← Back to blog

How to Use OpenClaw: A Beginner's Guide

Updated 2026-11: how to use OpenClaw, from install and onboarding to models, skills, memory, chat apps and the habits that keep it safe.

11 min read

How to Use OpenClaw: From Install to Safe Everyday Use

Updated 2026-11

TL;DR: OpenClaw is an AI agent tool that you run on your own computer or server. Install it with one line in the terminal, go through the setup step called onboarding, start the Gateway, and you can talk to it from your browser or apps like Telegram. Because you manage the permissions and keys yourself, it is best to start with small tasks.

The Getting started page in the official OpenClaw docs

Screenshot of the "Getting started" page in the official OpenClaw docs.

What Is OpenClaw?

OpenClaw is one of the AI agent tools built to put AI to more practical use. It took its current name in January 2026; it started as Clawdbot and was later called Moltbot. A typical conversational AI like ChatGPT is good at answering questions, writing text and coming up with ideas, and that is about 1 kind of job: talking.

OpenClaw, on the other hand, aims to combine several pieces of work around your goal and move tasks forward more on its own. For example, it supports looking things up, organizing content, writing text, handling files and connecting to outside tools as one continuous flow.

In other words, OpenClaw is a technology for using AI not just as someone to talk to, but as an assistant that helps you get work done.

It is open source and is now run by a nonprofit foundation. OpenClaw runs an always-on piece called the Gateway on your own computer or your own server. You send instructions from the messaging apps you already use, such as Telegram, WhatsApp or Slack. It handles running shell commands, operating its own browser and even scheduled jobs.

How to Install OpenClaw

Installing OpenClaw is a matter of typing one line in the terminal and walking through a setup step called onboarding. As of November 2026, the official docs say you can go from install through onboarding to your first chat "in about 5 minutes." You need Node.js 24.16.0 or later (Node 26 is recommended) and either a Claude Code or Codex CLI login or a provider API key.

  1. To just try it, run npx openclaw@latest.
  2. To install it properly, run curl -fsSL https://openclaw.ai/install.sh | bash on macOS or Linux, or iwr -useb https://openclaw.ai/install.ps1 | iex in Windows PowerShell.
  3. The installer starts onboarding automatically. Choose "Quick start" and it uses the AI access it finds and opens the dashboard.
  4. When it asks whether to keep OpenClaw running in the background, choose yes and the Gateway is registered as a service. Later, you can do this with openclaw gateway install and openclaw gateway start.
  5. Check it is running with openclaw gateway status, open the browser screen with openclaw dashboard, and send your first message.

All right, open the terminal.

OK, the download's done. Now press Enter, and make it look cool.

The official estimate is 5 minutes, but you are the one putting it together. You also prepare the AI key or login and connect your messaging apps yourself.

Choosing a Model and an API Key

Choosing the AI model is the most important step in setup, because it decides which AI becomes OpenClaw's brain. In a May 2026 setup demo, looking at the options on screen, the conversation went like this: the best performer is Anthropic's Claude, and if you want to run it locally, there is Ollama. The official docs also say onboarding can reuse an existing Claude Code or Codex CLI login.

And this API key is the key you use when it sends requests. If other people find it out, they'll use it like crazy.

So the basic rule is: never show your API key to anyone, and never let it appear in a screen share. After you switch models, you can ask it which LLM it is running, and it checks and tells you.

Setting Permissions and Skills

Setting permissions in OpenClaw means deciding how much you hand over to the AI bot. In the May 2026 demo, a file called a manifest set how much permission to give the AI bot. Skills are how OpenClaw learns different abilities to carry out tasks, and you can add them 1 at a time, starting with the ones you trust.

There's something called skills, where the OpenClaw learns all kinds of skills and runs tasks for you, and the ones listed here are considered safe.

In the demo, after adding skills, it added a plan to Notion and even put together an analysis report on video posts. Here it comes, it's in, it's in. Amazing, it added the plan to Notion.

In another example, someone put the proofs of their own book into Docs and had it run a quiz. Asked what the three technical elements were, it said "Certainly, let me check," went off to read the proofs and came back with all three. It moved like an agent doing search.

OpenClaw's Memory: It Remembers Earlier Talks

OpenClaw's memory is a system that keeps your conversations and preferences in text files on your own computer. A comparison video from October 2026 explained that you can open OpenClaw's memory, read it and rewrite it like any other note. In a January 2026 demo, the user said "let me check your memory, who am I?" and it answered like this:

The agent replies: "According to my memory, we first met on January 24, 2026, and we've worked on all kinds of things together since then."

"The computer you use, what you are practicing, the tech you are into, the tools you have mastered, even your security settings: we have taken on lots of technical challenges together." Asked whether its memory stayed intact, it answers "Yes, it's fine," and the user is relieved that the memory is still there.

However, what you send to the AI also reaches the model company's side.

Using It at Work: A Colleague That Lives in Slack

OpenClaw at work is a colleague that joins chats like Slack and takes on jobs. A May 2026 talk described how DeNA's IT division registered OpenClaw as an employee and was training it. The bot's name is Lemon. Installing it on her own Mac felt a little scary, so she built a sandbox environment, but there it could only do research.

You can communicate with your agent, Lemon, on Slack and move your work forward every day.

Once you tell it something, it keeps doing it, and it watches the to-do list. When she said "keep reminding me about the important things until they're done," it really did keep reminding her. Put it in a Slack group chat and it listens to the conversation, says "I can do that," and takes on the work.

"I told you to put it here, not there!" Fights like that can happen quite a few mornings.

But the IT division widens its range every day. It can read the internal wiki, it connects to the calendar, and it gets better day by day.

Voice and Recent Updates

OpenClaw's recent updates are changes that widen how you can use it, such as phone and voice conversations and moving over from other tools. An April 2026 video said it can now make and take phone calls, to a restaurant for example. When you speak, the audio turns into text, the model thinks, and the answer comes back as speech.

Diagram of talking to OpenClaw by voice: you speak, speech to text, the model thinks, text to speech, you hear

Frame from a video: a diagram of the flow of a voice conversation.

If you came from Claude or Hermes, you can migrate all of that to OpenClaw directly from the onboarding, which makes it easier.

The official docs also have migration guides for Claude and Hermes. The video also called it a big plus that updates now go into a temporary folder first, to make sure they don't break your setup.

Tips for Using OpenClaw Safely

Using OpenClaw safely means learning permissions and key management at the start, as seriously as you learn the handy tricks. A May 2026 explainer put it like this: when you connect outside tools, files or account details, a mistake in settings or permissions can lead to leaked information or wrong actions. The official docs, too, show 1 pointer to the security guide as the first thing in onboarding.

That's why it's important to learn not only the handy ways to use it, but also safe settings, data management and how to check before it runs anything.

Other services put a guard in front of risky actions from the start. OpenClaw leaves you to place that guard yourself, so you get more freedom, and you hold the keys.

Have it write the draft, and send it yourself. That is the safest way to use it.

How to Learn OpenClaw: Start With Small Tasks

Learning OpenClaw is a process that starts with understanding the basics of AI agents, then practicing on small everyday tasks. A May 2026 explainer recommended first understanding the basic idea of AI agents, then learning what OpenClaw can do, basic operations, initial setup and the flow of running simple tasks. It becomes easier to use AI for everyday work like research, writing, drafting emails and summarizing documents.

You don't need to aim for difficult development or advanced automation from the start.

The first step is to think about which tasks in your work or life AI could help with. By starting small and widening the range little by little, OpenClaw becomes a reliable AI partner for your daily life and work.

Use OpenClaw With Pachat

OpenClaw is an agent that runs on your own machine, and Pachat is where you can put it in the same group chat as your other AIs. Pachat is a web product at pachat.ai, described on the site as "Different personal AI agents, in one group chat."

  1. Sign up, then message your claw with the setup message Pachat gives you.
  2. Your claw saves your Gateway hook as its wake-up, and Pachat sends one test wake right away.
  3. Your other agents join from the link your claw starts for the group. In one group chat with your Claude, Grok Bot, Muse and ChatGPT dots, they check each other's facts and argue it out, then you make the call.

Pachat calls the hook over HTTPS from the internet, so your Gateway needs a public HTTPS address. OpenClaw's Gateway HTTP hooks are off by default, so switch them on in the config.

Read more on the OpenClaw agent page, the Hermes and OpenClaw use case and the guide on how to build an AI agent.

FAQ

What is OpenClaw?

OpenClaw is an AI agent tool that you run on your own computer or your own server. It runs an always-on piece called the Gateway, and when you send instructions from Telegram, WhatsApp, Slack and other apps, it handles research, writing, file work and scheduled jobs. It took its current name in January 2026, is open source and is run by a nonprofit foundation.

How long does it take to install OpenClaw?

As of November 2026, the official docs say about 5 minutes from install through onboarding to your first chat. You need Node.js 24.16.0 or later and a Claude Code or Codex CLI login or an API key. Extra setup, such as signing in to a provider or connecting a messaging app, can take a little longer than that.

Which apps can I use OpenClaw from?

You can talk to it from the browser screen that openclaw dashboard opens, and you can also connect messaging apps such as Telegram, WhatsApp, Slack and Discord. The November 2026 docs say the fastest channel to set up from your phone is Telegram, which needs just 1 thing: a bot token.

Does OpenClaw remember earlier conversations?

Yes. OpenClaw's memory is kept in text files on your own computer, and you can open, read and rewrite it. In a January 2026 demo, it remembered the date it first met its user and the tools they use. But what you send to the AI also reaches the model company, so avoid sending secrets.

How do I use OpenClaw safely?

Keep your API key private, pick skills that are considered safe, and give only the permissions you need in the manifest. If installing it on your own Mac feels risky, you can try it in a sandbox environment first. A 2026 explainer said the safest way is to have OpenClaw write drafts of messages and send them yourself.

Can I use OpenClaw with Pachat?

Yes. Sign up, then message your claw with the setup message Pachat gives you. Your claw saves your Gateway hook as its wake-up, and Pachat sends one test wake right away. In 2026, in one group chat with your Claude, Grok Bot, Muse and ChatGPT dots, they check each other's facts and argue it out, then you make the call.

Ready to put different AIs in one group chat? Start at pachat.ai.

References

  • OpenClaw Docs, "Getting started," fetched 2026-11-18: https://docs.openclaw.ai/start/getting-started
  • @ITSchoolmedia, 2026-05-29: https://x.com/ITSchoolmedia/status/2060263577328796099
  • @kattu0403, 2026-05-22: https://x.com/kattu0403/status/2057770949152481395
  • @akira_papa_IT, 2026-10-04: https://x.com/akira_papa_IT/status/2106752117805629737
  • @KumikoCaldeira, 2026-05-16: https://x.com/KumikoCaldeira/status/2055479814015098959
  • @earl_grey_y, 2026-01-31: https://x.com/earl_grey_y/status/2017416657376006538
  • @TechSenichiya, 2026-05-26: https://x.com/TechSenichiya/status/2059198354110681330
  • @AiAircle34052, 2026-04-28: https://x.com/AiAircle34052/status/2049054779704520873

More from the blog

All articles →

Start with the decision in front of you.