Cloudways Brings Managed OpenClaw and Hermes Deployments to General Availability

That combination matters because an agent is not just a static application. It may run persistently, use tools, retain state, and interact with external systems. The infrastructure around it therefore becomes part of the product experience—and part of the risk profile.

Who the service is for

Cloudways is targeting developers, engineers, agencies, and businesses, including customers already using the platform for application hosting, e-commerce, or agency work. The intended user is not necessarily unable to run OpenClaw or Hermes manually; rather, it may not want to own the recurring operational work required to keep an agent available and maintained.

That makes the launch a bet on the prototype-to-production gap. A team can experiment with an open-source agent independently, then use managed infrastructure when the experiment needs to become a dependable service or internal workflow.

Why OpenClaw and Hermes matter

Cloudways said its two launch agents together represented more than 614,000 GitHub stars at launch: approximately 386,000 for OpenClaw and 228,000 for Hermes. Those figures show substantial developer interest, but GitHub stars are time-sensitive signals of visibility and community attention—not proof of security, reliability, or production readiness.

Hermes Agent, launched by Nous Research on February 25, 2026, is presented as a self-improving open-source agent with a built-in learning loop. Its repository describes capabilities including creating skills from experience, improving them during use, persisting knowledge, and searching past conversations.

OpenClaw’s rapid adoption has also been accompanied by a documented security debate. Reporting has described exposed API tokens, remote-code-execution vulnerabilities, and malicious community skills. The project later moved under the OpenClaw Foundation, a 501(c)(3) nonprofit. The foundation says OpenAI supports inference and security work through Codex Security and Claw Labs, while NVIDIA has introduced NemoClaw with the OpenShell secure runtime and open Nemotron models.

That history explains why Cloudways emphasizes isolation and validated updates. Those controls can reduce infrastructure and lifecycle friction, but they should not be interpreted as a guarantee that an agent, its tools, models, extensions, or permissions are safe.

Cloudways’ message and planned expansion

Suhaib Zaheer, Cloudways’ senior vice president and general manager, framed the launch as part of a broader effort to make AI infrastructure simpler and more accessible. The company’s stated goal is to make running an agent feel closer to deploying another managed workload than assembling and maintaining an infrastructure stack from scratch.

Cloudways also said it plans to add more open-source agents and capabilities over time. The initial catalog is therefore as important as the underlying hosting model: the company is testing whether users want a managed entry point for popular open-source agents, rather than only fully proprietary assistants.

The middle ground in the agent market

Cloudways is positioning Managed AI Agents between two established approaches:

1. Self-hosting, which provides control over infrastructure, software, and deployment choices but leaves customers responsible for configuration, security, updates, and reliability.
2. Vertically managed proprietary agents, which simplify operations but may impose tighter limits on platform access, account requirements, geography, or customization.

Google’s Gemini Spark illustrates the second model. Google’s documentation says Spark is unavailable in the European Economic Area, Nigeria, Switzerland, and the United Kingdom, and currently requires a personal Google Account rather than a work or school account.

Cloudways’ proposition is different: users retain access to open-source agents while delegating much of the hosting lifecycle. The practical value is operational convenience—not a claim that OpenClaw or Hermes automatically become enterprise-ready. Teams still need to evaluate permissions, data access, extensions, model providers, update policies, and their own security requirements before putting an agent near sensitive systems.