Best VPS for OpenClaw: requirements and which KVM plan to pick

With a local model, which runs on the server itself, the math changes. The documentation sets a floor of 8 GiB of memory (about 8.6 GB) for its smallest local recipe, with a 64K context. KVM 2, with 8 GB, sits right at the limit. KVM 4, with 16 GB, is the first with room to spare.

If you don't want to look after a server, look at the managed plan. With the link it costs US$3.99/month over the first 24-month period, while KVM 1 costs US$6.49/month without the link (with the link, its price is shown in the cart). Hostinger installs and updates the assistant.

The official installation and local-model pages, read on October 21, 2026, say the following:

• Node version. OpenClaw needs Node 24.16 or newer, or Node 26.1 or newer. Node is the program that runs OpenClaw, and the npm installation requires that version. The page recommends Node 26. If Node is not installed, the official installer brings the version (Node 26 on macOS and Node 24 LTS on Linux).
• Local model. For the smallest recipe offered by the documentation itself, the floor is 8 GiB of host memory, with a 64K context. The page says those floors do not guarantee that the model fits or runs at a useful speed.
• Hostinger VPS. Hostinger installs OpenClaw on the VPS with Docker, a program that runs each application isolated in its own container. The VPS is a rented virtual server; with root access, you have full control of it.

It gives no RAM, processor (vCPU) or disk number for OpenClaw without a local model, nor a server minimum for the API key or credits scenario.

A detail on the units: 8 GiB is about 8.6 GB. KVM 2 is advertised with 8 GB, which is about 7.5 GiB, a bit below the floor, and that does not count the memory the system and Docker use. The documentation does not say whether this difference in units counts toward the floor.

RAM is the server's working memory: it is where the program and, in the case of a local model, the model itself stay while running. The documentation gives a number only for the local model.

Two details about the local model. The context is the amount of text (tokens, pieces of words) the model reads at once. When less than 20% of the context window is left, OpenClaw warns, with a floor of 8,000 tokens; below 10%, it blocks the reply, with a floor of 4,000 tokens.

In the KVM plan pages I read, none lists a graphics card (GPU). The documentation also does not say what speed a local model would have without one.

KVM is the name of Hostinger's virtual server plans. The table has the four, with prices for 24-month plans read on October 21, 2026. The price with the referral link is shown in the cart, because we did not read it on the global page. Read each plan against the documentation:

Enlarge

• KVM 1 (4 GB). Starting point for the API key or credits scenario. For a local model, it is below the official floor.
• KVM 2 (8 GB). With an API key, there is room to spare. With a local model, it sits at the floor's limit, with no margin for the system.
• KVM 4 (16 GB). The first plan with room above the official 8 GiB floor. The documentation does not say how well it performs with a local model.
• KVM 8 (32 GB). Only makes sense with a larger local model. The documentation does not say how much memory each large recipe needs.

The referral link gives 20% off the first period only, on plans of 12 months or more, and it does not stack with other coupons. Renewal is charged at the normal price. See how Hostinger's referral link works and check the amount in the cart before paying. Hostinger is running a Black Friday campaign and prices change: see the campaign.

The three paths run the same program, OpenClaw, with different costs and chores. The choice depends on three things: whether you want to deal with a server, whether the computer can stay on 24 hours a day, and whether the AI model is cloud-based or runs on your own equipment.

On the VPS, Hostinger installs OpenClaw with Docker, a program that runs each application isolated in a container. From then on, maintenance is yours.

Oracle has a free plan, Always Free. It is not from Hostinger and does not replace the paid VPS. According to Oracle's documentation, read on October 21, 2026:

• Up to two Micro instances (AMD processor), with 1/8 OCPU and 1 GB of memory each.
• Arm Ampere A1 instances: 2 OCPUs and 12 GB of memory in total, which you split as you like. The monthly limit is 1,500 OCPU-hours and 9,000 GB-hours.
• 200 GB of block storage in total, for the system disks and extras, with up to five backups.
• Free instances must be created in the account's home region. Creating them elsewhere generates charges.

The rule that matters: Oracle says it can reclaim an idle instance. It is considered idle when, over a 7-day period, CPU usage at the 95th percentile stays below 20%, network usage stays below 20% and, on the A1 instances, memory usage also stays below 20%. An assistant that sits idle most of the time can fall under this rule. The documentation does not describe a notice before reclamation.

Prices of 24-month plans, read on October 21, 2026: managed costs US$3.99/month with the link. The VPS prices with the link are shown in the cart; without the link, KVM 1 is US$6.49 and KVM 2 is US$8.79. On renewal, managed and KVM 1 go to US$11.99/month, and KVM 2 to US$14.99/month.

The rule: the 20% applies only to the first period, on plans of 12 months or more, and does not stack with other coupons. There is no code to type, because the discount goes in the link. Check the amount in the cart before paying, because Black Friday changes prices. See how Hostinger's referral link works and Hostinger's page on the site.

Em Casa com Cecília runs on Hostinger, as a Node.js Web App, and Bruno has been a customer since 2024, with hosting and a KVM 2 VPS. No one on the team uses OpenClaw: this text gathers what Hostinger, Oracle and OpenClaw's official documentation say, read on October 21, 2026.