OpenClaw Hardware
Requirements

Everything you need to know to self-host OpenClaw — the open-source AI assistant gateway that connects Telegram, Slack, WhatsApp, Discord, iMessage, and 40+ other channels to 60+ AI model providers.

View Requirements ↓ GitHub Repo ↗
v22.19+
Node.js
Required
2 GB
Min RAM
(practical)
60+
Model
Providers
Docker
amd64 &
arm64

✅  The short answer: OpenClaw is remarkably lightweight

OpenClaw acts as a gateway — it routes messages and manages AI agent sessions but does not run model inference locally by default. When you connect a cloud provider like Anthropic, OpenAI, or Groq, no GPU is required. Most modern laptops, low-cost VPS instances, and even Raspberry Pi boards handle it comfortably.

Quick Answer

Hardware needed by deployment scenario
Deployment scenario Min RAM GPU needed? Install time
Cloud API (Anthropic, OpenAI, etc.) 2 GB No GPU ~5 min
Raspberry Pi (Pi 4 / Pi 5) 4 GB (rec.) No GPU ~30 min
VPS / cloud server (Docker) 2–4 GB No GPU ~10 min
Local model inference (Ollama) 8–16 GB+ Optional GPU Varies
About

What is OpenClaw?

Your own personal AI assistant — any OS, any platform, the lobster way.

Official description

"Self-hosted gateway that connects Discord, Google Chat, iMessage, Matrix, Microsoft Teams, Signal, Slack, Telegram, WhatsApp, Zalo, and more to AI coding agents." — github.com/openclaw/openclaw

OpenClaw is a self-hosted gateway and agent runtime written in Node.js. It acts as the central control plane for your AI setup — bridging your existing messaging apps to any of 60+ model providers, while supporting local skills, voice, a live canvas, and autonomous multi-agent workflows.

40+ Messaging Channels

WhatsApp, Telegram, Slack, Discord, iMessage, Signal, Matrix, Microsoft Teams, Google Chat, IRC, LINE, and many more.

60+ Model Providers

Anthropic, OpenAI, Google Gemini, Groq, OpenRouter, Mistral, DeepSeek, xAI, Ollama, and dozens of cloud and local options.

Fully Self-Hosted

Your data stays on your hardware. No third-party cloud relay. Run on your laptop, server, Pi, or any VPS provider.

Cross-Platform

macOS (native), Linux (systemd), Windows (WSL2). Docker images for linux/amd64 and linux/arm64 via GitHub Container Registry.

Voice & Canvas

Wake-word voice detection and continuous voice mode on macOS/iOS/Android, plus a live interactive Canvas workspace.

Skill Ecosystem

Extend with skills from ClawHub — the official skill directory. Browser automation, integrations, persistent memory, and more.

Hardware Requirements

Minimum & Recommended Specs

Requirements for running the OpenClaw gateway with a cloud model provider. Local model inference requires additional resources — see the Local Model section below.

⚠️ Transparency note on spec sources

OpenClaw's official documentation specifies hardware requirements for Raspberry Pi deployments. Specs for general desktop and VPS use are practical estimates derived from the Pi documentation and community experience — they are clearly labeled below. Where the official docs give a number, we quote it.

Minimum Requirements Bare minimum
  • Node.js v22.19+  Official
  • RAM 2 GB  Est.
  • CPU cores 1 core  Official
  • Disk free 500 MB  Official
  • OS 64-bit  Official
  • Storage SSD preferred  Est.
  • Internet Required  Est.
  • API key For cloud providers  Est.
Recommended Comfortable
  • Node.js v24  Official
  • RAM 8–16 GB  Est.
  • CPU cores 4+ cores  Est.
  • Disk free 10 GB+ SSD  Est.
  • OS 64-bit  Official
  • Storage SSD (NVMe ideal)  Est.
  • Internet Broadband  Est.
  • Multi-agent 32 GB RAM  Est.

Raspberry Pi Hardware Compatibility Official docs

Model RAM Status Notes
Raspberry Pi 5 4–8 GB Best Fastest Pi option; plenty of headroom for multi-channel + skills
Raspberry Pi 4 4 GB Good Sweet spot for most users; smooth performance
Raspberry Pi 4 2 GB OK Works with swap enabled; may feel sluggish under load
Raspberry Pi 4 1 GB Tight Technically possible with swap; not recommended for production
Raspberry Pi 3B+ 1 GB Slow Functional but noticeably slow; 64-bit OS required
Raspberry Pi Zero 2 W 512 MB Not rec. Insufficient RAM; not recommended

Pi tip: Use a USB SSD instead of an SD card for dramatically better startup speed and reliability. The official docs specifically recommend this. Raspberry Pi install guide →

Local Model Inference

Power-User & Local LLM Requirements

Running AI models directly on your hardware via Ollama, LM Studio, vLLM, or SGLang. Requirements scale with model size.

ℹ️ Local inference is completely optional

OpenClaw supports Ollama, LM Studio, vLLM, and SGLang as local inference backends — but you do not need to run local models. All 60+ cloud providers work without any additional hardware beyond what the gateway itself needs. Local inference is a power-user option for privacy, offline use, or cost optimization at scale.

Tier 1
Small Models
1B – 3B parameters
RAM 4–8 GB
Storage 2–4 GB/model
GPU Not required
Examples Llama 3.2 1B, Phi-3 mini
Tier 2 · Popular
Mid Models
7B – 8B parameters
RAM 8–16 GB
Storage 4–8 GB/model
GPU Optional (faster)
Examples Llama 3.1 8B, Mistral 7B
Tier 3
Large Models
13B – 34B parameters
RAM 16–32 GB
Storage 8–20 GB/model
GPU Strongly rec.
Examples Llama 3.3 70B (Q4), Mixtral
Tier 4
Very Large Models
70B+ parameters
RAM 64 GB+
Storage 40–80 GB/model
GPU Required (VRAM)
Examples Llama 3.1 405B, DeepSeek V3

All tier specs above are Practical estimates — model sizes vary by quantization level. Apple Silicon (M-series) is excellent for local inference, leveraging unified memory efficiently. NVIDIA GPUs with CUDA provide the fastest dedicated inference on Linux/Windows.

Docker

Docker Requirements & Quick Start

Official Docker images are published to GitHub Container Registry and support both x86-64 and ARM64 architectures.

Quick start

# Pull the latest slim image
docker pull ghcr.io/openclaw/openclaw:latest

# Or the slim variant (smaller footprint)
docker pull ghcr.io/openclaw/openclaw:latest-slim

# Alternatively, install via npm
npm install -g openclaw@latest
openclaw onboard --install-daemon

Image registry: ghcr.io/openclaw/openclaw ↗

Supported architectures Official

Architecture Tag suffix Use case
linux/amd64 -amd64 Standard x86-64 servers, most VPS
linux/arm64 -arm64 Apple Silicon, AWS Graviton, Raspberry Pi 64-bit

Image variants

Tag Description
:latest Latest stable release (full)
:latest-slim Lightweight variant — smaller image size
:YYYY.M.D Pinned date-versioned release
:YYYY.M.D-beta.N Pre-release / beta builds

Docker is the recommended production deployment method

Docker provides the easiest path for cloud deployments on DigitalOcean, Oracle Cloud, Hetzner, Fly.io, Railway, Render, Azure, GCP, and other providers officially documented by OpenClaw. The same RAM and CPU estimates from the requirements section apply when running the container.

Platform Support

Windows, macOS & Linux Compatibility

OpenClaw supports all major desktop and server operating systems, with different feature availability per platform.

macOS
Full native support
  • Native install via npm
  • Menu bar companion app
  • Voice Wake detection
  • Live Canvas workspace
  • launchd daemon supervision
  • Apple Silicon (arm64) native
  • Docker (colima / Docker Desktop)
Linux
Full native support
  • Native install via npm
  • systemd service supervision
  • Docker (linux/amd64 + arm64)
  • Raspberry Pi officially supported
  • Kubernetes & Podman
  • VPS / cloud server deployment
  • Ansible & Nix provisioning
Windows
Via WSL2
  • WSL2 required (Windows Subsystem for Linux)
  • Install OpenClaw inside WSL2 Linux
  • Docker Desktop (WSL2 backend)
  • Windows Scheduled Tasks for daemon
  • Native Windows install not officially supported
  • Menu bar / Voice Wake not available

OS support details sourced from the official OpenClaw repository and documentation.

Model Providers

Cost-Saving Options & Provider Guide

OpenClaw supports 60+ model providers. Cloud APIs need no additional hardware — just an API key.

💡 No GPU needed for cloud providers

Every cloud API provider listed below handles inference on their own hardware. Your OpenClaw gateway only processes routing and channel messages — a 2 GB RAM VPS or even a Raspberry Pi is sufficient when using these providers.

Cloud API providers

All providers below work without a GPU. Hardware requirements are determined by the gateway only.

Anthropic Claude OpenAI Google Gemini Groq ⚡ Free tier OpenRouter 🔀 200+ models Together AI Fireworks AI Mistral DeepSeek xAI Grok Perplexity Venice (privacy) Azure OpenAI AWS Bedrock Cohere + 45 more

Local inference providers

These require additional hardware — see the Local Model Requirements section.

Ollama LM Studio vLLM SGLang

Cost-saving strategies

Strategy Provider / approach Benefit
Use a model gateway OpenRouter Single API key for 200+ models; compare cost per token easily
Fast free inference Groq Generous free tier; LPU-based inference is extremely fast
Use efficient models Claude Haiku, GPT-4o-mini, Gemini Flash 10–50× cheaper than flagship models for routine tasks
Competitive pricing DeepSeek Very low per-token cost with strong benchmark performance
Local inference Ollama + small model Zero ongoing cost; requires capable local hardware

See the full provider list at docs.openclaw.ai/providers ↗

FAQ

Frequently Asked Questions

Answers to the most common OpenClaw hardware and setup questions.

What are the minimum hardware requirements to run OpenClaw?
OpenClaw's officially documented minimum for Raspberry Pi is 1 CPU core, 1 GB RAM, 500 MB free disk space, and a 64-bit OS (Node.js v22.19+). For a comfortable desktop or VPS deployment, a practical minimum is 2 GB RAM, 2 CPU cores, and 2 GB free SSD space. You also need a stable internet connection and an API key if using cloud model providers.
Does OpenClaw require a GPU?
No. When you use cloud-based model providers — Anthropic Claude, OpenAI, Google Gemini, Groq, or any of the 60+ supported providers — all AI inference runs on the provider's servers. Your machine only handles the gateway routing. A GPU is only needed if you choose to run local AI models via Ollama, LM Studio, vLLM, or SGLang directly on your hardware.
Can I run OpenClaw on a Raspberry Pi?
Yes — OpenClaw has official Raspberry Pi documentation. The recommended configuration is a Pi 4 or Pi 5 with 4 GB+ RAM. A Pi 4 with 2 GB works with swap enabled. The Pi Zero 2 W (512 MB) is not recommended. Always use 64-bit Raspberry Pi OS, and prefer a USB SSD over an SD card for better performance and longevity.
Does OpenClaw support Docker?
Yes. Official Docker images are published at ghcr.io/openclaw/openclaw. Both standard and slim variants are available. Supported container architectures are linux/amd64 (x86-64 servers, most VPS) and linux/arm64 (Apple Silicon, AWS Graviton, Raspberry Pi 64-bit). Pull with: docker pull ghcr.io/openclaw/openclaw:latest
What Node.js version does OpenClaw require?
OpenClaw requires Node.js v22.19 or higher. Node.js v24 is the officially recommended version for best performance and compatibility. The project uses modern ESM and TypeScript, so a recent LTS release is essential. If installing from npm, run: npm install -g openclaw@latest
Does OpenClaw run on Windows?
Yes, via WSL2 (Windows Subsystem for Linux 2). Native Windows execution is not officially supported. Install WSL2 first, then install OpenClaw inside your Linux distribution. Docker Desktop for Windows (which runs on WSL2) also works for the containerized version. macOS and Linux are fully supported natively.
What AI model providers does OpenClaw support?
OpenClaw supports 60+ model providers. Cloud providers include Anthropic Claude, OpenAI, Google Gemini, Groq, OpenRouter, Together AI, Fireworks AI, Mistral, DeepSeek, xAI Grok, Perplexity, Venice, and many more. For local inference, it supports Ollama, LM Studio, vLLM, and SGLang. OpenRouter acts as a model gateway giving access to 200+ models through a single API key.
How much disk storage does OpenClaw need?
The official documentation notes 500 MB as the minimum for Raspberry Pi installs. In practice, plan for 2–5 GB to comfortably accommodate the runtime, logs, skill installations, and cached files. If running local AI models via Ollama, each model requires additional storage: small models (1–3B) need 1–4 GB, 7B models need 4–8 GB, and larger models (70B) can need 40–80 GB.
Can I self-host OpenClaw on a VPS or cloud server?
Yes. OpenClaw officially documents deployment on DigitalOcean, Oracle Cloud, Hetzner, Fly.io, Railway, Render, Azure, GCP, and more. The documentation recommends SSD-backed storage for better startup performance. The Docker image (ghcr.io/openclaw/openclaw) is the easiest path for cloud deployments, with both amd64 and arm64 variants available.
What is OpenRouter and how does it save money with OpenClaw?
OpenRouter is a model gateway that provides access to 200+ AI models from multiple providers through a single API key and unified billing. OpenClaw supports it as a model provider. It's a practical cost-saving option — you can route different tasks to different models (e.g., cheap fast models for simple tasks, powerful models for complex ones) without managing separate API keys. Groq is another cost-effective option with a generous free tier and extremely fast LPU-based inference.