Does OpenClaw require a GPU?

No — not for cloud API mode. OpenClaw is a Node.js orchestration gateway. When you use Claude, GPT-4, or Gemini via API, the model runs on the provider's hardware. Your machine only routes messages. A GPU is only needed if you run local models via Ollama. For local models, an NVIDIA GPU with 8–24 GB VRAM accelerates inference significantly over CPU-only, but CPU+RAM works at slower speeds.

Is a Mac Mini M4 with 16 GB enough for OpenClaw?

Yes — if you use cloud APIs. The gateway runs in under 200 MB RAM, so the 16 GB Mac Mini handles multi-channel operation, automations, and multiple agents without strain. Upgrade to 24 GB only if you also want to run local models via Ollama. At 24 GB you can run 13–14B models comfortably. Memory is soldered and cannot be upgraded later, so buy for your intended use case.

How much RAM do I need to run Ollama with OpenClaw?

RAM for Ollama depends on model size, not on OpenClaw itself. OpenClaw uses ~200 MB RAM; the rest goes to the model. Rule of thumb: 8 GB total = 7B models (Llama 3.1 8B, Mistral 7B); 16 GB = 13–14B models comfortably; 24 GB unified memory (Mac) = 32B models at reduced speed; 32 GB+ = 32B comfortably or 70B slowly. Browse model sizes at ollama.com/library. On Apple Silicon, unified memory means the full RAM pool is available for Ollama — no split between system and GPU.

Can I run OpenClaw on Windows without WSL2?

Yes — native Windows is supported without WSL2, using Node.js and PowerShell. All core features work: the gateway, CLI, and all channels except iMessage. WSL2 is still strongly recommended for better daemon management and full Linux compatibility. If WSL2 is unavailable (corporate restrictions or preference), native Windows works. iMessage requires macOS regardless.

What is the minimum VPS spec to run OpenClaw 24/7?

Minimum for a single-channel cloud setup: 1 vCPU, 2 GB RAM, 20 GB SSD, Ubuntu 22.04 or Debian 12. Important: 512 MB plans (Render Starter, fly.io shared) will OOM-kill the process during startup — 2 GB RAM is the true minimum. Recommended for comfortable multi-channel 24/7 operation: 2 vCPU, 4 GB RAM, 40 GB SSD. Providers with documented setups at $5–7/month: Hetzner, DigitalOcean, Vultr.

Can I run OpenClaw on an old laptop?

Yes, for cloud-model use. Any laptop with 4 GB+ RAM running macOS, Linux, or Windows can run the OpenClaw gateway. The key limitation: automations and scheduled briefings only fire when the machine is awake. A laptop that sleeps will miss cron jobs. For 24/7 automations, use a dedicated always-on device — see system requirements tiers.

Should I get a Raspberry Pi 4 or Pi 5 for OpenClaw?

Pi 5 is strongly recommended. Both run OpenClaw for cloud-model use, but Pi 5 is 2–3x faster on CPU tasks and handles complex multi-step automations without lag. Pi 4 (4 GB or 8 GB) works but feels slow on busier setups. Pi 3 and earlier are not officially supported. Local model inference is impractical on any Pi — use cloud APIs.

How much does it cost in electricity to run OpenClaw 24/7?

Very little. Raspberry Pi 5 draws 3–7 W idle (~$3–5/year); Mac Mini M4 draws 10–15 W idle (~$10–15/year); a VPS costs $5–10/month with zero home electricity overhead. The bigger cost for cloud-API use is the AI provider API itself — typically $1–10/month. For local models, the machine draws 25–150 W under inference load, but only while actively generating a response.

Does OpenClaw work on ARM Linux or ARM servers?

Yes. OpenClaw publishes multi-architecture Docker images including native ARM64 (aarch64). Raspberry Pi 5 running Raspberry Pi OS 64-bit or Ubuntu 24.04, NVIDIA Jetson Orin Nano, and ARM cloud VMs (AWS Graviton, Oracle Ampere) all run OpenClaw without emulation. `docker pull openclaw/openclaw:latest` fetches the correct ARM64 image automatically.

Does OpenClaw work on Windows?

Yes. OpenClaw runs on Windows with Node.js 22.16 or later. WSL2 is strongly recommended for the best experience. The iMessage channel is the only exception — it requires macOS. All other channels work on Windows.

Does OpenClaw work on Raspberry Pi?

Yes, for cloud-model setups. Raspberry Pi 4 or 5 with 2 GB+ RAM (4 GB recommended) and a 64-bit OS. Since models run in the cloud via API, even a modest Pi handles the gateway workload well. Not suitable for local models.

What Node.js version does OpenClaw require?

Node.js 22.16 or later. Node.js 24 is recommended. Older versions (18, 20) are not supported. Check your version with `node --version` and upgrade via nvm if needed.

Can I run OpenClaw in Docker?

Yes. OpenClaw runs in Docker with Docker Engine or Docker Desktop + Compose v2. Your host needs at least 2 GB RAM — 1 GB hosts may OOM-kill the build. The iMessage channel requires macOS and won't work inside a Linux container. See the OpenClaw Docker guide.

What do I need for multi-agent orchestration with local models?

A single 24 GB machine is the absolute floor — the official docs describe it as suitable only for lighter prompts with higher latency. For serious multi-agent setups running 24/7, the docs reference two or more maxed-out Mac Studios or an equivalent GPU rig as the recommended ceiling for full local stacks.

What Mac should I use for 24/7 multi-agent OpenClaw?

A Mac Mini with 24 GB RAM and 512 GB SSD is a solid starting point for multi-agent orchestration with cloud models. If you also want to run local models (Ollama) alongside agents, 24 GB is the minimum — expect higher latency on larger models. For heavy local model workloads, a Mac Studio with 64 GB+ unified memory is recommended.

OpenClaw Server Requirements (Node.js, ARM64, Docker)

Exact OpenClaw requirements per use case — cloud APIs vs. local Ollama. Covers Node.js, RAM, macOS, Linux, Windows, ARM64 servers, Raspberry Pi, Docker.

Requirements depend heavily on what you want to do. A basic personal assistant and a 24/7 multi-agent orchestration setup with local models are very different workloads. Use the tiers below to find your target.

Tier 1 — Basic personal assistant (cloud models)

This is the default. Your AI provider (Anthropic, OpenAI, Google, etc.) does the heavy lifting. OpenClaw is just the gateway — it routes messages and runs automations.

Component	Minimum	Notes
Node.js	22.16	Node 24 recommended
RAM	2 GB	4 GB+ for comfortable headroom
Disk	No official minimum stated	Budget 4 GB+ for install + logs; Raspberry Pi docs recommend 16 GB+ storage
OS	macOS, Linux (64-bit), Windows (WSL2)

Works on: Raspberry Pi 4/5, any cheap VPS ($5/month), a spare laptop, Mac Mini.

When to use this tier: you want a personal assistant, you're connecting messaging apps, you want scheduled automations. You don't need a GPU. You're paying API costs per request.

→ OpenClaw install overview · Supported AI providers

Tier 2 — Always-on, 24/7, multi-channel (cloud models)

Adding more channels, more automations, and always-on operation doesn't change the model compute requirements — that still runs in the cloud. What changes is reliability and uptime.

Component	Recommended
RAM	4–8 GB
Disk	20 GB+ (logs, session files, media accumulate over time)
Machine	Dedicated always-on device — not a laptop that sleeps
Network	Stable connection; no sleep/hibernate

Good choices: Mac Mini, Raspberry Pi 5 (4 GB), a VPS with SSD storage.

Important: Cron jobs, scheduled briefings, and automations only fire when the gateway is running. A laptop in sleep mode won't trigger them. A Mac Mini set to "prevent sleep" or a VPS is the right approach for this tier.

→ VPS deployment guide · Raspberry Pi setup · Mac Mini specs

Tier 3 — Local models via Ollama (no cloud API)

Running models locally means your machine does all the inference. Requirements scale with the model you want to run.

Model size	RAM / Unified Memory needed	Examples
7–8B	8 GB	Mistral 7B, Llama 3.1 8B
13–14B	16 GB	Llama 2 13B, Qwen 14B
32B	32 GB+	Qwen2.5-Coder 32B, DeepSeek-R1 32B
70B+	64 GB+	Llama 3.3 70B

The official OpenClaw docs note: "A single 24 GB GPU works only for lighter prompts with higher latency."

This means 24 GB unified memory (Mac) or VRAM (GPU) is the floor for meaningful local model use — not the sweet spot. At 24 GB you can run 13–14B models well, or 32B models slowly.

Disk: Local model weights are large. A 32B model is ~20 GB on disk. A 70B model is ~40 GB+. Budget 512 GB SSD minimum if you plan to run multiple models. You can browse available models and their sizes at ollama.com/library.

→ OpenClaw + Ollama setup · OpenClaw local models guide · Ollama model library

Tier 4 — Multi-agent orchestration with local models, 24/7

This is the most demanding use case: multiple agents running in parallel, each making inference calls, with memory consolidation, automations, and always-on operation.

The official docs are direct: "≥2 maxed-out Mac Studios or equivalent GPU rig (~$30k+)" is the recommended hardware for a full local multi-agent stack.

For most users, the practical entry point is:

Component	Minimum	Recommended
RAM / Unified Memory	24 GB	64 GB+
Disk	512 GB SSD (more if running large local models)	1 TB+ NVMe SSD
CPU	Apple Silicon M-series or modern x86	M4 Pro / M4 Max or better
Machine	Mac Mini (24 GB)	Mac Studio (64 GB+)
Uptime	Dedicated always-on machine	Never sleeps, UPS recommended

Mac Mini 24 GB + 512 GB SSD is the realistic entry point for running multi-agent orchestration 24/7 with cloud models, plus moderate local model use (up to ~14B comfortably, 32B with patience). It's a solid machine for this workload at a reasonable price.

If you want to run large local models (32B+) as the backbone of your agents — not just occasionally — a Mac Studio with 64 GB+ unified memory is the right tool.

On Apple Silicon: Unified memory means RAM and VRAM are the same pool. A 24 GB Mac Mini gives you the full 24 GB for model inference, not split between system and GPU like a PC with a discrete card. This makes Apple Silicon unusually efficient for local model workloads at this price point. Apple's Mac Mini and Mac Studio spec pages show exact memory options and configurations.

→ OpenClaw multi-agent concepts · Mac Mini specs · Mac Studio specs

Node.js

OpenClaw requires Node.js 22.16 or later. Node.js 24 is recommended.

Check your version:

node --version

Install or upgrade via nvm:

nvm install 24
nvm use 24

Or download directly from nodejs.org.

→ OpenClaw Node.js install guide · Node.js release schedule · nvm on GitHub

Supported operating systems

macOS

Any modern macOS version supported for the gateway
macOS 15+ required for the optional companion menu-bar app
All channels supported including iMessage
ARM (Apple Silicon M1/M2/M3/M4) and Intel supported

→ macOS release history

Linux

Any modern 64-bit distribution with Node.js 22.16+
Debian Bookworm or equivalent recommended
All channels supported except iMessage

→ Ubuntu download · Debian download

Windows

WSL2 strongly recommended — more stable than native Windows
Native Windows CLI is supported but still improving
All channels supported except iMessage
iMessage requires macOS on the machine running the gateway

→ Install WSL2 — Microsoft docs · OpenClaw Windows guide

Raspberry Pi

Raspberry Pi 4 or 5 with 2 GB+ RAM (4 GB recommended)
64-bit OS required — Raspberry Pi OS Bookworm 64-bit or Ubuntu 24.04
Pi 3 and earlier are not officially supported
Suitable for Tier 1–2 (cloud models only) — not for local model inference

→ OpenClaw Raspberry Pi guide · Raspberry Pi OS download · Pi 4 vs Pi 5 specs

Docker

OpenClaw runs in Docker with Docker Engine or Docker Desktop + Compose v2.

docker pull openclaw/openclaw:latest
docker run -d \
  -v ~/.openclaw:/root/.openclaw \
  -p 3000:3000 \
  openclaw/openclaw:latest

Minimum RAM for Docker: 2 GB — 1 GB hosts may OOM-kill the build (exit 137).

Disk: Watch media/, session JSONL files, and cron/runs/*.jsonl — these grow over time. Use SSD-backed storage.

iMessage in Docker: Not supported. iMessage uses the macOS Messages framework, which isn't available in Linux containers.

→ OpenClaw Docker guide · Docker Engine install · Docker Compose v2

Always-on deployment

For automations and scheduled briefings to fire reliably, the Gateway must be running continuously:

Machine must be powered on and not sleeping
Stable network connection required for API calls
Automations will not fire if the machine hibernates

See Module 6: Autonomous Tasks for deployment options including VPS and Mac Mini setups.

→ VPS deployment guide · Tailscale for remote access

Checking your setup

Run the built-in diagnostics after installation:

openclaw doctor

This checks Node.js version, config validity, AI provider connectivity, and channel status.

→ OpenClaw install guide · Troubleshooting

FAQ

Does OpenClaw require a GPU?: No — not for cloud API mode. OpenClaw is a Node.js orchestration gateway. When you use Claude, GPT-4, or Gemini via API, the model runs on the provider's hardware. Your machine only routes messages. A GPU is only needed if you run local models via Ollama. For local models, an NVIDIA GPU with 8–24 GB VRAM accelerates inference significantly over CPU-only, but CPU+RAM works at slower speeds.
Is a Mac Mini M4 with 16 GB enough for OpenClaw?: Yes — if you use cloud APIs. The gateway runs in under 200 MB RAM, so the 16 GB Mac Mini handles multi-channel operation, automations, and multiple agents without strain. Upgrade to 24 GB only if you also want to run local models via Ollama. At 24 GB you can run 13–14B models comfortably. Memory is soldered and cannot be upgraded later, so buy for your intended use case.
How much RAM do I need to run Ollama with OpenClaw?: RAM for Ollama depends on model size, not on OpenClaw itself. OpenClaw uses ~200 MB RAM; the rest goes to the model. Rule of thumb: 8 GB total = 7B models (Llama 3.1 8B, Mistral 7B); 16 GB = 13–14B models comfortably; 24 GB unified memory (Mac) = 32B models at reduced speed; 32 GB+ = 32B comfortably or 70B slowly. Browse model sizes at ollama.com/library. On Apple Silicon, unified memory means the full RAM pool is available for Ollama — no split between system and GPU.
Can I run OpenClaw on Windows without WSL2?: Yes — native Windows is supported without WSL2, using Node.js and PowerShell. All core features work: the gateway, CLI, and all channels except iMessage. WSL2 is still strongly recommended for better daemon management and full Linux compatibility. If WSL2 is unavailable (corporate restrictions or preference), native Windows works. iMessage requires macOS regardless.
What is the minimum VPS spec to run OpenClaw 24/7?: Minimum for a single-channel cloud setup: 1 vCPU, 2 GB RAM, 20 GB SSD, Ubuntu 22.04 or Debian 12. Important: 512 MB plans (Render Starter, fly.io shared) will OOM-kill the process during startup — 2 GB RAM is the true minimum. Recommended for comfortable multi-channel 24/7 operation: 2 vCPU, 4 GB RAM, 40 GB SSD. Providers with documented setups at $5–7/month: Hetzner, DigitalOcean, Vultr.
Can I run OpenClaw on an old laptop?: Yes, for cloud-model use. Any laptop with 4 GB+ RAM running macOS, Linux, or Windows can run the OpenClaw gateway. The key limitation: automations and scheduled briefings only fire when the machine is awake. A laptop that sleeps will miss cron jobs. For 24/7 automations, use a dedicated always-on device — see system requirements tiers.
Should I get a Raspberry Pi 4 or Pi 5 for OpenClaw?: Pi 5 is strongly recommended. Both run OpenClaw for cloud-model use, but Pi 5 is 2–3x faster on CPU tasks and handles complex multi-step automations without lag. Pi 4 (4 GB or 8 GB) works but feels slow on busier setups. Pi 3 and earlier are not officially supported. Local model inference is impractical on any Pi — use cloud APIs.
How much does it cost in electricity to run OpenClaw 24/7?: Very little. Raspberry Pi 5 draws 3–7 W idle (~$3–5/year); Mac Mini M4 draws 10–15 W idle (~$10–15/year); a VPS costs $5–10/month with zero home electricity overhead. The bigger cost for cloud-API use is the AI provider API itself — typically $1–10/month. For local models, the machine draws 25–150 W under inference load, but only while actively generating a response.
Does OpenClaw work on ARM Linux or ARM servers?: Yes. OpenClaw publishes multi-architecture Docker images including native ARM64 (aarch64). Raspberry Pi 5 running Raspberry Pi OS 64-bit or Ubuntu 24.04, NVIDIA Jetson Orin Nano, and ARM cloud VMs (AWS Graviton, Oracle Ampere) all run OpenClaw without emulation. `docker pull openclaw/openclaw:latest` fetches the correct ARM64 image automatically.
Does OpenClaw work on Windows?: Yes. OpenClaw runs on Windows with Node.js 22.16 or later. WSL2 is strongly recommended for the best experience. The iMessage channel is the only exception — it requires macOS. All other channels work on Windows.
Does OpenClaw work on Raspberry Pi?: Yes, for cloud-model setups. Raspberry Pi 4 or 5 with 2 GB+ RAM (4 GB recommended) and a 64-bit OS. Since models run in the cloud via API, even a modest Pi handles the gateway workload well. Not suitable for local models.
What Node.js version does OpenClaw require?: Node.js 22.16 or later. Node.js 24 is recommended. Older versions (18, 20) are not supported. Check your version with `node --version` and upgrade via nvm if needed.
Can I run OpenClaw in Docker?: Yes. OpenClaw runs in Docker with Docker Engine or Docker Desktop + Compose v2. Your host needs at least 2 GB RAM — 1 GB hosts may OOM-kill the build. The iMessage channel requires macOS and won't work inside a Linux container. See the OpenClaw Docker guide.
What do I need for multi-agent orchestration with local models?: A single 24 GB machine is the absolute floor — the official docs describe it as suitable only for lighter prompts with higher latency. For serious multi-agent setups running 24/7, the docs reference two or more maxed-out Mac Studios or an equivalent GPU rig as the recommended ceiling for full local stacks.
What Mac should I use for 24/7 multi-agent OpenClaw?: A Mac Mini with 24 GB RAM and 512 GB SSD is a solid starting point for multi-agent orchestration with cloud models. If you also want to run local models (Ollama) alongside agents, 24 GB is the minimum — expect higher latency on larger models. For heavy local model workloads, a Mac Studio with 64 GB+ unified memory is recommended.