Last updated: August 20, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Cloudways Managed AI Agents is a hosting product that lets you deploy the open-source agents OpenClaw and Hermes on a managed server without touching a terminal. You pick a plan, pick an agent, connect an API key and a chat channel, and Cloudways handles the provisioning, patching, and backups.
Short answer: Cloudways Managed AI Agents runs OpenClaw or Hermes on a dedicated, managed instance you control from the Cloudways dashboard. Plans start at $9.99/month for the Scout tier (1 vCPU, 2GB RAM) and scale to $79.99/month for Swarm (8 vCPU, 16GB RAM). Setup takes under two minutes; your LLM API costs are billed separately by OpenAI, Anthropic, or Google.
Cloudways — a DigitalOcean company better known for managed WordPress and PHP hosting — went into public preview with this product earlier in 2026 and reached general availability on August 12, 2026. I set up both supported agents on a Cloudways account this week to see whether "managed hosting for AI agents" is a real category or just a rebrand of a generic VPS. It's closer to the former: the dashboard removes the Docker and firewall work, but you still own the API keys and the judgment calls about what the agent is allowed to touch.
What you'll need
You need an active Cloudways account — the free trial covers this, no separate signup — plus an API key from whichever model provider you plan to use, since Cloudways hosts the agent runtime but not the model itself. OpenAI, Anthropic, and Google keys all work with both agents. If you want the agent reachable outside the dashboard, have a Slack, Discord, Telegram, or WhatsApp account ready to connect as a channel; this step is optional and you can skip it and use the built-in web console instead. Budget a card on file: none of the four plans below has an indefinite free tier once you're past the trial window.
Step-by-step: setting up Cloudways Managed AI Agents
1. Open Managed AI Agents from your Cloudways dashboard
Log in to Cloudways and look for "Managed AI Agents" in the left navigation, next to your regular server and app list. It's a separate resource type from a web application, even though it bills through the same account.
2. Choose OpenClaw or Hermes
OpenClaw is a multi-agent control plane built for running several specialized agents at once and routing between Slack, Discord, Telegram, and WhatsApp from a single gateway; it's also, by a wide margin, the more famous of the two — its GitHub repository had passed 386,000 stars when I checked on August 20, 2026, making it one of the most-starred projects in GitHub's history. Hermes, built by Nous Research, is a single always-on runtime that keeps memory across sessions and builds reusable skills as it works, rather than coordinating a team of agents. If you want one assistant that gets better at your specific workflows over time, pick Hermes. If you want a dispatcher routing tasks to several purpose-built agents, pick OpenClaw.
3. Pick a plan
Every plan runs your chosen agent on an isolated instance with automated security patching and daily encrypted backups. Here's what each tier actually gets you, confirmed directly on Cloudways' pricing page:
| Plan | Price | vCPU | RAM | Storage | Bandwidth |
|---|---|---|---|---|---|
| Scout | $9.99/mo | 1 | 2GB | 50GB | 2TB |
| Operator | $19.99/mo | 2 | 4GB | 80GB | 3TB |
| Squad | $39.99/mo | 4 | 8GB | 160GB | 5TB |
| Swarm | $79.99/mo | 8 | 16GB | 320GB | 6TB |
In my testing, Scout was fine for a single Hermes instance answering questions in one Slack channel. It started to feel cramped once I pointed OpenClaw at three channels at once — Operator was the tier where response times stopped lagging under that load.
4. Connect your LLM provider
Paste in an API key from OpenAI, Anthropic, or Google. Cloudways doesn't bundle model access or credits — you're billed by the model provider directly, on top of the hosting fee. This is the line item people forget to budget for; a chatty agent running all day can cost more in tokens than the server does in hosting.
5. Connect a channel and set permissions
Add Slack, Discord, Telegram, or WhatsApp if you want the agent reachable there, or skip this and use the dashboard console. Set what the agent can do before you connect anything real — file access, shell commands, and outbound messages are the three permissions worth reviewing line by line rather than accepting the defaults.
6. Launch and verify over SSH
Click launch and the instance is live in under two minutes, per Cloudways' own setup claim, which matched what I saw both times I tried it. Full SSH and SFTP credentials appear in the dashboard, so you can log in directly if you need to debug a stuck process or read logs the web console doesn't surface.
Example prompts you can copy
These are close to what I sent each agent after setup, to confirm the channel connection and permissions actually worked:
- First message to a new OpenClaw instance: "Introduce yourself in this channel, then list the tools and integrations you currently have access to."
- Testing Hermes memory: "Remember that I prefer terse answers with no preamble. Confirm you've saved that, then answer my next three questions that way."
- Scoped task for OpenClaw: "Check this GitHub repo's open issues labeled 'bug' and summarize the three oldest ones. Do not comment on any issue — just report back."
- Recurring job for Hermes: "Every weekday at 9am, pull yesterday's closed support tickets and post a two-sentence summary here. Ask before you post anything longer."
Keep the first few prompts scoped and reversible. An agent that can only read and report can't do much damage while you're still learning what it does with ambiguous instructions.
Common mistakes to avoid
The mistake I made first: connecting a channel before reviewing permissions, which meant OpenClaw had file and shell access in a live Slack channel before I'd decided whether it should. Fix the permission scope first, connect the channel second. Second, picking Scout for anything beyond a single low-traffic channel — the $9.99 tier is real, but 1 vCPU runs out of headroom fast once an agent is juggling more than one conversation. Third, forgetting that the model provider bills separately; the hosting invoice looks small right up until the token bill arrives. Fourth, skipping the SSH credentials entirely — the dashboard console is fine for normal use, but when an agent hangs mid-task, direct server access is the fastest way to see what it was actually doing. Fifth, deleting an agent instance without checking Cloudways' retention window: deleted data is kept for 14 days and then it's gone for good, so don't delete on a whim if you might want the conversation history back.
Tools that make this easier
If OpenClaw or Hermes are your first AI agents, how to use ChatGPT agents is a gentler on-ramp that doesn't require picking a hosting plan first. For more on why permission scoping matters before you connect a live channel, see why AI agents lie, cheat, and steal — the reward-hacking behavior it covers is exactly why step 5 above isn't optional. If you'd rather run an agent in a disposable sandbox instead of a persistent managed server, Docker sandboxes for AI agents covers that alternative. Cloudways' one-click MCP integration is worth pairing with MCP in Claude if you want your agent talking to tools beyond chat channels. And if hosting fees aren't in the budget yet, my best AI tool for code roundup and free AI tools guide both cover no-cost or low-cost ways to get agent-assisted work done in the meantime.
Frequently Asked Questions
Is Cloudways Managed AI Agents free?
Not indefinitely. Cloudways ran a free public preview, but general availability, reached on August 12, 2026, moved the product to paid plans starting at $9.99/month. You still pay your model provider separately for API usage on top of that.
How long does it take to set up Cloudways Managed AI Agents?
Around two minutes for the instance itself, based on Cloudways' own claim and what I saw testing it twice. Budget more time than that for picking a plan, connecting your API key, and — more importantly — deciding what permissions the agent should have before you connect a live channel.
What is the easiest way to get started?
Start with Hermes on the Scout plan if you want one assistant in one channel. It's the smallest resource commitment and the simplest mental model — one agent, one memory, no routing logic to configure like you'd need with OpenClaw's multi-agent setup.
Which AI agents does Cloudways support?
Two at general availability: OpenClaw, a multi-agent control plane, and Hermes, a single always-on runtime from Nous Research. Cloudways has said more open-source agents are coming, but as of this update those are the only two live options.
Do I need my own OpenAI or Anthropic API key?
Yes. Cloudways hosts the server the agent runs on, not the underlying model, so you connect your own key from OpenAI, Anthropic, or Google and pay that provider directly for tokens. There's no bundled model credit in any of the four hosting plans.