OpenClaw Setup: 5-Minute Quickstart for Beginners

The fastest path to a working OpenClaw assistant is a single installer command, one onboarding run, and a Gateway status check. Here is what success looks like before you type anything:

• Gateway listening on port 18789
• Model provider API key configured and inference verified
• Dashboard reachable at http://localhost:18789
• First chat reply received

macOS / Linux / WSL2:

curl -fsSL https://openclaw.ai/install.sh | bash

Windows PowerShell:

iwr -useb https://openclaw.ai/install.ps1 | iex

Both commands provision Node if needed, install the CLI, and drop you straight into the onboarding wizard.

***

Key Takeaways

A working OpenClaw assistant requires one installer command, one onboarding run, and a Gateway status check — the whole process takes under five minutes on a supported system.

***

Table of Contents

• What do you need before starting the openclaw setup?
• How do the installer scripts work for each platform?
• How do you run onboarding and install the Gateway daemon?
• How do you verify the Gateway and send your first message?
• What are the alternative ways to install OpenClaw?
• Troubleshooting common OpenClaw installation errors
• How do you secure your OpenClaw configuration and Gateway?
• Self-host or managed hosting: which fits your situation?
• What should you do next after the setup is complete?
• The part most guides skip
• Skip the manual setup with Clawbase managed hosting
• Sources

What do you need before starting the openclaw setup?

Node version is the most common trip wire. The installer supports Node 22.22.3+, 24.15+, and 25.9+, with Node 26 as the recommended runtime. The installer provisions Node automatically if your system lacks a supported version, but confirming with node --version first saves time.

• OS support: macOS, Linux (native and WSL2), and Windows (native PowerShell path)
• API key: Have an OpenAI, Anthropic, or compatible model provider key ready. Onboarding asks for it interactively.
• Alpine/musl caution: Alpine Linux can expose a vulnerable system SQLite version. Use an official node:26-alpine container or a glibc-based host instead.

Pro Tip: *If you already have a workspace, re-running onboarding is safe. The wizard preserves your existing config and workspace unless you explicitly pass --reset.*

***

How do the installer scripts work for each platform?

The three installer scripts cover every major platform. Each detects your OS, provisions Node when absent, and installs OpenClaw. The default flow ends with onboarding unless you pass a skip flag.

Key flags worth knowing:

• --no-onboard skips the wizard after install
• --no-prompt / -NoOnboard runs non-interactively for CI pipelines
• OPENCLAW_INSTALL_METHOD=npm or =git forces the install method
• --json outputs machine-readable status for automation
• DryRun (PowerShell) previews actions without writing anything

For a headless CI run on Linux, the pattern looks like:

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

***

How do you run onboarding and install the Gateway daemon?

Once the CLI is installed, the Getting Started quickstart directs you to a single command:

openclaw onboard --install-daemon

This runs the full onboarding sequence and registers the Gateway as a persistent background service. Onboarding covers model and auth configuration, workspace initialization, Gateway port and bind settings, channel connections, health checks, and skill installation.

The daemon behavior differs by OS:

• macOS: Registered as a LaunchAgent under your user account
• Linux / WSL2: Installed as a systemd user unit (systemctl --user)
• Windows: Registered as a Scheduled Task; falls back to the Startup folder if Task Scheduler is unavailable

For automation, two flags matter most. --non-interactive skips all prompts and uses defaults or environment variables. --gateway-token-ref-env lets you pass a secret reference instead of a plaintext token, which is the right call for any shared or server environment. The --accept-risk flag acknowledges known configuration risks without halting the wizard.

You can also use openclaw setup --baseline to create config and workspace locations without running the full wizard. This is useful for container bootstrapping where you want a known directory structure before injecting secrets.

***

How do you verify the Gateway and send your first message?

Run these four commands in order after onboarding completes:

openclaw --version        # confirm CLI is installed correctly
openclaw gateway status   # check Gateway is listening
openclaw doctor           # surface any broken config keys
openclaw dashboard        # open the web UI

A healthy Gateway reports listening on 0.0.0.0:18789 (or your configured bind address). The dashboard loads at http://localhost:18789 by default. Send a short message in the chat interface. If the assistant replies, your openclaw installation is fully working.

If the dashboard does not load:

• Confirm the daemon is running (systemctl --user status openclaw on Linux, or check LaunchAgent on macOS)
• Check for port conflicts on 18789
• Verify firewall rules are not blocking localhost traffic
• If you use Tailscale, note that it is off by default in QuickStart mode

***

What are the alternative ways to install OpenClaw?

Not every environment suits the one-liner installer. Here are the supported paths and when each makes sense.

npm global install is the simplest manual method:

npm install -g openclaw@latest

pnpm and bun work as drop-in replacements. This approach suits developers who already manage their own Node environment and want explicit control over the install location.

Git / source checkout is appropriate for contributors or anyone who wants to run unreleased code:

git clone https://github.com/openclaw/openclaw.git
cd openclaw
pnpm install && pnpm build && pnpm link

For live development, the advanced setup docs describe pnpm gateway:watch, which runs the Gateway in a hot-reload loop. Attach a local OpenClaw app in Local mode to test changes without daemonizing.

install-cli.sh installs to a local prefix at ~/.openclaw without touching system paths. This is the right choice for Docker containers, Nix environments, and cloud CI runners where you want no global side effects.

Alpine/musl warning: The installer docs flag that Alpine Linux can expose a vulnerable system SQLite version. Use the official node:26-alpine container image or switch to a glibc-based host to avoid this.

***

Troubleshooting common OpenClaw installation errors

Most failures fall into three categories. Here is how to resolve each quickly. For a broader list of open-source AI setup errors, the Clawbase fix guide covers additional edge cases.

openclaw not found after install:

• Run npm prefix -g to find your global bin directory
• Add $(npm prefix -g)/bin to your PATH in ~/.bashrc or ~/.zshrc
• Reopen your terminal and retry

Node version mismatch:

• Run node --version and confirm it is 22.22.3+, 24.15+, 25.9+, or 26+
• The installer provisions Node automatically, but a conflicting system Node can shadow it
• Use a version manager like nvm or fnm to switch cleanly

Git missing for source installs:

• On Windows, the installer includes a MinGit bootstrap; if it fails, install Git for Windows manually and retry
• On Linux, apt install git or the equivalent resolves this immediately

Pro Tip: *Run openclaw doctor before re-running onboarding. It surfaces legacy or corrupted config keys that would cause the wizard to fail silently. On PowerShell, add -Verbose or enable tracing to see exactly where a script stalls.*

***

How do you secure your OpenClaw configuration and Gateway?

Security defaults in OpenClaw are reasonable, but a few deliberate choices make a production deployment meaningfully safer.

• Never expose the Gateway to the open internet without TLS and auth. The QuickStart binds to localhost by default. If you need remote access, put a reverse proxy (nginx, Caddy) with HTTPS in front.
• Store secrets as environment-backed refs. Onboarding supports SecretRef and --gateway-token-ref-env so plaintext API keys never land in your config file.
• Keep personal config in ~/.openclaw/openclaw.json and workspace files in ~/.openclaw/workspace. The setup docs explicitly recommend this to prevent repo updates from overwriting your customizations.
• Run the daemon under your user account, not root. The LaunchAgent and systemd user unit defaults do this correctly. Avoid running with elevated privileges.
• Set firewall rules to restrict Gateway port access to localhost or a known VPN range.
• Monitor with openclaw doctor and Gateway logs regularly. Automated updates and daily encrypted backups (available on managed plans) remove most of this burden for non-self-hosters.

***

Self-host or managed hosting: which fits your situation?

Self-hosting gives you full control over configuration, model routing, and data residency. That control comes with real operational overhead: you manage Node upgrades, daemon restarts, API key rotation, and backup schedules yourself.

Managed hosting is worth considering when:

• You have limited sysadmin time or no dedicated server
• You need 99.9% uptime without writing your own monitoring
• Automated encrypted backups and updates matter for your workflow
• You want multi-model routing across 50+ models without manual configuration
• Channel integrations (Telegram, Discord, Slack, WhatsApp) need to stay connected without babysitting

Clawbase offers one-click OpenClaw deployment on dedicated encrypted servers. You get persistent memory management, daily encrypted backups, automated updates, and browser relay integration out of the box. For teams and busy professionals who want the assistant's capabilities without the infrastructure work, that tradeoff is clear. For developers who want to modify the source or run bleeding-edge builds, self-hosting remains the right call.

***

What should you do next after the setup is complete?

A working Gateway is the starting line, not the finish. Here is where to go next:

• Connect a channel. Telegram is the fastest integration to configure. Run openclaw dashboard, navigate to Channels, and follow the bot token flow.
• Confirm your model provider. Verify your API key is active and the model you selected during onboarding is responding correctly.
• Install recommended skills. The Skills marketplace adds capabilities like web search, file management, and calendar access. Browse from the dashboard.
• Set up monitoring. Schedule a weekly openclaw doctor check and review Gateway logs monthly. If you self-host, configure log rotation.
• Read the docs. The most useful next reads are the Channels reference, the Models configuration page, the Gateway options doc, and the Plugins/Skills guide.

For a deeper look at what the assistant can actually do in daily workflows, the OpenClaw use cases page is a practical starting point.

***

The part most guides skip

The openclaw setup itself takes five minutes. What takes longer is deciding how much of the operational layer you actually want to own.

Most beginners underestimate the ongoing work: keeping the daemon alive after OS updates, rotating API keys before they expire, and debugging Gateway connectivity when a firewall rule changes. None of it is hard, but all of it is time. The openclaw doctor command catches most issues early, and keeping your config in ~/.openclaw/openclaw.json prevents the most common data-loss scenario (a repo pull that overwrites your workspace). Those two habits alone prevent the majority of support questions.

The honest advice: self-host if you want to learn the internals or need custom model routing. Choose managed hosting if the assistant's output is the point, not the infrastructure.

***

Skip the manual setup with Clawbase managed hosting

If the steps above look like more infrastructure than you signed up for, Clawbase is the direct alternative. You get a fully configured OpenClaw assistant on a dedicated encrypted server, deployed in one click, with no terminal required.

• 99.9% uptime with daily encrypted backups and automated updates
• 50+ AI models with multi-model routing, plus Telegram, Discord, Slack, and WhatsApp integrations pre-connected

Plans start with affordable monthly rates and include a free trial period. See what the assistant can do for your workflow on the Clawbase use cases page, or go straight to Clawbase to start your trial.

***

Sources

• Install
• installer.md (OpenClaw docs on GitHub)
• OpenClaw — Personal AI Assistant

Recommended

• What Is OpenClaw and How ClawBase Runs It for You | ClawBase
• Blog — OpenClaw Hosting Tips, Tutorials & Updates | ClawBase