Cloudways has moved Managed AI Agents out of public preview and into general availability. You pick OpenClaw or Hermes, choose a region and a tier, plug in your own LLM API key, and Cloudways provisions a dedicated instance and keeps it patched and backed up — from the same dashboard you would use to launch a WordPress server. That is a different proposition from the preinstalled-on-a-VPS agent offers that showed up earlier this year, because the ongoing upkeep genuinely sits on the provider's side. Here is what is included, what is not, and who this is actually worth paying for.
What Cloudways Managed AI Agents is
This is a hosting product for agent runtimes rather than websites. Deployment is a four-step flow in the dashboard: select the agent, choose region and tier, connect your LLM API keys, then configure and deploy. Cloudways states there is no server configuration, no Docker and no command-line work involved, with the instance live in minutes.
You bring your own model keys — OpenAI, Anthropic and Google Gemini are supported — so the LLM relationship stays yours. Once running, the agent connects to Slack, Discord, Telegram and WhatsApp directly from the dashboard, which is the part that turns an agent from something you poke at over SSH into something a team can talk to. Full details are in Cloudways' general-availability announcement.
OpenClaw and Hermes: two different shapes of agent
The two supported runtimes are not interchangeable. Cloudways positions OpenClaw as a multi-agent control plane — multi-channel routing for managing persistent teams of specialised agents. In other words, several named agents with different jobs, each reachable on a different channel. Hermes, an open-source runtime from Nous Research, is the opposite bet: a single always-on runtime with closed-loop learning, intended to become more efficient over time, and built for safe, isolated multi-agent work.
The practical split is straightforward. If you want a roster of agents that each own a task and answer in their own channel, OpenClaw fits. If you want one agent running continuously against a single domain and improving as it goes, Hermes fits. Both sit on the same managed infrastructure with the same patching, LLM integrations and channel support, so trying the other one is a redeploy rather than a migration.
If you have been following agent hosting since spring, the useful contrast is HostArmada's Openclaw Hosting — same runtime, preinstalled on a VPS, but explicitly self-managed once the server is up.
Decide on the job first, the hosting second
The reason to pay for hosted agent infrastructure is not the hosting. It is having a job worth running around the clock. An agent on your laptop stops when the lid closes, drops its channel connections, and cannot be handed to a colleague — fine for experiments, useless for anything anyone depends on.
The workloads that justify a permanent instance tend to look like content operations (drafting, repurposing and scheduling on a fixed cadence), competitor or price monitoring that only pays off if it runs daily, a lightweight team CRM that watches inboxes and channels, or server operations such as log triage and alert summaries. Pick one, confirm you would genuinely miss it if it stopped, and the $9.99–$79.99/mo question answers itself. Skip that step and you are paying rent on a demo.
What is managed — and what is not
On Cloudways' side: security patching for both the agent software and the underlying infrastructure, automated backups on every deployment, validated agent version updates, managed SSL/TLS, and SSH/SFTP access as standard. That is the meaningful gap versus a VPS with the runtime preinstalled, where the initial setup is automated but every patch afterwards is yours.
What stays on you deserves to be said plainly. The LLM bill is not in the tier price — you supply the keys, and for a chatty always-on agent the token spend can comfortably exceed the hosting line. Scope and permissions are yours: what the agent may touch, which credentials it holds, which channels it can post into. And an always-on agent holding API keys with Slack or WhatsApp access is, in security terms, a production server with a chat interface. Treat it like one.
Plans and pricing
Cloudways lists four tiers, each with dedicated resources and no bundled LLM credits:
| Plan | Price | vCPU | RAM | Storage | Bandwidth |
|---|---|---|---|---|---|
| Scout | $9.99/mo | 1 | 2 GB | 50 GB | 2 TB |
| Operator | $19.99/mo | 2 | 4 GB | 80 GB | 3 TB |
| Squad | $39.99/mo | 4 | 8 GB | 160 GB | 5 TB |
| Swarm | $79.99/mo | 8 | 16 GB | 320 GB | 6 TB |
Scout is enough to run a single agent and find out whether the job holds up. OpenClaw setups with several agents across multiple channels, or anything that spawns parallel work, will want Squad or above. The number of agents per instance is not capped by tier — what limits you is the RAM and vCPU on the plan.
One timing note worth checking before you commit: the free usage period from the public preview has ended and standard pricing is now in effect, with discounted rates for instances that were deployed during the preview. Cloudways is separately promoting a 50% launch discount held for the lifetime of the agent, though the public pricing page currently shows the rates above with no discount badge. Confirm which price your checkout actually applies — the pricing has already changed once, when the product left preview.
Also new: per-application PHP version control
Not agent-related, but it closes a long-standing Cloudways annoyance. PHP version used to be a server-level setting, so every application on a server shared one. You can now set the PHP version per application, included in existing plans at no extra cost.
The rollout has edges. It is live for servers launched from 8 August 2026 only; existing servers get it in a later phase, once the required network-level changes are done. Per-application PHP extensions and settings are not part of this release, and there are no audit logs for version changes at launch. If you run a legacy app pinned to an old PHP release next to something modern, this is the change that finally lets both share a server — but for now only on a new one. Cloudways documents the steps in its help centre.
Also new: Cloudflare Enterprise free for 30 days
Cloudways' Cloudflare Enterprise add-on — enterprise-grade CDN, WAF and DDoS protection wired into the platform — is currently offered free for 30 days. If you have wondered whether the add-on measurably changes your TTFB and bot traffic, a month is long enough to judge it against your own analytics instead of a vendor benchmark. Note the add-on's regular price before the trial lapses so the first renewal is not a surprise.
Summary
The notable part here is not that another host will run an agent for you. It is that upkeep of the agent runtime is now the provider's problem — managed patching, validated runtime updates and backups are precisely the work nobody wants to do by hand at 2am, and precisely what a preinstalled-on-a-VPS offer leaves on your plate.
Two things to stay clear-eyed about: the LLM spend sits outside the plan price and is the number that actually scales with use, and launch pricing has already shifted once. If you have a job worth running continuously, Scout at $9.99/mo is a cheap way to find out whether it survives contact with production.

