OpenClaw Minimum Requirements: RAM, CPU, and Storage
What are OpenClaw's system requirements in 2026? Minimum and recommended RAM, CPU, and storage for different configurations including local LLMs.
Understanding OpenClaw's system requirements helps you choose the right hardware or cloud instance. Requirements vary significantly depending on whether you're using cloud LLM APIs (lightweight) or running local models via Ollama (memory-intensive). This guide covers both scenarios.
Cloud LLM Configuration (Minimum Setup)
When using external LLM APIs (Anthropic, OpenAI, DeepSeek, etc.), OpenClaw itself is lightweight:
| Requirement | Minimum | Recommended |
|---|---|---|
| RAM | 512MB | 1–2GB |
| CPU | 1 vCPU | 2 vCPU |
| Storage | 5GB | 20GB |
| Network | 10 Mbps | 50 Mbps |
| OS | Linux/macOS/Windows | Ubuntu 24.04 LTS |
A $5/month cloud VPS is sufficient for a single OpenClaw instance using cloud APIs.
Local LLM Configuration (Ollama)
Running local models requires much more memory. The entire model must fit in RAM (or VRAM for GPU-accelerated setups):
| Model Size | RAM Required | Examples |
|---|---|---|
| 3B params (Q4) | 4GB | Phi-3.5 mini, Llama 3.2 3B |
| 7B params (Q4) | 8GB | Mistral 7B, Llama 3.1 8B |
| 13B params (Q4) | 16GB | — |
| 34B params (Q4) | 24GB | CodeLlama 34B |
| 70B params (Q4) | 48GB | Llama 3.1 70B |
Add 1–2GB on top of the model RAM requirement for the OS, OpenClaw, and Ollama overhead.
Storage Requirements
| Component | Storage |
|---|---|
| OpenClaw core | ~500MB (venv + code) |
| Per-skill | 10–100MB |
| Agent memory (1 year of use) | 500MB–5GB |
| Local LLM model (7B Q4) | ~4GB |
| Local LLM model (70B Q4) | ~40GB |
Budget storage generously if you plan to run local models.
CPU vs GPU for Local Models
CPU only: Works well for 7B models and smaller, producing 10–40 tokens/second on a modern CPU (acceptable for personal use).
GPU (NVIDIA): Dramatically faster — a midrange GPU runs 7B models at 100+ tokens/second. RTX 3060 12GB handles 13B models well.
Apple Silicon: Excellent for local LLMs. M4 Pro with 48GB unified memory handles 70B models well via Metal acceleration.
Frequently Asked Questions
Can OpenClaw run on a Raspberry Pi?
Yes for cloud LLMs — a Raspberry Pi 4 (4GB RAM) runs OpenClaw adequately. Expect slightly sluggish web terminal performance. Not suitable for local models.
Does OpenClaw use much CPU at idle?
With cloud LLMs, OpenClaw is essentially idle between messages — CPU usage is under 1% when waiting for input. The LLM API calls themselves are I/O-bound, not CPU-bound.
What's the smallest VPS that works reliably?
A 1GB RAM VPS works for single-instance OpenClaw with cloud LLM. 2GB is recommended for comfort and to handle memory spikes during complex skill executions.
nacre.sh
Run OpenClaw without the server headaches
Dedicated instance, automatic TLS, nightly backups, and 290+ LLM integrations. Live in under 90 seconds from $12/month.
Deploy your agent →