messaging hub online.
The Gateway owns the live connection
OpenClaw says its Gateway keeps sessions and routes messages to chat connections. On a VPS, back up its settings and work files.
Check the current OpenClaw server guideCompare five OpenClaw VPS setupsKeep the messaging data folder
Hermes documents a normal install and a container, which is a packaged place for the app to run. Its data folder must stay safe when that package is replaced.
Check the current Hermes container guideCompare Hermes and OpenClaw Check a VPS for Hermes AgentMinimum server size
Treat the minimum as a starting point.
These numbers come from product guides checked on August 25, 2026. A vCPU is a virtual processor, RAM is working memory, and disk is storage. The minimum does not promise a smooth setup.
| Product setup | Published minimum | Suggested size for more room |
|---|---|---|
| Basic OpenClaw Gateway | 1 vCPU, 1 GB RAM, about 500 MB disk | 1–2 vCPU and 2 GB+ RAM |
| Hermes container gateway | 1 core, 1 GB RAM, 500 MB data | 2 cores, 2–4 GB RAM, and 2 GB+ data |
Controlling a browser, using several chat channels, running separate agent jobs, or running the AI model on the server can need more memory. Check the tools you will really use.
Who does each job?
“Hosted” does not mean every job is managed.
Hosting company
Runs the physical hardware and network. A managed plan may promise more.
You
Choose who can use it, the secret service keys, chat apps, allowed actions, tools, and the AI model.
Self-managed VPS owner
Also updates the server and app. The owner must watch the server, make backups, and block unwanted network connections.
Five checks
Protect the door, keys, work area, and backup.
- 01
Keep the dashboard private. Use a private or protected network connection.
- 02
Approve users. Approve each person who connects, or keep a list of allowed users.
- 03
Protect secret keys and access codes. Never publish them or put them in public code.
- 04
Limit the work area. Give the agent a limited account, not the root account that can control everything. Keep risky work in a separate area.
- 05
Back up saved information. Test that you can restore the agent settings and work files.
The full agent bill
The server is only one cost.
Add any domain, secure network tool, or paid search and browser service that your setup uses. Prices can change, so use the current checkout instead of a number from this guide.
Sources and limits
Read the current product guides before you choose.
- OpenClaw Linux server guide
- OpenClaw first-run FAQ
- OpenClaw security guide
- Hermes Docker guide
- Hermes model provider guide
- Hermes security guide
Sources checked August 25, 2026. Server requirements and managed-plan duties can change. We did not speed-test these server setups.
Straight answers
AI agent hosting questions.
Do I need a VPS for an AI agent?
No. Start on your own computer. Use a VPS when the agent must stay online without that computer.
Can normal web hosting run an AI agent?
Do not assume it can. The plan must support the required software and a service that can keep running in the background.
Does a VPS include the AI model?
Not always. The agent and the AI model can be separate services. Check which outside AI service and secret key the setup expects.
How much RAM does an AI agent need?
RAM is the server’s working memory. The amount depends on the tools. Calling an outside AI model usually needs less memory than browser tasks, several agent jobs running at once, or an AI model running on the server.
Is self-hosting completely private?
Not automatically. You control the host, but prompts may still go to the model provider and other services that you configure.
What does managed AI agent hosting mean?
It means the provider manages specific listed tasks. Check exactly who handles the app, system updates, secure access, and backups.