OpenClaw VPS Requirements: How Much RAM, CPU & Storage Do You Need?
OpenClaw has become the dominant open-source AI agent framework in 2026 with 269K+ GitHub stars and over 300,000 active users. It runs 24/7 on your own server — browsing the web, executing shell commands, managing files, and connecting to platforms like Telegram, Discord, and Slack. But the official docs give you minimums without context. This guide breaks down exactly what hardware you need based on how you actually plan to use OpenClaw, backed by real deployment data.
The short answer: a basic text-only assistant runs fine on 2 vCPU / 4 GB RAM / 40 GB SSD. Browser automation doubles the RAM requirement. Running local AI models via Ollama quadruples it. Below, we cover every scenario in detail.
OpenClaw Minimum Server Requirements
The official OpenClaw documentation specifies these absolute minimums:
- CPU: 2 vCPU cores
- RAM: 4 GB
- Storage: 40 GB SSD
- OS: Ubuntu 22.04 or Debian 12
- Docker: Version 24 or higher
- Node.js: Version 22 or higher
These minimums assume a text-only assistant using cloud-based AI APIs (Anthropic, OpenAI, etc.) without browser automation or local models. The OpenClaw Docker image itself consumes 2–4 GB of disk space in the standard configuration. If you enable the Chromium browser automation layer, the image grows to 6–8 GB.
At these specs, OpenClaw will start and respond to queries, but you will have minimal headroom for OS processes, logging, and Docker overhead. For anything beyond a personal experiment, plan for more.
RAM Requirements by Use Case
RAM is the single most important resource for OpenClaw. CPU and storage matter, but running out of RAM causes the process to crash or get OOM-killed — and there is no graceful recovery. Here is what each deployment scenario actually consumes:
| Use Case | RAM | CPU | Storage | Est. Cost |
|---|---|---|---|---|
| Text-only assistant | 4 GB | 2 vCPU | 40 GB | $4–6/mo |
| Browser automation | 8 GB | 4 vCPU | 80 GB | $8–15/mo |
| Ollama 7B model | 16 GB | 4 vCPU | 100 GB | $15–30/mo |
| Ollama 13B model | 32 GB | 8 vCPU | 200 GB | $40+/mo |
Text-only assistant (4 GB): OpenClaw sends prompts to a cloud API and returns text responses. The Node.js process itself uses 200–500 MB. Docker overhead adds another 200–400 MB. The OS and background services need ~500 MB. With 4 GB total, you have comfortable headroom for logging, temporary spikes, and running docker compose updates without swapping.
Browser automation (8 GB): OpenClaw's browser skill launches a headless Chromium instance to interact with web pages. Chromium is notoriously memory-hungry — each session consumes 2–4 GB of RAM depending on the complexity of the pages being loaded. With OpenClaw itself using ~1 GB and the OS needing another ~1 GB, you need 8 GB minimum to avoid OOM kills. Do not attempt browser automation on a 4 GB VPS.
Ollama local models (16–32 GB): Running inference locally via Ollama eliminates API costs but demands significant RAM. A 7B parameter model (like Llama 3 8B or Mistral 7B) requires 8–16 GB of RAM just for the model weights. A 13B parameter model needs 16–32 GB. Add OpenClaw's own overhead, and you are looking at 16 GB minimum for 7B models and 32 GB for 13B models.
CPU: How Many Cores Does OpenClaw Need?
OpenClaw's CPU usage is bursty rather than sustained. Most of the time, the agent is waiting for API responses or user input, consuming near-zero CPU. Spikes occur during:
- Message processing: Parsing and formatting prompts — brief, single-core spikes
- Docker operations: Image pulls, container restarts — moderate multi-core load
- Browser automation: Chromium rendering is CPU-intensive, benefits from 4+ cores
- Ollama inference: Model inference is heavily parallelized, scales linearly with cores
For a text-only setup, 2 vCPU is sufficient. The Node.js event loop is single-threaded, so additional cores provide diminishing returns unless you are running concurrent tasks. Browser automation benefits meaningfully from 4 vCPU since Chromium's rendering engine parallelizes across cores. Ollama inference benefits from every core you can provide — 4 vCPU is the practical minimum, and 8 vCPU provides noticeably faster token generation for 13B models.
ARM vs. x86 matters here. ARM cores (like Oracle Cloud's Ampere A1) offer excellent performance-per-dollar for Node.js workloads. OpenClaw runs natively on ARM64 via Docker. If you are considering Oracle Cloud's free tier (4 ARM CPUs, 24 GB RAM, 200 GB storage), it is one of the strongest options for running OpenClaw with Ollama at zero cost.
Storage: SSD vs NVMe and How Much Space
OpenClaw's storage requirements are modest for the application itself but grow significantly with Docker images and local models. Here is the breakdown:
- OS + system packages: 5–8 GB
- Docker engine + layers: 2–3 GB
- OpenClaw Docker image (standard): 2–4 GB
- OpenClaw Docker image (with Chromium): 6–8 GB
- Ollama model files (7B): 4–8 GB per model
- Ollama model files (13B): 8–15 GB per model
- Logs, configs, conversation history: 1–5 GB over time
A text-only deployment fits comfortably on 40 GB. Browser automation needs 80 GB to accommodate the larger Docker image plus headroom for Chromium's temporary files and cache. Ollama deployments should plan for 100–200 GB depending on how many models you want stored locally.
NVMe is recommended over standard SSD. OpenClaw's Docker operations are I/O-sensitive — pulling images, building layers, and restarting containers all involve heavy sequential and random reads. NVMe drives deliver 3–5x the throughput of SATA SSDs, which translates to noticeably faster docker pull and docker compose up operations. On a SATA SSD, a full OpenClaw image pull can take 3–5 minutes; on NVMe, it completes in under a minute.
Avoid HDD-backed VPS plans entirely. The random I/O latency makes Docker operations painfully slow and can cause timeout issues during container startup.
Operating System and Software Requirements
OpenClaw officially supports two Linux distributions:
- Ubuntu 22.04 LTS (Jammy Jellyfish) — the most popular choice, best community support
- Debian 12 (Bookworm) — lighter footprint, equally stable
Both distributions ship with the required kernel version and package manager support for Docker and Node.js. Other Linux distributions may work but are not officially tested.
The software stack requires:
- Docker 24+ — Install via the official Docker repository, not the distro's older package. Run
curl -fsSL https://get.docker.com | shfor the latest version. - Docker Compose v2 — Included with Docker 24+ as
docker compose(no hyphen). OpenClaw'sdocker-compose.ymlrelies on Compose v2 features. - Node.js ≥ 22 — Required if running OpenClaw outside Docker. Use
nvm install 22or the NodeSource repository. The Docker image bundles its own Node.js, so this is only needed for bare-metal installs. - Git — For cloning the repository and pulling updates.
Windows and macOS are supported for local development only. Production deployments should always use Linux for stability, performance, and Docker compatibility.
Setup and Maintenance Time
Plan for 4–8 hours for initial setup, including VPS provisioning, OS hardening, Docker installation, OpenClaw configuration, and testing. Ongoing maintenance averages 2–4 hours per month for updating Docker images, reviewing logs, applying OS security patches, and managing API keys. See our security checklist for a hardening guide.
Recommended VPS Specs by Budget
Based on our testing across 40+ VPS providers, here are the best value configurations for each OpenClaw deployment tier:
| Budget | Specs | Best For | Example Providers |
|---|---|---|---|
| $4–6/mo | 2 vCPU, 4 GB RAM, 40 GB SSD | Text-only personal assistant | Hetzner CX23, Contabo, InterServer |
| $8–15/mo | 4 vCPU, 8 GB RAM, 80 GB NVMe | Browser automation, multi-platform | Contabo, OVHcloud, InterServer |
| $15–30/mo | 4 vCPU, 16 GB RAM, 100 GB NVMe | Ollama with 7B models | Hetzner CPX31, Contabo VPS M |
| $40+/mo | 8 vCPU, 32 GB RAM, 200 GB NVMe | Ollama with 13B models, production | Hetzner CPX41, Contabo VPS L |
| Free | 4 ARM CPU, 24 GB RAM, 200 GB | Ollama 7B, browser automation | Oracle Cloud Always Free Tier |
The Oracle Cloud free tier deserves special mention: 4 ARM Ampere CPUs, 24 GB RAM, and 200 GB storage at zero cost. This is enough to run OpenClaw with Ollama 7B models. The catch is availability — free-tier instances are capacity-limited and may require repeated provisioning attempts. But if you can get one, it is the best deal in the market by a wide margin.
For all other tiers, use our OpenClaw VPS comparison tool to find the cheapest plans that meet your exact requirements. Plans are pre-filtered for OpenClaw compatibility and sorted by price.
Ready to deploy? Compare VPS plans pre-filtered for OpenClaw requirements.
Browse OpenClaw VPS Plans →Frequently Asked Questions
Can I run OpenClaw on 2 GB RAM?
Technically, OpenClaw can start on 2 GB, but the official minimum is 4 GB RAM for a reason. With 2 GB, the OS, Docker daemon, and OpenClaw's Node.js process leave almost no headroom. Any spike — a large API response, a log rotation, or a Docker image update — can trigger the OOM killer and crash your agent. Stick with 4 GB minimum for a stable deployment.
Does OpenClaw need a GPU?
No. OpenClaw itself does not use GPU acceleration. It runs entirely on CPU. However, if you add Ollama for local AI inference, a GPU dramatically speeds up token generation. That said, Ollama works on CPU-only servers — it is just slower (roughly 2–5 tokens/second for a 7B model on 4 vCPU versus 30–50 tokens/second on a mid-range GPU). Most VPS providers do not offer GPU instances at reasonable prices, so CPU-only Ollama is the practical choice for self-hosting.
How much bandwidth does OpenClaw use?
Very little. A text-only OpenClaw instance uses 1–5 GB of bandwidth per month in normal operation. The largest bandwidth consumers are Docker image pulls (2–8 GB per update) and Ollama model downloads (4–15 GB per model). Browser automation adds more variable bandwidth depending on the pages being loaded. Any VPS plan with 1 TB/month or more is more than sufficient. Bandwidth is almost never the bottleneck for OpenClaw deployments.
Can I run OpenClaw on a free VPS?
Yes. Oracle Cloud's Always Free tier offers 4 ARM Ampere CPUs, 24 GB RAM, and 200 GB of storage at zero cost — permanently, not just a trial. This is powerful enough for OpenClaw with browser automation and even Ollama 7B models. The main challenge is provisioning: free-tier capacity is limited by region, and you may need to try multiple regions or retry during off-peak hours. Google Cloud and AWS also offer free tiers, but their free instances (e2-micro / t2.micro) have only 1 GB RAM, which is insufficient for OpenClaw.