The practical minimum: 1 vCPU and 2 GB RAM

Scout is the practical entry point for OpenClaw on Cloudways. Its 1 vCPU and 2 GB RAM are intended for light checks and one or two simple workflows. That can cover a personal agent that receives occasional messages, performs basic lookups, runs modest scheduled tasks and keeps a limited amount of local state. It is also the right place to validate whether a workflow deserves a permanent server.

Two gigabytes should not be interpreted as a universal OpenClaw requirement. It is a tier for a particular workload shape. If several tools run together, the browser is active, local files grow or multiple messages arrive at once, the remaining headroom can disappear quickly. An agent that technically runs is not necessarily comfortable or responsive under the usage pattern you intend.

Use Scout when the goal is learning, testing or light production. If you already know the agent will serve a team, stay busy through the day or coordinate sub-agents, starting at Operator or Squad avoids optimizing around a minimum you are likely to outgrow immediately.

Why 4 GB RAM is a strong everyday baseline

Operator doubles the Scout compute to 2 vCPUs and 4 GB RAM. For an OpenClaw instance that is expected to be genuinely useful every day, this is often the more comfortable baseline. The extra memory gives room for the agent, helper processes, logs and overlapping activity, while the second vCPU improves the ability to handle concurrent work without everything competing for one processor share.

Four gigabytes does not guarantee smooth performance for every workload, but it reduces the chance that a normal daily routine is constantly close to the limit. An agent connected to one or two channels, running scheduled checks and performing browser or terminal actions can fit this profile. The deciding factor should be sustained activity, not whether the user considers the tasks simple.

If monitoring shows that memory and CPU remain low, you may be overprovisioned and can evaluate Scout. If the agent regularly launches several workflows or sub-agents, Operator may be too small. The goal is to have enough headroom for normal bursts without paying for a tier whose resources sit unused all month.

Why sub-agents and parallel workflows favor 8 GB

Squad provides 4 vCPUs and 8 GB RAM, and Cloudways explicitly associates it with multiple sub-agents coordinating on tasks. OpenClaw can be used as a broad agent that delegates work, so this tier matters once parallelism is part of the normal design. A single request can fan out into several jobs, and the server must handle the combined runtime rather than only the visible conversation.

Eight gigabytes is also useful when the agent connects to several busy channels while scheduled work continues in the background. The workload may consist of many individually small tasks, but concurrency turns them into meaningful aggregate demand. In that situation, the higher tier can improve responsiveness and reduce the need to queue everything sequentially.

Do not jump to Squad solely because you might use sub-agents someday. Start with the architecture you have. If delegation becomes important, track how often tasks overlap and whether Operator experiences pressure. Scaling should follow a change in real workload or an explicit production requirement.

Storage requirements for OpenClaw

Cloudways pairs 50 GB, 80 GB, 160 GB and 320 GB SSD storage with the four tiers. Many OpenClaw deployments will not be storage-heavy at first because model computation happens through an external provider. Storage is used by the agent environment, logs, local files, downloaded artifacts, skills and any other data your workflows keep on the server.

Storage can become important when the agent works with many documents, media files, repositories or long-running logs. If files are temporary, implement cleanup rather than buying a larger compute plan only for disk space. If the files are durable business data, consider whether they belong in dedicated object storage, a repository or another system that provides clearer retention and backup policies.

Backups protect the hosted environment, but storage planning should still include retention. Decide how long logs are kept, which generated files are disposable and which artifacts are authoritative. An agent server should not become an accidental archive simply because it has free disk space.

Messaging channel requirements are mostly about overlap

WhatsApp, Slack, Discord and Telegram do not each require a published fixed amount of OpenClaw RAM. Their effect is behavioral. More channels create more opportunities for requests to arrive and more sessions that may overlap with scheduled or tool-driven work. A private dashboard agent can be quiet for hours; a team agent in several channels may be active unpredictably throughout the day.

If channels are central to the use case, choose a plan with concurrency headroom. Operator is a reasonable starting point for regular multi-channel use, and Squad becomes more attractive when several people, channels and workflows can overlap. The calculator reflects this by increasing the score as connected channels rise rather than pretending a channel has a precise memory cost.

Permissions should be reviewed at the same time. Every channel adds credentials and a new user-access surface. Keep tokens private, restrict bot permissions to what is necessary and require confirmation for high-impact tool actions. Resource planning and access planning are both consequences of making the agent more reachable.

CPU requirements rise with tools and continuous activity

CPU matters when OpenClaw is actively executing local work, orchestrating tools, handling several tasks or running supporting processes. Model reasoning itself may happen through a remote API, but the agent still has to prepare requests, process results, manage tools and maintain its runtime. A 1 vCPU Scout instance is appropriate for light work; 2 or 4 vCPUs provide more room as concurrent activity increases.

Continuous use is a stronger signal than a short burst. A nightly job that briefly uses high CPU may be fine on a small tier if the rest of the day is idle. Several jobs plus interactive requests running every hour create a different profile. That is why Cloudways advises looking at usage over time before scaling and why this calculator asks about runtime pattern.

If CPU is the bottleneck, check whether the workload can be staggered or optimized before upgrading. If the activity is useful and genuinely parallel, more CPU is the correct answer. If one task loops or launches unnecessary processes, a larger server can hide inefficiency and increase cost without improving the workflow design.

A requirements checklist before you deploy OpenClaw

Before choosing a plan, write down the expected number of active workflows, maximum useful concurrency, messaging channels, scheduled jobs, sub-agents and local storage needs. Then identify whether the agent is experimental, an everyday personal tool or a production service. Those answers are more valuable than copying a generic minimum requirement from another user with a different workload.

Next, separate the Cloudways server requirement from the LLM requirement. You will need a supported model-provider API key, and the provider bills that usage separately. Decide how credentials are stored, who can rotate them and whether the model account has spending controls. The server can be perfectly sized while the overall deployment is still unsafe or financially uncontrolled.

Finally, plan how you will validate the choice after launch. Monitor resource use for several representative days, note the busy periods and record any latency or failures. If the workload remains comfortable, keep the plan. If sustained pressure appears, scale. If the agent is mostly idle, consider downsizing. Requirements become reliable when they are connected to measurement.

For production use, repeat the review whenever you add a major channel, tool or automation path. Requirements can change without the number of users changing at all. A new scheduled browser workflow or several new sub-agents can shift the server profile more than adding another occasional chat user, so capacity reviews should follow architecture changes as well as traffic growth.

OpenClaw production-readiness requirements beyond RAM

RAM and CPU are only part of a production-ready OpenClaw deployment. You also need a stable model-provider account, protected credentials, a recovery path, a plan for messaging-channel access and clear rules for tools that can change external systems. A server with generous resources can still be fragile if the agent depends on undocumented tokens or if nobody knows how to restore the workflow after an integration fails.

For team use, document ownership. Identify who can change the OpenClaw configuration, who manages Cloudways billing, who controls model spending and who can approve high-impact actions. If the agent is connected to Slack or another shared channel, decide which users may invoke sensitive workflows. These operational requirements are often more important to reliability than adding another vCPU.

Finally, test recovery before the agent becomes critical. Confirm that backups exist, important configuration is documented outside the instance and key integrations can be reconnected. A production requirement is not merely that the agent runs today; it is that the team can understand, restore and safely operate it when something changes.