Client area
Deploy OpenClaw on VPS in one click
Run your self-hosted OpenClaw agent on a VPS that stays online 24/7. Pick the preinstalled image at checkout and get working hosting in 5-15 minutes.
What can you do with self-hosted OpenClaw on a VPS
OpenClaw is an AI agent in your messaging apps that runs commands on your server. On a VPS hosting, it does that without your laptop being on.
One agent across Telegram, Slack, Discord, and more
Connect Telegram, WhatsApp, Slack, Discord, Signal, iMessage, Microsoft Teams, Matrix, and several other channels. Every channel reaches the same agent with the same memory, so a thread you start on your laptop continues from your phone, and it keeps answering after you close the lid.
Bring your own model, swap it in one line
Point OpenClaw at any OpenAI-compatible API: Claude, GPT, Gemini, or a model served through OpenRouter. Changing providers means switching with a /model command in chat, or setting the default in a single config line, which keeps you off a single vendor's pricing and rate limits.
Run a local model when the data stays in the box
Connect OpenClaw to Ollama on the same VPS, and prompts never reach an external API. You pay nothing per token, no third party logs your inputs, and inference stays inside the server you rent. Larger plans carry the RAM this needs.
Run real shell commands, get real results
OpenClaw executes shell commands, reads and writes files, browses the web, and ships code. Root access on your own box lets it apply a fix instead of only describing one, which is the reason people self-host it over a hosted chatbot.
Automate the recurring work
Schedule cron-style jobs and let the agent watch repos, inboxes, and endpoints on its own. It can commit state to a private repo overnight, summarize what changed, and message you the result by morning. The server runs the loop while nobody sits at a terminal.
Pull ready-made skills, or write your own
OpenClaw ships with built-in web search, file operations, code execution, and browser control. Thousands more community skills install from the ClawHub registry with one command. The registry is unvetted, so review a skill before running it on your VPS.
Snapshot before a risky skill, roll back after
Agentic loops misbehave, and a bad skill can hammer an API or wipe a directory while you watch. An isolated VPS with free weekly backups lets you snapshot before an experiment and restore the whole instance when a run goes sideways.
Choose a VPS plan for OpenClaw
OpenClaw needs at least 2 GB RAM, so the entry point is Start VPS at $10.19/mo, which handles basic chat with some memory. Browser skills, background jobs, and headless tasks make memory the first thing to run out, so Medium VPS (3 vCPU, 4 GB RAM, 40 GB NVMe, 1 IPv4) is the recommended starting point.
- 40 CountriesFrom$10.19 / 1 month$9.68 with is*smart
- Xeon 2x2.20 GHz CPU
- 2 Gb RAM
- 30GB SSD Drive
- 3 Tb Bandwidth
- 40 CountriesFrom$21.24 / 1 month$20.18 with is*smart
- Xeon 3x2.20 GHz CPU
- 4 Gb RAM
- 40GB SSD Drive
- Unmetered Bandwidth
- 40 CountriesFrom$33.99 / 1 month$32.29 with is*smart
- Xeon 4x2.20 GHz CPU
- 8 Gb RAM
- 50GB SSD Drive
- Unmetered Bandwidth
Boost Your VPS Server Resources
Check out how you can change the configuration of your VPS. You can purchase additional drive space, increase CPU performance, add RAM, and take advantage of additional services without having to switch to a new plan.
How to deploy OpenClaw on a VPS
The fast path uses the preinstalled image and the onboarding wizard.
Step 01.a. Order a VPS and SSH to it
Start meets the 2 GB requirement; pick Medium or higher for automation. At checkout, select the OS image Ubuntu 24 + OpenClaw.
Wait for provisioning. Access details arrive by email, usually within 5 to 15 minutes. SSH in:
ssh root@YOUR_VPS_IPStep 02.a. Run onboarding
openclaw onboardThe wizard walks you through model setup and messenger connections.
Step 03.a. Connect a messenger
Connect one messenger first, send a test message, and confirm that it replies. Once that first reply lands in your chat app, you are operational, and everything after it is tuning.
Step 01.b. Via scripts or Ansible
If you provision via scripts or Ansible and want a vanilla Ubuntu box, skip the image and use the upstream installer:
curl -fsSL https://clawd.bot/install.sh | bashexec bashopenclaw setup --wizardYou save rebuilding a base stack and spend a little more time on setup. For the fastest time to first message, the preinstalled image still wins.
Common issues
Field-tested gotchas.
Verify the upstream installer URL and the exact setup/onboard subcommand against the current OpenClaw docs, since the project was renamed.
is*hosting infrastructure is a solid base for OpenClaw
Your cores, your RAM via KVM
KVM virtualization with dedicated resources on each VPS means what you pay for is what runs. There is no shared pool deciding when your gateway gets OOM-killed.
No middlemen. Full control
We run VPS on our own hardware in trusted data centers across 40+ locations, with no resold capacity behind an intermediary. Root on every plan and scale resources by request.
Backed by Tier 3+ data centers
Every location meets Tier 3 standards or higher — redundant power and network, plus physical security — so the server hosting your agent keeps running.
Frequently asked questions about OpenClaw hosting
What is OpenClaw, and is it the same as Clawdbot?
OpenClaw is an open-source, self-hosted AI agent that lives in your messaging apps and runs commands on your server. The project carried the names Clawdbot and briefly Moltbot before this one, so any guide written for “Clawdbot on VPS” applies here too.
What are the VPS requirements for OpenClaw?
2 GB RAM covers basic chat with some memory retained. Automation raises the floor: browser skills, background jobs, and headless tasks push 4 GB and up. For CPU, 3 vCPU stays responsive under load, and NVMe keeps installs and logs quick. Those numbers describe our Medium plan.
Which plan should I pick?
For chat and light use, Start (2 vCPU, 2 GB RAM) at $10.19/mo meets OpenClaw's 2 GB minimum. For serious automation, Medium (3 vCPU, 4 GB, NVMe) at $21.24/mo is the recommended default. To run a local model through Ollama, choose Premium or higher for the added RAM.
Can I run a local model instead of a cloud API?
Yes, on larger plans. With enough RAM, you can run a local model through Ollama, keeping prompts, files, and inference on the VPS. You can also point OpenClaw at Claude, GPT, or Gemini through an API key, so the data path stays your choice.
Is it safe to expose OpenClaw on a public server?
Keep the admin and control interface off the public internet, since a random port provides no real protection. Reach it through an SSH tunnel or a reverse proxy with real authentication, log in with SSH keys instead of passwords, store messenger tokens as carefully as passwords, and start automation with minimal permissions. Skills carry their own risk: the ClawHub registry is unvetted, so read a skill before you run it, and lean on the VPS isolation and weekly backups to contain one that misbehaves.
How is OpenClaw different from Hermes Agent?
Both are open-source, self-hosted agents that live in messaging apps, keep persistent memory, and run scheduled jobs, so on a VPS, they cover the same ground. OpenClaw runs a single planning loop with human-authored skills and the widest messenger coverage, plus a large skill library. Hermes Agent (Nous Research) centers on a self-improving loop that turns completed tasks into reusable skills and stores memory in searchable vector form. If you want broad integrations and fast setup, OpenClaw fits; if you want an agent that compounds skills over time, Hermes is worth a look.
More from is*hosting
VPS
Deploy VPS instances across 40+ locations. Manage multiple servers from one account — useful for running background jobs or a separate local-model node.
Explore
From $5.94 /mo
Dedicated server
When a VPS isn't enough — dedicated hardware for high-traffic production, GPU workloads (local models with Ollama), or large databases.
Explore
From $75.00 /mo