Key Takeaways

• OpenClaw is an open-source AI agent that executes tasks across your calendar, email, files, and messaging apps — not just answers questions.
• It runs on a personal device or a VPS (virtual private server), with VPS hosting recommended for 24/7 availability.
• Channel integration (WhatsApp, Telegram, Discord, Slack) lets you interact with one agent across multiple platforms without switching apps.
• A basic VPS setup costs roughly $5–10/month and can replace multiple SaaS subscriptions (notes, to-do lists, habit trackers).
• Security is critical: bind the gateway to localhost, use SSH tunneling, rotate tokens, run as a non-root user, and never expose the gateway to the open internet.
• OpenClaw supports multiple AI providers (Anthropic Claude, OpenAI, Google Gemini) and lets you assign different models to different tasks.
• Skills (plugins) extend functionality — web browsing, calendar management, image generation, code review, and more.

Software development – illustrative photo. Image credit: ThisisEngineering via Unsplash, free license

What Is OpenClaw and Why Does It Matter?

OpenClaw — formerly known as Clawdbot and later Moltbot — is an open-source AI agent designed to carry out tasks on your behalf. Unlike standard chatbots that generate text responses and stop there, OpenClaw connects to your tools and acts on them. It can send emails, update calendars, run shell commands, manage files, post to Slack, and track deliveries without requiring you to switch between apps or perform each step manually.

The core advantage is channel integration. You can message the same agent on WhatsApp during your commute, continue on Telegram from your desktop, and check in through a web dashboard later. Your conversation history, memory, and active tasks persist across all channels. This cross-platform continuity is the primary reason OpenClaw has gained traction over alternatives where each AI service lives in its own isolated interface.

Prerequisites: What You Need Before Installation

Before installing OpenClaw, confirm you have the following ready.

Hardware: Any Mac, Windows, or Linux machine works for local use. For always-on operation, a VPS with at least 2 vCPUs and 8 GB RAM is ideal. Providers like Hostinger offer suitable plans starting at $5/month.

Software: Node.js version 22 or higher is required. Check your version by running node --version in your terminal. If it returns a number below 22 or says “command not found,” download the LTS version from nodejs.org.

AI provider account: OpenClaw needs an API key from at least one AI provider. Supported options include Anthropic (Claude), OpenAI (GPT-4), and Google Gemini. You can configure multiple providers and assign different models to different task types — for example, Claude Opus 4.5 for complex reasoning and a lighter model for quick lookups.

Time commitment: The initial setup takes approximately 30 minutes for a basic configuration. Adding messaging channels and advanced security measures may take another 30–60 minutes.

OpenClaw logo on the project’s title page.

How to Install OpenClaw Step by Step

Run the Installer

On Mac or Linux, open a terminal and run:

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

On Windows, open PowerShell and run:

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

The installer downloads OpenClaw, resolves dependencies, and sets up the binary. This typically takes 2–5 minutes. Once finished, verify the installation with openclaw --version. A version number confirms success.

Run the Onboarding Wizard

Start the setup wizard with:

openclaw onboard --install-daemon

The --install-daemon flag configures OpenClaw to run as a background process that survives terminal closures and system restarts. The wizard walks you through naming your assistant, selecting an AI provider, entering your API key, choosing a model, and optionally connecting messaging channels.

During this process, you will be asked for your API key. For Anthropic, generate one at console.anthropic.com under “API Keys.” For OpenAI, visit platform.openai.com/api-keys. Paste the key into the terminal prompt — it won’t display on screen, but it registers correctly.

Verify the Gateway Is Running

After the wizard completes, check the status:

openclaw gateway status

You should see confirmation that the gateway is active on port 18789. Open the web dashboard by running openclaw dashboard or navigating to http://127.0.0.1:18789/ in your browser. Send a test message in the chat interface to confirm the AI responds.

Setting Up Messaging Channels

Channel integration is where OpenClaw differentiates itself. Here is how to connect the most popular platforms.

WhatsApp

Run openclaw channels login whatsapp, then scan the QR code that appears in your terminal using WhatsApp’s “Linked Devices” feature on your phone. By default, the agent only responds to approved numbers. Approve contacts with:

openclaw pairing approve whatsapp +15555551234

For multi-user households, you can route different numbers to separate agents. One user gets the “main” agent with full access to private files, while another gets a “shared” agent with limited scope. This separation is configured through agent bindings in openclaw.json.

Telegram

Create a bot through Telegram’s @BotFather, copy the bot token, and add it to your OpenClaw config under channels.telegram.botToken. Restart the gateway with openclaw gateway restart and message your bot to test.

Discord

Create an application at discord.com/developers, add a bot, enable the Presence, Server Members, and Message Content intents, and copy the bot token into your config. Generate an OAuth2 invite URL with Send Messages, Read Messages, and Read Message History permissions, then use that URL to add the bot to your server.

Additional Channels

OpenClaw also supports Slack, Google Chat, Signal, iMessage (Mac only), and Matrix. Each requires its own authentication flow, documented at docs.openclaw.ai.

Security Configuration: Protecting Your Agent

An AI agent with shell access, API keys, and browser control can cause serious damage if misconfigured. Security is not optional — it is the first thing to configure.

Lock Down the Gateway

Never expose the OpenClaw gateway to the open internet. Bind it to localhost only and access it through SSH tunneling:

ssh -L 18789:127.0.0.1:18789 user@your-vps

If someone gains access to your gateway, they control your calendar, email, files, and anything else the agent can reach.

Rotate the Authentication Token

The gateway auth token functions as your master password. Use a strong random string, store it in your .env file, and rotate it regularly. Never hardcode tokens in configuration files.

Run With Least Privilege

Create a dedicated system user for OpenClaw with minimal group memberships. Never run the agent as root. If you only need calendar access, do not grant shell access. Keep an explicit allowlist of approved shell commands and block destructive operations like rm -rf or dd.

Isolate the Environment

Run OpenClaw inside a Docker container to limit the blast radius of any misconfiguration. If one component is compromised, container isolation prevents it from reaching the rest of your system.

Three Deployment Tiers

Docker Deployment for 24/7 Operation

For always-on reliability, Docker on a VPS is the recommended approach. A working docker-compose.yml configuration includes several critical details that are easy to miss.

Shared memory for Chromium: If you enable web browsing, the container needs shm_size: '2gb' and /dev/shm mounted. Docker’s default shared memory allocation (64 MB) causes Chromium to crash silently.

Stale lock cleanup: Chrome leaves SingletonLock files after unclean shutdowns, preventing the browser from relaunching. Add a cleanup step to your container entrypoint:

find /data/.openclaw/browser -name SingletonLock -delete 2>/dev/null
```

**Config drift protection:** OpenClaw rewrites `openclaw.json` on startup and can strip custom settings like WhatsApp allowlist entries and agent bindings. A "golden config" pattern solves this: after startup completes (with a ~15-second delay), overwrite the config with a known-good backup. Pair this with a cron-based watchdog that checks for configuration drift every minute and auto-restores from the golden copy if needed.

**Port binding:** Bind the gateway port to localhost only (`127.0.0.1:${PORT}:${PORT}`), never to `0.0.0.0`.

---

## Replacing SaaS Subscriptions With Markdown and Git

One of the most practical outcomes of running OpenClaw is consolidating paid tools into simple markdown files managed through Git. A workspace folder structure like this replaces dedicated apps for notes, to-do lists, and habit tracking:
```
workspace/
├── HABITS.md       # Daily habit tracking
├── MEMORY.md       # Long-term notes and context
├── NOTES.md        # Categorized notes and links
├── PROJECTS.md     # Active projects and tasks
├── PROFILE.md      # Personal preferences the agent references
├── USER.md         # Quick-reference identity card
└── drafts/         # Article drafts, ideas

A nightly cron job pushes everything to a private GitHub repository. The agent also commits changes in real time as files are updated throughout the day.

This approach eliminates vendor lock-in. The files are readable in any text editor, version-controlled through Git (with full diff, blame, and revert history), and free to host on GitHub’s private repositories. The agent reads and writes these files directly — no SDK integration, no sync conflicts, no API dependency.

Practical Use Cases Worth Setting Up First

Morning Briefing

Schedule a daily prompt at a fixed time using cron or OpenClaw’s built-in scheduler. The agent pulls weather data, your first calendar events, and top headlines, then sends a sub-150-word summary to your preferred channel.

Voice Note Transcription

Send audio messages through WhatsApp or Telegram. OpenClaw transcribes them (using Gemini or another speech-to-text provider), extracts actionable items, and files them into the appropriate workspace documents. No typing required.

Email Triage

The agent scans unread messages, categorizes them by urgency, and drafts replies for high-priority items. Start with a single label or folder before granting broader inbox access.

Receipt-to-Spreadsheet Logging

Photograph a receipt and send it to the agent. OpenClaw extracts the vendor, date, amount, and category, then logs the entry in a spreadsheet that matches your expense format.

Server Health Monitoring

Set threshold alerts for disk usage (above 85%), sustained CPU load (above 80% for five minutes), and RAM usage (above 90%). The agent monitors these metrics continuously and notifies you only when intervention is needed.

CI/CD Pipeline Monitoring

Connect to GitHub Actions, GitLab CI, or Jenkins. Get alerts when builds fail, deployments complete, or test suites exceed time limits — delivered to your chat channel with commit messages and direct links.

The SuperAgent Configuration: Giving Your Agent a Phone Number

An advanced setup involves giving the agent its own phone number through a low-cost eSIM service (such as Tello, at roughly $5/month). Connected to WhatsApp, this lets the agent make calls, send texts, and handle MFA challenges independently.

The practical benefits include allowing other people to interact with the agent by texting a real number, enabling the agent to make reservations and manage appointments, and avoiding the limitations of virtual VoIP numbers that some services block. This configuration moves OpenClaw from a private assistant into something closer to a front-facing representative that can handle scheduling, inquiries, and basic communication on your behalf.

Skills and Extensions

Skills are plugins that expand what OpenClaw can do. Browse available options at clawhub.ai and install them from the command line:

openclaw skills install web-search

Recommended starting skills include calendar management, weather lookups, web browsing (with the security caveats noted above), and file organization. For image generation, the bundled nano-banana-pro skill connects to Google’s Gemini image models, while openai-image-gen connects to DALL·E.

Browser automation is available but carries higher risk. Web pages can contain hidden instructions that exploit prompt injection, potentially causing the agent to leak data or take unintended actions. Limit browser automation to internal tools you control and avoid using it for payments, account changes, or sensitive data entry.

Customizing the Agent’s Personality

Edit the SOUL.md file in your workspace folder to define how the agent communicates. You can specify tone, areas of expertise, behavioral guidelines, and recurring conversation topics. After saving changes, restart the gateway for them to take effect. The PROFILE.md and USER.md files serve as persistent context the agent references during every interaction, giving it stable awareness of your preferences and identity.

Common Troubleshooting

“Command not found” after installation: Close and reopen your terminal to refresh the PATH, or run the installer again.

Gateway won’t start: Another process may be using port 18789. Try openclaw gateway --port 18790 or identify the conflicting process.

WhatsApp keeps disconnecting: Keep your phone connected to the internet and avoid logging out of WhatsApp. Re-scan the QR code if the session drops.

High API costs: Set daily token limits in your config to prevent runaway spending:

"limits": {
  "maxTokensPerDay": 100000,
  "alertWhenOver": 50000
}

Chromium crashes in Docker: Confirm shm_size: '2gb' is set and /dev/shm is mounted in your compose file. Check for stale SingletonLock files.

Config keeps resetting: Implement the golden config pattern with a cron-based watchdog that detects drift and auto-restores from a known-good backup.

Final Word

OpenClaw is open-source, self-hosted, and capable of acting across the platforms you already use daily. The difference between OpenClaw and a typical chatbot is execution. It does not just suggest what to do; it does it, whether that means updating your calendar, committing notes to a Git repository, alerting you about a failing CI/CD pipeline, or transcribing a voice note into categorized action items.

That said, the gap between a working demo and a reliable personal agent is filled mostly by security discipline and configuration maintenance. The golden config pattern, SSH-only gateway access, least-privilege permissions, and regular token rotation are not advanced extras — they are baseline requirements. Skipping them turns a productivity tool into a liability.

The financial math is straightforward. A $5/month VPS plus API usage costs replaces a stack of $3–$10/month SaaS subscriptions while giving you version-controlled data in portable formats you own permanently. If a service shuts down tomorrow, your markdown files and Git history remain intact.

Start small. Pick two or three use cases — a morning briefing, voice note transcription, and email triage make a strong starting trio. Get those running reliably, verify the security configuration holds up across restarts, and expand from there. The agent becomes more useful as you add integrations, but only if the foundation is solid first.

If you are interested in this topic, we suggest you check our articles:

• How Much Does OpenClaw Cost per Month?
• How Simple Is It to Build an AI Agent from Scratch?
• How Simple is it for Someone With No Development Experience to Create an AI Agent?

Sources: Hostinger, Sre Chakra Yeddula on Medium, Reddit

Written by Alius Noreika