A server for AI agents that work around the clock
Personal agents such as OpenClaw and Hermes Agent, and the agents you build with current frameworks, need a machine that never sleeps: to answer on Telegram, WhatsApp or Slack, run scheduled tasks and call tools. Run them on a HyperDC server with full root access, the gateway kept private and tools sandboxed in Docker.
- OpenClaw and Hermes Agent, always on
- Agents built with LangGraph, CrewAI, Agno and more
- Any model provider, or your own models on a GPU
- Gateway on loopback, tools in Docker sandboxes
- StackOpenClaw needs Node.js 24.16 or later (26 recommended); Hermes Agent and most frameworks are Python.
- Node.js 24+, Python 3.10+
- Default portsBoth bind to loopback by default. Reach them over SSH, a VPN or an authenticated reverse proxy; never expose a gateway without authentication.
- OpenClaw 18789, Hermes 8642 (local)
- MinimumHermes Agent recommends 2 cores and 2–4 GB, and at least 2 GB for browser automation.
- Hermes Agent: 1 core, 1 GB RAM
- Your data
- Memory, skills and API keys in a data folder
- Official docs
- docs.openclaw.ai · hermes-agent.nousresearch.com
Agents and frameworks
- OpenClaw
- Hermes Agent
- LangGraph
- CrewAI
- Microsoft Agent Framework
- OpenAI Agents SDK
- Google ADK
- PydanticAI
- Agno
- OpenHands
Facts from the project’s official website, documentation and repository, checked in October 2026.
Plans are being prepared
We are preparing ready-to-use plans for AI agents. Tell us how you will use it and how many users you expect, and we will reply with a server that fits. You can also start today on a Linux VPS and install it with our guide.
Which server size fits?
Starting points for vCPU, memory and disk. Grow the server when your data and users grow.
| Feature |
Personal agent
One assistant on your chat apps
|
Recommended Team agents
Several agents, browser automation
|
Local models
Agents and the model on your hardware
|
|---|---|---|---|
| vCPUVirtual processor cores of the server. | 1–2 | 2–4 | 8+ |
| MemoryMemory for the app, its database and the operating system. | 2 GB | 4–8 GB | 16–32 GB |
| DiskNVMe or SSD storage for the app, its data and local backups. | 20 GB | 40 GB | 100 GB+ |
| ModelWhere the language model runs. | Hosted API | Hosted API | Ollama or vLLM on a GPU |
| Server type | Linux VPS | VPS or VDS | GPU server |
-
Personal agent
One assistant on your chat apps
- vCPUVirtual processor cores of the server.
- 1–2
- MemoryMemory for the app, its database and the operating system.
- 2 GB
- DiskNVMe or SSD storage for the app, its data and local backups.
- 20 GB
- ModelWhere the language model runs.
- Hosted API
- Server type
- Linux VPS
-
Recommended
Team agents
Several agents, browser automation
- vCPUVirtual processor cores of the server.
- 2–4
- MemoryMemory for the app, its database and the operating system.
- 4–8 GB
- DiskNVMe or SSD storage for the app, its data and local backups.
- 40 GB
- ModelWhere the language model runs.
- Hosted API
- Server type
- VPS or VDS
-
Local models
Agents and the model on your hardware
- vCPUVirtual processor cores of the server.
- 8+
- MemoryMemory for the app, its database and the operating system.
- 16–32 GB
- DiskNVMe or SSD storage for the app, its data and local backups.
- 100 GB+
- ModelWhere the language model runs.
- Ollama or vLLM on a GPU
- Server type
- GPU server
Hermes Agent documents 1 core and 1 GB of RAM as its minimum, 2 cores and 2–4 GB as recommended and at least 2 GB for browser automation; OpenClaw publishes no figures for running the gateway. Local models need a GPU server or a large CPU server.
What an agent server does for you
Agents are only useful when they can act at any hour, and only safe when you control what they can reach.
Always reachable
OpenClaw and Hermes Agent answer through Telegram, WhatsApp, Slack, Discord and more, and run scheduled jobs while your computer is off.
Your own agents as services
Deploy agents written with LangGraph, CrewAI, Microsoft Agent Framework, PydanticAI or Mastra as APIs and workers, with their memory in your database.
Tools in sandboxes
Let agents run code and shell tools inside Docker containers instead of directly on the host, with only the secrets they need.
MCP servers
Host the Model Context Protocol servers your agents use, next to the data they need, instead of on a laptop.
Your data stays yours
Prompts, files and databases stay on a server you control, in the location you choose, instead of a shared SaaS account.
Full root access
Install what the app needs, change any setting and run more services next to it. Nothing is locked behind a panel.
Near your users
Choose a data center in the United States, Europe or Asia. The order form estimates the latency from where you are to each location.
Grow without starting over
Start on a VPS, then move to a bigger plan, a VDS with NVMe storage or a dedicated server when the workload grows.
From order to first login
Order the app preinstalled on your server, or install it yourself with our guide.
-
Pick the server
Choose a size from the table above and the data center closest to the people who will use the app.
-
Add OpenClaw or Hermes Agent
Select OpenClaw or Hermes Agent as an app option when you order, or install it on a clean Ubuntu or Debian server with our guide.
-
Connect your channels and model
Run the onboarding, add your model provider key or local model endpoint, and pair the chat apps you want to use.
-
Lock it down
Keep the gateway on loopback, reach it over SSH or a VPN, run the built-in security audit and give tools a Docker sandbox.
Step-by-step setup guides
Install, secure and update the app with our guides, written for current Ubuntu and Debian releases.
-
How to self-host the OpenHands AI agent platform with Docker
Deploy OpenHands Agent Canvas, the open-source platform for AI coding agents, with Docker Compose, protect it with an API key and an IP allowlist, connect a model and keep the agent's sandbox tight.
40 min Advanced
Related solutions
Frequently asked questions
Why run an AI agent on a server?
Agents that reply to messages, watch inboxes or run scheduled tasks have to be online all the time. A server keeps them running when your computer sleeps, gives them a stable address and keeps their tools and data away from your personal machine.
What are OpenClaw and Hermes Agent?
Both are open-source personal AI assistants. OpenClaw, earlier known as Clawdbot and Moltbot, runs a gateway that connects your model to chat apps such as WhatsApp, Telegram and Slack. Hermes Agent by Nous Research keeps memory across sessions, writes reusable skills from experience and offers an OpenAI-compatible API.
Which agent frameworks are maintained in 2026?
Actively released in October 2026: LangGraph, CrewAI, Microsoft Agent Framework, the OpenAI Agents SDK, Google’s Agent Development Kit, PydanticAI, Agno and Mastra. Microsoft AutoGen is in maintenance mode, with Microsoft Agent Framework as its successor, and the Letta V1 server has been retired in favour of its new code base.
Do I need a GPU for agents?
No, as long as the agent calls a hosted model API: the agent itself is light. You need a GPU server, or a large CPU server for small models, only if the language model should run on your own hardware too.
How do I keep an agent safe?
Follow the projects’ own advice: never expose the gateway unauthenticated, keep it on loopback and reach it over SSH or a VPN, treat incoming messages as untrusted, require pairing for unknown senders, run tools in a Docker sandbox and keep API keys readable only by the agent’s user. OpenClaw includes a security audit command.
Why does Docker not respect my firewall for the agent’s port?
Ports published by Docker bypass ufw and firewalld rules. Publish the agent’s port on 127.0.0.1 only, or add your rules to the DOCKER-USER chain, as both Docker and OpenClaw recommend.
Which models can my agents use?
Any model the framework supports: hosted providers through their APIs, or open models on your own Ollama, vLLM or LocalAI server through an OpenAI-compatible endpoint. Pick the model per task and keep provider keys in the agent’s private configuration.
Can I run coding agents such as OpenHands?
Yes. OpenHands now ships as Agent Canvas: a self-hosted control centre for coding agents such as OpenHands, Claude Code and Codex. It listens on 127.0.0.1 port 8000 by default, so reach it over SSH or a VPN, and give the agents a separate user and Docker sandboxes.
Can I host MCP servers for my agents?
Yes. Model Context Protocol servers are small services that give agents tools and data. Run them on the same server or next to the systems they connect to, and protect remote ones with authentication and HTTPS.
Give your agents a home that never sleeps
Tell us the app, how many people will use it and where they are, and we will suggest a server for it.