How to Set Up OpenClaw for Your Business (Step-by-Step)

Setting up OpenClaw takes about 10 minutes for a basic installation. Setting it up correctly for business use, with proper security, the right LLM provider, useful skills, and team access, takes more thought. This guide covers both: the quick start for testing, and the production setup for real business deployment.

Prerequisites

Before starting, you need three things. First, Node.js version 20 or later. Older versions fail silently, so verify your version with node --version before proceeding. Second, an LLM API key from at least one provider: Anthropic (Claude), OpenAI (GPT-4o), DeepSeek, or another supported model. We recommend starting with Claude for its strong reasoning capabilities. Third, a machine to run OpenClaw on: your laptop for testing, or a cloud server (Ubuntu 22.04 or later recommended) for production.

Step 1: Install OpenClaw

Open your terminal and run the official installer. On macOS or Linux, use the curl-based installer from the OpenClaw documentation. On Windows, use the MSI installer available from the GitHub releases page. The installer downloads the OpenClaw runtime, sets up the gateway

process, and creates the configuration directory at ~/.openclaw/. Verify the installation succeeded by running openclaw --version. You should see version 0.5.x or later.

Step 2: Configure Your LLM Provider

OpenClaw needs an LLM to think. Open the configuration file at ~/.openclaw/config.yaml and add your API key for your chosen provider. You can configure multiple providers and set a primary and fallback. For business use, we recommend configuring at least two providers: a primary (Claude or GPT-4o for high-quality reasoning) and a fallback (DeepSeek or an open-source model for cost-sensitive tasks or provider outages). Set cost limits in the configuration to prevent runaway API spending. The daily_limit and monthly_limit fields cap your API expenditure automatically. Start conservative: a $5 daily limit and $100 monthly limit covers most small business workloads.

Step 3: Security Hardening (Critical for Business Use)

This step is non-negotiable for any business deployment. The default OpenClaw configuration is designed for personal use and is not production-ready without hardening. First, bind the gateway to localhost only. In config.yaml, set gateway.host to 127.0.0.1. Never set it to 0.0.0.0, which exposes your agent to the entire internet. Second, generate a strong gateway token: a random string of at least 64 characters. Set this in gateway.token. Third, block port 18789 at your firewall. Even with localhost binding, belt-and-suspenders security is essential. Fourth, enable OpenShell for sandboxed execution. This isolates every agent action inside a secure container, restricting file system access to whitelisted directories and filtering network requests through policy rules. Fifth, if you need remote access, use Tailscale or a VPN tunnel instead of exposing the gateway port publicly.

Step 4: Install Business-Critical Skills

Skills turn OpenClaw from a general-purpose agent into a tool tailored to your operations. Use the clawhub CLI to browse and install skills. Start with these categories based on your needs: for email management, install the email-processor and email-sender skills. For CRM integration, install the HubSpot or Salesforce connector skills. For browser automation, install the browser skill for web research and form filling. For document processing, install the pdf-reader and document-extractor skills. For reporting, install the sheets and data-analysis skills. Important security note: vet every skill before installation. Check the skill's GitHub repository, review the code, look at the star count and recent commit activity, and verify it passed ClawHub's SHA-256 hash verification and VirusTotal scan. Never install skills from untrusted sources or with unreviewed code.

Step 5: Connect Your Communication Channels

OpenClaw supports multiple messaging channels as its user interface. For business use, we recommend Slack as the primary channel because it integrates naturally into team workflows and provides channel-based access control. Configure the Slack integration by creating an OpenClaw app in your Slack workspace and adding the bot token to your OpenClaw configuration. You can also set up WhatsApp, Discord, Telegram, or email as additional channels depending on your team's preferences.

Step 6: Define Permissions and Guardrails

Before letting your team use the agent, define what it can and cannot do. OpenClaw's permission system lets you whitelist and blacklist specific actions, directories, URLs, and API endpoints. Set file system access to only the directories the agent needs. Restrict network access to approved domains and APIs. Configure approval gates for high-impact actions such as sending emails to external contacts, making purchases, or modifying production databases. Start restrictive and expand permissions as you build confidence in the agent's behavior.

Step 7: Test with a Real Workflow

Do not automate your entire business on day one. Pick one annoying, repetitive task. Something your team spends time on every day that follows a roughly consistent pattern but has enough variation that simple rule-based automation does not work well. Good first candidates: summarizing incoming support emails and drafting responses, researching new leads and enriching CRM records, generating daily or weekly status reports from multiple data sources, and monitoring competitor pricing or product changes. Run the workflow manually alongside the agent for the first week. Compare the agent's output to what your team produces. Tune prompts, adjust permissions, and refine skill configurations based on what you observe.

Step 8: Onboard Your Team

Your team needs to understand three things about working with the OpenClaw agent. First, how to give it instructions: clear, specific goals with context produce better results than vague requests. Second, when to trust its output: for the first few weeks, review agent actions before they go external (emails, CRM updates, customer responses). Third, how to escalate: teach the team how to flag incorrect behavior and override agent decisions. This feedback loop is how the agent improves.

When to Bring in an Expert

For a broader introduction, read our OpenClaw business guide.

The setup described above works for a single-agent deployment with standard skills. If you need multi-agent orchestration, custom skill development, integration with legacy systems without APIs, compliance-ready deployment for regulated industries, or high-availability architecture with failovers, these require architectural expertise beyond a basic setup. At Cubitrek, we handle the entire deployment lifecycle so your team can focus on using the agent rather than maintaining it.

Keep exploring

• our OpenClaw business guide
• OpenClaw vs n8n vs Zapier: Which Automation Tool Wins in 2026?
• OpenClaw for Customer Support: Automating Tickets & Responses