OpenClaw Tutorial: Installation to First Chat Setup

What Is OpenClaw and Why Should You Care?

Most AI tools sit behind someone else's servers, play by someone else's rules, and forget everything the moment you close the tab. OpenClaw takes a different approach entirely. Created by Peter Steinberger and maintained as an open-source project on GitHub, OpenClaw is a self-hosted AI agent, meaning it runs directly on your own machine or a server you control. Instead of logging into a web interface, you talk to it through messaging apps you already use: Telegram, WhatsApp, Discord, iMessage, and others.

What sets it apart from a typical chatbot isn't the conversation, it's what happens after the conversation. OpenClaw doesn't just reply. It acts. It can run a web search, draft a Markdown report, write to your filesystem, execute shell commands, and remember context across multiple sessions.

Think of it less like a search engine with a personality and more like a capable assistant who shares your computer and actually follows through on what you ask.

Name History

OpenClaw has gone through a few names — ClawdBot, then MoltBot, now OpenClaw. If you see older references to those names in screenshots or documentation, they're talking about the same project.

Core Capabilities at a Glance

Live Web Search

Fetch and summarize current information from the web on demand.

File System Access

Create, read, and edit files in directories you specify.

Command Execution

Run shell commands and automate system-level workflows.

Persistent Memory

Retains context across conversations, no repeating yourself.

Multi-Platform Chat

Controlled through Telegram, WhatsApp, Discord, and more.

Fully Open Source

Free to use and modify. Third-party API costs are the only expense.

What You Need Before You Start

The good news: the bar is genuinely low here. OpenClaw's installation script handles most of the heavy lifting automatically, including dependency detection. Here's what you actually need going in:

• A Telegram account (or access to another supported messaging platform)
• Basic comfort navigating your terminal — nothing advanced required
• Node.js installed, or willingness to let the installer handle it for you
• An account with an AI provider (Google, OpenAI, Anthropic, etc.), you'll set this up during onboarding
• About 20–40 minutes of focused setup time

No Docker expertise, no DevOps background, no manual config file editing. If you can paste a command into a terminal and follow prompts, you're good.

Installing OpenClaw in One Command

The entire installation process starts with a single terminal command. Open your terminal and run:

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

Once executed, the installer takes over and works through several steps on your behalf:

OS Detection

Identifies your operating system so it can apply the right installation path automatically.

Dependency Check

Scans for missing software, including Node.js, and installs anything that's absent.

Automated Setup

Handles the full setup without requiring manual repository clones or config file edits.

Terminal UI Launch

When complete, OpenClaw opens an interactive Terminal UI (TUI) where the guided onboarding begins.

You don't need to babysit this process. Once the command is running, it completes without much input until the TUI appears.

First-Run Onboarding & QuickStart Mode

The interactive TUI walks you through initial configuration. The first prompt asks you to confirm that you understand OpenClaw can take real actions on your system, not just generate text. This isn't boilerplate; it's a meaningful checkpoint. Take it seriously.

After that, you'll be asked to choose an onboarding mode. For almost everyone reading this guide, the right choice is QuickStart.

Why QuickStart?

QuickStart applies sensible defaults that get the bot operational fast, without requiring you to understand every option upfront. You can revisit and refine individual settings later once you've seen how OpenClaw behaves in practice.

The philosophy here is deliberate: get something working first, then customize. Trying to configure everything before you've seen the agent in action tends to lead to decision fatigue and setup abandonment.

Choosing Your AI Model

With QuickStart complete, OpenClaw prompts you to select an AI provider. The options currently include:

• Google Gemini — authenticate via OAuth, then pick from available models (a fast Gemini option works well for everyday tasks)
• OpenAI — GPT-based models via API key
• Claude (Anthropic) — strong for reasoning and longer context windows
• Other supported providers depending on your version

Not a Permanent Decision

Your model choice isn't locked in. You can switch providers or add fallback models later. Pick one you have access to right now and move forward. If you're new to API-based AI access, Google's OAuth path tends to be the smoothest starting point since it doesn't require managing a standalone API key upfront.

Skills, API Keys & Hooks (and When to Skip Them)

During setup, OpenClaw introduces the concept of Skills — modular tools that extend what the agent can do. Think of them as plugins: email integration, file management routines, custom scripts, and more.

When asked about skills on your first run, here's the recommended path:

• Choose Yes when asked if you want to configure skills (to initialize the system)
• Select npm as the package manager
• Then choose Skip for now

Similarly, if you're prompted for optional API keys (for search integrations or external services) or for Hooks — skip them. The goal right now is to get the agent running end-to-end. Layering in advanced features before the basics are working is a reliable path to confusion.

Come Back to Skills Later

‍Skills are where OpenClaw gets genuinely powerful. Once you understand how the agent reasons and responds, you'll have much better intuition for which skills are worth adding and how to configure them correctly.

Control UI vs Terminal UI: Which Should You Use?

OpenClaw offers two ways to monitor and manage the agent:

Control UI (Browser)

A web-based dashboard accessible via the URL shown in your terminal. Good for ongoing monitoring, configuration review, and visual overviews of what the agent is doing.

Terminal UI (TUI)

Shows you exactly how OpenClaw is reasoning — tool calls, decisions, outputs. Ideal for first-time setup and for understanding the agent's behavior patterns before trusting it with more complex tasks.

For initial setup, stay in the TUI. When prompted to choose where to launch, select "Hatch in TUI." The transparency it offers, seeing the agent's reasoning process in real time, is genuinely valuable when you're still learning what the system does and doesn't do well.

Giving Your Bot a Voice and Identity

Once the agent is hatched, OpenClaw asks a handful of short questions to establish its personality and interaction style:

• What should the bot be called?
• How should it address you?
• What tone should it use — formal, casual, direct, conversational?

Based on your answers, it generates a personalized confirmation response, essentially introducing itself with the identity you've defined. This isn't cosmetic. The tone and memory context established here shape how the agent communicates across all future sessions.

At this point, your OpenClaw instance is fully operational. Everything after this is about extending what it can do.

Sending Your First Message

To activate the Telegram connection and hand off control from your terminal to your phone, open the Telegram bot you created, send /start, and follow the pairing prompts displayed. Once confirmed, you can close the terminal entirely.

A few good first messages to try:

• Ask about the weather in your city
• Ask it to explain what it can do
• Request a quick summary of any topic you're curious about
• Ask it to remind you of something later (tests the memory layer)

Keep the first tests simple and observable. You want to build confidence in how the agent interprets your requests before giving it tasks with real consequences.

Enabling Live Web Search

Out of the box, OpenClaw has limited access to current information. To unlock full live web search, you'll need a Brave Search API key and the easiest way to set it up is to just ask the bot how to do it.

Ask the Bot

Send a message like: "How do I enable web search for you?" It will walk you through obtaining the key.

Get Your Brave Search API Key

Sign up at Brave's developer portal and generate a free API key (the free tier is sufficient for most personal use cases).

Send the Key via Chat

Paste the API key back to OpenClaw through the Telegram conversation. It registers it automatically.

Once enabled, OpenClaw can search the live web, fetch articles, and reference current data — no more relying solely on whatever knowledge the underlying model was trained on.

File Creation and Filesystem Access

This is where OpenClaw shifts from impressive to genuinely useful. You can ask it to research a topic and save the findings directly to a file on your machine and it will handle the entire workflow autonomously.

Try asking something like: "Research the history of containerization in software and save a summary as a Markdown file."

What happens next is a good example of the agent's full loop:

• It searches the web for relevant sources
• It extracts and organizes key points
• It creates a properly formatted Markdown file locally
• It includes source references for verification

That's a real workflow, one that would normally take 20–30 minutes manually, handled in under a minute. Used well, this is where OpenClaw earns its place as a genuine productivity tool rather than a novelty.

Important Caution

‍File access is powerful in both directions. OpenClaw will create, modify, and delete files based on what you ask. Be specific about which directories it should work in, and don't test new capabilities in folders that contain anything important.

Security: What You Absolutely Need to Know

This section deserves more than a skim. OpenClaw is a system-level agent with real access to your files, commands, and processes. That's what makes it powerful, and it's also why sloppy setup can cause genuine damage.

The right mental model: treat OpenClaw like a capable junior engineer who executes exactly what they're told, without judgment about whether the instruction makes sense. If you're vague, it'll make assumptions. If those assumptions are wrong, the results might be difficult to undo.

Understand the Risk Profile

Local Machine

Risk of overwriting personal files, exposing config data, or modifying directories you didn't intend to touch.

VPS / Server

Risk of disrupting services, exposing production data, or creating security vulnerabilities at the infrastructure level.

Best Practices Before You Go Further

• Run OpenClaw as a non-root user — never as admin or root
• Set a dedicated working directory and stay explicit about file paths in every request
• Start with private chats only — don't expose the bot to groups or public channels until you fully trust its behavior
• Test any new capabilities on a disposable or non-critical environment first
• Apply the same permission logic you'd apply to a shell script, if you wouldn't grant a script those permissions, don't grant them to the agent

This Isn't Just a Chatbot

‍OpenClaw will do exactly what you ask, even when the request is ambiguous or risky. The defaults are reasonable, but the tool is only as careful as the person using it. Start small, observe behavior, and expand access deliberately.

Frequently Asked Questions

What exactly does OpenClaw do differently from a standard chatbot?

Standard chatbots generate text. OpenClaw takes actions, it can search the web, write files to your filesystem, run shell commands, and automate workflows, all triggered through natural language conversation in apps like Telegram. It also retains memory across sessions, so context isn't lost between conversations.

Is OpenClaw actually free to use?

Yes, the OpenClaw software itself is open source and free. The only costs come from optional third-party services: your AI model provider (Google, OpenAI, Anthropic, etc.) and any external APIs you choose to connect, like Brave Search. Many providers offer free tiers that cover personal use comfortably.

How safe is it to run on my main machine?

Reasonably safe when configured correctly, but it requires intentional setup. Running it as a non-root user, specifying working directories, and being precise in your requests significantly reduces risk. Don't treat it as a harmless chat app; treat it as a script runner that takes instructions literally.

Can I use OpenClaw with apps other than Telegram?

Yes. OpenClaw supports WhatsApp, Discord, iMessage, and other messaging platforms. Telegram tends to be the easiest to configure from scratch, making it the recommended starting point, but the setup flow is comparable across platforms.