How to Install OpenClaw (Mac, Windows, WSL2): Step-by-Step Guide

To install OpenClaw locally, you need Node.js (v22.16+). For Mac and Linux, run the command curl -fsSL https://openclaw.ai/install.sh | bash. For Windows, run iwr -useb https://openclaw.ai/install.ps1 | iex in PowerShell.

Building your own AI assistant is no longer just for enterprise tech teams. Self-hosted AI agents like OpenClaw give developers incredible local control over their workflows. But before we dive into terminal commands and local deployment guides, it is crucial to determine if a local CLI deployment is the right fit for your business goals.

For teams that want the outcome of an AI assistant without spending days on WSL2, dependencies, API limits, Docker, and local security hardening, Buda offers a faster zero-setup path with a free trial so you can test real assistant workflows before committing to a self-hosted OpenClaw stack. Best of all, Buda currently offers a free trial, allowing you to experience a fully automated workflow today with zero upfront risk.

How to Download OpenClaw Safely

The safest way to download OpenClaw is from its official repository, installer, or documentation-linked source. Avoid random ZIP files, reposted shell scripts, unofficial mirrors, or “one-click” installers that are not linked from the project itself.

A typical download pattern looks like this:

git clone <official-openclaw-repository-url>
cd openclaw

Then install dependencies according to the current documentation:

pnpm install

or, if the current version uses npm:

npm install

This step matters because OpenClaw is not a simple desktop app. It is closer to an agent runtime. A working installation depends on Node.js, package dependencies, model APIs, browser automation, messaging integrations, and local permissions.

In my research, some of the most frustrating failures happened after the project was already downloaded. On Windows, for example, onboarding could fail during the messaging-platform step because required Node packages were missing. The practical lesson: downloaded does not mean installed. Always verify dependency installation before starting the onboarding flow.

A good pre-install checklist is:

Official source confirmed
Node.js version checked
pnpm or npm installed
Git installed
Docker installed only if needed
API provider selected
One messaging channel selected

For most beginners, the first setup should be simple: OpenClaw plus one model provider plus Telegram.

How to install OpenClaw on Mac

Mac users, particularly those on M-series chips, enjoy a highly optimized OpenClaw experience. Follow these steps to install OpenClaw on Mac:

1. Verify Prerequisites: Ensure you have Node.js 24 (or at least 22.16+) installed. The script will attempt to handle this if you don’t.
2. Open Terminal: Use macOS Terminal or iTerm2.
3. Run the Script: Paste the following command and hit Enter: curl -fsSL https://openclaw.ai/install.sh | bash

Install but don’t run the New User Guide: curl -fsSL https://openclaw.ai/install.sh | bash -s — –no-onboard

4. Follow the Onboarding: When prompted, type “Yes” to acknowledge the security risks of powerful AI agents, and select the QuickStart profile for safe default settings.

5. Enable Auto-Completion: Opt into bash/zsh shell completion when asked; it will make your daily usage significantly faster.

Practical Mac installation experience

Mac avoids many Windows-specific problems: PowerShell execution policy, Windows Defender blocking scripts, path confusion, WSL filesystem boundaries, and native Windows browser automation issues.

However, Mac setup is not risk-free. The most common Mac issues are package-manager path problems, permission prompts, browser automation permissions, and confusion about whether the AI work is happening locally or through cloud APIs.

My recommended first Mac stack is:

Mac
+ Node.js v22
+ pnpm
+ OpenClaw
+ Telegram
+ one API model

Do not start with WhatsApp, Gmail, file access, browser control, and local models all at once. Start with one Telegram message loop. If that works, add tools one by one.

How to install OpenClaw on Windows

The best way to install OpenClaw on Windows is usually through WSL2. Native Windows installation can work, but it is more likely to break because of PowerShell policies, Defender warnings, path issues, missing dependencies, and automation limitations.

To install OpenClaw natively on Windows:

1. Launch PowerShell: Open Windows PowerShell as an Administrator.
2. Execute the Command: Copy and paste the following web request script: iwr -useb https://openclaw.ai/install.ps1 | iex

Install but don’t run the New User Guide:

WSL2(Bath)

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard

Windows (PowerShell)

& ([scriptblock]::Create((iwr -useb https://openclaw.ai/ins

3. Confirm Prompts: Use your arrow keys to select “Yes” for the security warning and choose the “QuickStart” option.

Case study: $200 CAD Windows mini PC sandbox

Another practical Windows setup used a dedicated mini PC as an isolated OpenClaw sandbox:

Cost: $200 CAD
CPU: Core i5
RAM: 16GB
Storage: 256GB SSD
OS: Windows 11

The goal was not maximum performance. It was safety. Running OpenClaw on a separate machine reduced the risk of giving an experimental agent access to a main laptop.

The setup still required fixing PowerShell execution policy issues, Windows Defender blocks, Node/NPM/Git installation, winget setup, and remote access through tools such as Tailscale. But the sandbox approach made the risk easier to control.

For beginners, a separate test machine can be safer than installing OpenClaw directly on a primary work computer.

Verify the install

Bash：

openclaw --version      # confirm the CLI is available
openclaw doctor         # check for config issues
openclaw gateway status  # verify the Gateway is running

If you want managed startup after install:

• macOS: LaunchAgent via openclaw onboard --install-daemon or openclaw gateway install
• Linux/WSL2: systemd user service via the same commands
• Native Windows: Scheduled Task first, with a per-user Startup-folder login item fallback if task creation is denied

How to Set Up OpenClaw After Installation

After installation, set up OpenClaw in this order:

Model provider
Messaging channel
Execution environment
Safety controls

This sequence avoids most beginner mistakes.

Choose one model provider

Common model options include OpenAI, Anthropic Claude, MiniMax, Ollama, Qwen, and GLM-style local models.

For beginners, a cloud API is usually easier. Local models are better for privacy and control, but they require more hardware, memory, tuning, and patience.

One low-cost setup used MiniMax Coding Starter at $10/month, with 100 prompts per 5 hours. In practice, the prompt window was observed to reset around 4 hours. That workflow was tested about 50 times, which makes the result more useful than a one-off setup report.

The setup was:

Windows
WSL
MiniMax
Telegram
BotFather
OpenClaw

It was not the strongest possible model stack, but it showed that OpenClaw can be tested cheaply without buying a Mac Mini or high-end GPU.

Start with Telegram

Telegram is usually the best first communication channel. It has a clear bot setup flow through BotFather and does not require the same kind of phone-session persistence that WhatsApp often does.

A reliable first setup looks like:

OpenClaw + Telegram bot + one model provider

WhatsApp may be useful later, but it can be harder to maintain. Discord can work for communities or teams, but bot permissions and scheduled tasks may need more debugging.

Choose local, VPS, Docker, or sandbox

There are four common OpenClaw setup paths:

Local machine
VPS
Docker/container
Separate sandbox device

Each is useful for a different goal.

A local machine is best when you need local files, local browser sessions, or local development tools. The downside is security risk.

A VPS is best when you want an always-on assistant for reminders, summaries, lightweight web automation, and Telegram-based workflows. A practical beginner VPS setup can be:

1–2 vCPU
2–4GB RAM
$5–8/month VPS
$3–8/month light API usage
Under $15/month total

For many beginners, the correct first move is not more hardware. It is a smaller, safer workflow. Docker is helpful for isolation and repeatability, but it is not automatically easier. It can add problems around ports, volumes, browser shared memory, token persistence, and networking.

Security Lockdown

Always bind your openclaw.json gateway to 127.0.0.1 instead of 0.0.0.0, and enable the exec_approval flag to ensure the AI cannot execute risky terminal commands without your manual permission.

OpenClaw Safety Setup: Do This Before Real Tasks

OpenClaw can become risky because it may connect to messaging apps, browsers, files, calendars, code repositories, and cloud accounts. Set safety controls before giving it sensitive work.

A safer OpenClaw setup includes:

API spending limits
HTTPS for remote access
Authentication
Firewall rules
Separate browser profile
Separate bot accounts
Restricted file access
No public exposure of internal ports
Docker, VM, WSL2, or separate hardware isolation

API spending limits are especially important. In one complex coding case, a dashboard-building workflow cost $160 because the agent repeatedly wrote code, hit errors, rewrote code, and kept context alive. In other runaway automation scenarios, costs reached $300–600 when retry loops continued overnight.

That is why the first production rule should be: never run OpenClaw unattended with unlimited API access.

Before connecting Gmail, WhatsApp, GitHub, calendar tools, or local files, test with a limited account. Use a separate browser profile and avoid granting broad file-system access during early setup.

How to Get Started With OpenClaw

The best way to get started with OpenClaw is to pick one narrow workflow and make it reliable.

Good first tasks include:

Summarize a web page
Reply through Telegram
Draft a daily briefing
Search and report back
Create a simple task list
Run a small browser task in a separate profile

Practical beginner workflow

A good first OpenClaw workflow is:

Send Telegram message
→ OpenClaw receives it
→ Model answers
→ OpenClaw logs the result
→ You check API usage
→ You repeat with a slightly harder task

This sounds basic, but it prevents most setup confusion. When something breaks, you know which layer to inspect: Telegram, OpenClaw, model API, browser tool, or local permission.

Local model experience

For privacy-focused setups, OpenClaw can be paired with Docker and Ollama. One advanced setup used:

Docker
Ollama
GLM model
64k context
4-bit quantization
RTX 4070 Super

Another local route used:

Qwen 9B
RTX 4080

The benefit is more local control. The trade-off is slower performance, higher hardware requirements, more setup complexity, and sometimes weaker tool use than cloud models.

The practical answer to “Can OpenClaw run 100% locally?” is: partly, for some workflows. But beginners should usually start with an API model first, then move local once they understand the workload.

FAQ: OpenClaw Installation and Setup

How do I install OpenClaw on Windows?

Use WSL2. Install Ubuntu through WSL2, install Node.js, pnpm, and Git inside Ubuntu, download OpenClaw inside the Linux filesystem, then run the setup or onboarding command.

How do I install OpenClaw on Mac?

Install Git, Node.js, and pnpm with Homebrew, download OpenClaw from the official source, install dependencies, and run the onboarding command. Start with Telegram and one model provider.

How do I download OpenClaw?

Download OpenClaw only from the official repository, installer, or documentation-linked source. Avoid unofficial mirrors and copied install scripts.

How do I set up OpenClaw after installation?

Connect one model provider, one messaging channel, one execution environment, and safety controls. Start with Telegram and a cloud API before adding more tools.

If your goal is to validate AI assistant workflows quickly rather than spend time managing dependencies, WSL2, Docker, API limits, and local security settings, Buda offers a zero-setup cloud alternative with a free trial, making it easier to test real assistant use cases before committing to a self-hosted OpenClaw setup.

How do I get started with OpenClaw?

Start with one small workflow, such as sending a Telegram message and receiving a model response. Then test summaries, reminders, or simple browser tasks.

Can I install OpenClaw on Windows 10 or Windows 11 without using the terminal?

Not reliably. OpenClaw usually requires terminal commands, dependency installation, API keys, and configuration. WSL2 makes the process more predictable.

Should I use WSL2 or native Windows?

Use WSL2 for your first Windows installation. Native Windows is more likely to hit PowerShell, Defender, path, dependency, and automation issues.

Do I need a Mac Mini to run OpenClaw?

No. A Mac Mini is optional. A small VPS, existing computer, WSL2 setup, or sandbox mini PC can be enough depending on the use case.

Do I need Docker for OpenClaw?

Not always. Docker is useful for isolation, but it can add setup complexity. Beginners should use Docker only when needed or when isolation is the priority.

How much does OpenClaw cost to run?

A low-cost model setup can start around $10/month. A VPS plus light API use can stay under $15/month. Complex automation can cost far more, with real cases reaching $160 or even $300–600 during runaway retry loops.

Can OpenClaw control Excel, PowerPoint, or Windows apps?

Not automatically. Installing OpenClaw does not guarantee desktop app control. Windows software automation needs additional permissions, tools, and workflow design.

Is Telegram better than WhatsApp for OpenClaw?

For a first setup, yes. Telegram is usually easier because bot creation is straightforward and session management is simpler. WhatsApp may require more maintenance.

Is OpenClaw safe to connect to Gmail, GitHub, or local files?

Only after safety controls are in place. Use restricted accounts, separate profiles, API limits, authentication, firewall rules, and isolation through WSL2, Docker, a VM, or separate hardware.

Should I run OpenClaw locally or on a VPS?

Use local installation when you need local files or local browser access. Use a VPS when you want an always-on assistant for lightweight tasks such as reminders, summaries, and Telegram workflows.

If your goal is to validate AI assistant workflows quickly rather than spend time managing dependencies, WSL2, Docker, API limits, and local security settings, Buda offers a zero-setup cloud alternative with a free trial, making it easier to test real assistant use cases before committing to a self-hosted OpenClaw setup.