OpenClaw Tutorial for Beginners: Your First Day at Work

Written by

You keep hearing that AI agents can run on your own machine, reply in Slack or Telegram, and finish desk work while you are in a meeting. Then you open another ChatGPT tab, type a prompt, and close it. The gap is not motivation. It is a clear first day.

This OpenClaw tutorial for beginners is that first day. OpenClaw is an open-source, local-first AI assistant you run yourself. You install a Gateway on your computer or a small always-on box, finish onboarding, chat in the dashboard, then connect one messaging channel and give it a few real work jobs. By the end of this page you will know what it is, how to get it running, and which five tasks are worth trying before you touch advanced setups.

If you only wanted ChatGPT or Claude in a browser, use those guides instead. If you want an assistant that lives where you already message people, and that can use tools on infrastructure you control, stay here.

Want a free Launchpad lesson that turns agent basics into a weekly habit? Start on the RobotsHelp home page.

No password needed – your reset opens instantly.

No password. Just your email, and you’re in.

What OpenClaw is (and what it is not)

OpenClaw’s own docs describe it as an AI assistant you run on your own infrastructure, for yourself or a small team. It replies on messaging surfaces you already use, such as Discord, Slack, Telegram, WhatsApp, iMessage, Signal, and others, and it can also do voice plus hosted widgets. The Gateway is the always-on control plane. The assistant is the product. See the official FAQ: What is OpenClaw?.

Three practical differences from a normal chat tab:

Your machine, your history. Session and workspace state stay on infrastructure you choose (laptop, Mac mini, Linux box, VPS), instead of living only inside a hosted chat product.
Real channels. You message the agent from Slack or Telegram the same way you message a colleague.
Model-agnostic routing. You can point it at providers you already use (for example Anthropic, OpenAI, or OpenRouter) and keep a local-only path when that fits your risk rules.

It is not “ChatGPT with a new skin.” It is also not a magic employee that knows your company secrets on day one. You still choose which channels can talk to it, which tools it may use, and what “done” looks like for each job. Open source means you can inspect and self-host. It does not mean “skip security.”

Who this beginners path is for

This page is for professionals who already use a computer all day, have tried ChatGPT or Claude for drafts and summaries, and now want one assistant that can stay reachable from chat and act through a Gateway they control. You do not need to be an engineer. You do need patience for a short install and honest pairing rules (who is allowed to message the agent).

Skip this path for now if you only need a one-off rewrite in a browser. Start with the ChatGPT tutorial for beginners or the Claude tutorial for beginners, then come back when you want an always-reachable agent.

What you need before you install

OpenClaw’s getting-started guide is the source of truth for versions. As of their current docs you need:

Node.js on a supported 24.x or 26.x line (they recommend checking node --version before you start). Details: Getting started.
Model access you are allowed to use: an existing Claude Code or Codex CLI login, or a provider API key. Onboarding can reuse a login it detects.
A machine that can stay reachable when you want the agent online. A laptop works for learning. A small always-on box is better once you rely on it. For Mac hardware sizing, see OpenClaw Mac mini specs and OpenClaw Mac mini setup.

Write down one work outcome before you install. Example: “Summarize yesterday’s Slack thread into three action items and send them back in the same channel.” Vague goals produce toy demos.

Install OpenClaw without getting lost

For a quick try, the docs show a one-line path:

npx openclaw@latest

On a fresh install, choose Quick start after the short security pointer. If usable AI access is already on the machine, OpenClaw verifies it with a real completion, saves config, and opens the web dashboard. Quick start also asks whether to keep OpenClaw running in the background. Yes installs a managed copy under your home folder so clearing an npx cache does not kill the Gateway. No keeps the Gateway in that terminal until you stop it.

Installer scripts for a fuller setup:

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

# Windows (PowerShell)
iwr -useb https://openclaw.ai/install.ps1 | iex

Those lines come from the official getting-started page. Prefer a deep OS guide when you want menu-bar apps, WSL notes, or troubleshooting screenshots:

After install, verify the Gateway and open the Control UI:

openclaw gateway status
openclaw dashboard

You want a healthy Gateway (their docs mention the control plane listening on port 18789) and a dashboard that loads in the browser. Send one short message in the Control UI chat. If you get a reply, the core loop works. If setup feels stuck, their read-only triage path is openclaw triage or openclaw doctor (see the same getting-started page).

Onboarding: make the first chat useful

Onboarding is not decoration. It wires auth, workspace, and the Gateway so later channels have something to talk to. Prefer Quick start when you already have a working model login. Choose Custom setup when you need to pick providers by hand.

First message ideas that prove more than “hello”:

You are my work assistant for [ROLE].
Today I only want help with email and meeting cleanup.
When I paste a messy note, return:
1) 5-bullet summary
2) Action items with owner blanks
3) One risk I might be missing
Stay concrete. Do not invent deadlines I did not mention.

Paste a real note from yesterday. Read the output like a skeptical coworker. Correct tone and format in the next message instead of starting a brand-new agent personality every time. That correction habit is what makes an agent feel trained.

Connect one channel (Telegram is a good first)

The dashboard proves the brain works. A channel proves it fits your day. OpenClaw’s docs call Telegram one of the faster first channels because you mainly need a bot token, then you can add Slack, Discord, WhatsApp, or others from their Channels guide.

Beginner rules that save pain:

One channel first. Do not wire five messengers on day one.
Pairing and allowlists. Decide who may message the agent. A public bot with no gate is a security incident waiting to happen. Follow their pairing and safety docs before you share the handle with a team.
Same job, same prompt shape. Reuse the dashboard prompt structure in the channel so results stay comparable.

Smoke test from your phone: send one paste of a short email thread and ask for action items only. If the reply lands in the same chat with a clean list, you have a working loop. If nothing replies, check Gateway status before you rebuild anything.

Five beginner tasks that beat another chatbot tab

Use these in week one. Each one should take a real work artifact you already have.

Inbox triage paste. Paste five emails. Ask for: must-reply today, can wait, and archive candidates, with one-line reasons. You still send the replies. The agent sorts.
Meeting notes to owners. Paste raw notes. Ask for decisions, open questions, and action items with blank owners. Compare against what you remember. Fix misses once so the next meeting is faster.
Weekly status draft. Feed bullet chaos from the week. Ask for a status update in your team’s format (three wins, three risks, next week). Edit the voice. Do not publish the first draft untouched.
Research brief with sources required. Ask for a one-page brief on a vendor or topic, and require links you can open. Drop any claim without a source you trust.
Personal FAQ for repeat questions. Collect the five questions colleagues always ask you. Have the agent draft short answers. You approve them. Later you can point the channel at that approved pack.

Notice the pattern: you bring the material, you keep the send button, the agent does structure. That is how beginners stay safe and still save time.

OpenClaw vs a normal ChatGPT or Claude chat

Use a normal chat tab when the job is a single document in the browser and you do not need an always-on endpoint. Use OpenClaw when you want the assistant reachable from messaging, running on your Gateway, with tools and memory you configure.

If you are choosing between OpenClaw and Anthropic’s cowork-style products, read the comparison on this site: OpenClaw vs Claude Cowork. For coding-agent shape vs OpenClaw, see OpenClaw vs Claude Code. For no-code agent builders that are not OpenClaw, the parallel path is how to build an AI agent without code.

Beginner mistakes that waste the first weekend

Installing five times because the dashboard felt slow. Run openclaw gateway status and triage before you wipe config.
Opening every channel on day one. One channel. One allowlist. One clear job.
Asking for magic autonomy with no artifacts. “Run my business” fails. “Turn this paste into action items” works.
Skipping pairing rules. If anyone on the internet can talk to a tool-enabled agent, you have the wrong setup for work data.
Never correcting the agent. The second message (“shorter, no fluff, keep my bullet labels”) matters more than the first prompt.
Chasing Mac mini shopping before a laptop proof. Prove the workflow on the machine you have. Hardware comes after a job you repeat daily.

A 60-minute first-day plan

0-10 min: Check Node, decide the one work outcome, skim the official getting-started page.
10-25 min: Install or run npx openclaw@latest, finish Quick start, confirm dashboard chat.
25-40 min: Run the meeting-notes or inbox task on a real paste. Correct the format once.
40-55 min: Connect one channel with a tight allowlist. Repeat the same task from your phone.
55-60 min: Write three rules in a note: what the agent may see, who may message it, which jobs stay human-only (payments, legal send-offs, HR).

That hour beats a weekend of videos you never apply.

FAQ

Is OpenClaw free?

The software is open source. You still pay for the model provider (or bring a login you already have) and for any host you run. Budget for API usage the same way you would for heavy ChatGPT or Claude API work.

Do I need a Mac mini on day one?

No. A normal laptop is enough to learn. Move to always-on hardware when you rely on the agent while the laptop lid is closed. Use the Mac mini guides on this site when you are ready.

Is this the same as Zapier or n8n?

No. Zapier and n8n are workflow builders that connect apps with steps you draw. OpenClaw is an assistant Gateway you talk to, with tools and channels. Many teams use both. For a first n8n path, see the n8n tutorial for beginners.

Can my whole team share one agent tomorrow?

Technically the Gateway can serve more than one person. Practically, start alone or with one trusted teammate, lock pairing, and only then widen access. Shared agents without rules create shared incidents.

Where do I go when install fails?

Use the OS-specific install articles on this site, then the official triage and troubleshooting docs. Paste sanitized logs into a coding agent only if you are comfortable with what leaves the machine.

What to do next

You now have the beginners path: install, onboard, dashboard proof, one channel, five desk tasks. Keep the agent on short, sourced jobs for a week. When the same job repeats daily, tighten the prompt and the allowlist before you add more tools.

When you want structured practice beyond articles, join Launchpad free on the home page and keep building the habit with small weekly wins.