OpenClaw Tutorial for Beginners - Crash Course
By Adrian Twarog
Summary
## Key takeaways - **OpenClaw backed by OpenAI**: OpenClaw is an autonomous AI agent written in Typescript by Peter Steinberger, now with official backing from OpenAI as an open-source project. [00:06], [00:12] - **Install in 1-2 minutes**: To install OpenClaw, copy the quick start command and paste it into your terminal; the installation takes 1 or 2 minutes. [00:18], [00:24] - **17% skills are hacker honeybots**: Up to 17% of skills or tools are honeybots made by hackers to leak information, so use trusted man-in-the-middle like Zapier MCP instead of direct connections. [05:18], [05:30] - **Claude Opus racks up costs fast**: OpenClaw can rack up a lot of costs very quickly, with even day-to-day testing amounting to a few dollars and vastly more if running 24/7. [00:57], [05:56] - **Set identity for persistent memory**: Set up a name for OpenClaw and describe how you want it to act; this will be saved in memory and referred back to when interacting. [01:49], [02:00] - **Channels sync phone to home PC**: Channels like WhatsApp or Telegram let you interact with OpenClaw from your phone, with chat sessions directly in sync with your PC back at home. [03:40], [04:03]
Topics Covered
- OpenClaw Runs 24/7 Locally
- Granular Security Beats Direct Access
- 17% Skills Are Hacker Honeybots
- Local Models Slash 24/7 API Costs
- Sync Agents Across Devices via GitHub
Full Transcript
This is OpenClaw, an autonomous AI agent that runs 24/7 on your own PC or VPS.
Written in Typescript by Peter Steinberger, it now has official backing from OpenAI who are supporting this as an open- source project. So, for this video, I'm going to show you exactly how to set it up, configure it, and start
using it right now. To install OpenClaw, we'll head down to this quick start, copy this command, and paste it into our terminal. The installation takes 1 or 2
terminal. The installation takes 1 or 2 minutes. Once installed, we can do a
minutes. Once installed, we can do a quick configuration to get it up and running. If it's the first time you're
running. If it's the first time you're using it, I recommend just simply using quick start. Next, you'll need to
quick start. Next, you'll need to connect up an AI model such as OpenAI or Anthropic. I'm going to use Anthropic
Anthropic. I'm going to use Anthropic with its API keys. You can do that easily by heading over to the claw developer website. Click start building,
developer website. Click start building, get an API key, and create a key. Make
sure you give it a label like open claw, then copy over this key back into the terminal under the anthropic API key section. You can then pick which model
section. You can then pick which model to use. I'm going to use Claude Opus
to use. I'm going to use Claude Opus 4.6, but be aware that OpenClaw can rack up a lot of costs very quickly. Now, you
can configure a channel to communicate with OpenClaw, such as Telegram, WhatsApp, or others. For now, we'll skip this and we'll do it later in the video.
Next, we can set up some skills. I'll
also skip this section since it's not necessary and we can do this later in the video, too. There are a few more requests here like using Google maps or nano banana for videos or even notion or
openai for image gen but all of these we'll skip so that we can get it up and running. This takes us to the very last
running. This takes us to the very last option. You can now select how you use
option. You can now select how you use openclaw whether it's through the terminal or through a web UI. If you
selected the terminal one you can immediately start talking to openclaw right here in this instance. Selecting
web UI takes you to this website here with a session token and it lets you start talking with OpenClaw just like you would with chat GPT. Let's
initialize it. First, it's going to ask us about identity. We'll need to set up a name for OpenClaw as well as who we are. Here, I'm just going to ask it to
are. Here, I'm just going to ask it to refer to itself as claw and myself as Adrien. You can go into further detail
Adrien. You can go into further detail about how you want it to act and this will be saved in memory so it can always be referred back to when it's interacting with you. You can see that in action here where it's writing to
file different things like its own identity, working with the user, etc. Now let's create a channel from openclaw to WhatsApp. We can do this in the
to WhatsApp. We can do this in the terminal or during the setup process. If
you've already installed openclaw, you can run openclaw space channels space ad to configure this new chat channel. I'm
going to set up WhatsApp and then I'm going to set up telegram so you know what the process is like. WhatsApp is
pretty easy. You simply link up a QR code to the phone number you want OpenClaw to use. Once you scan it, it'll ask you if it's your personal phone number or a separate phone number, which is recommended if you're setting up
OpenClaw to run as its own agent. You'll
also need to fill out your own personal phone number here so that it knows which phone number to message you on. With
that done, now let's set up Telegram.
Run OpenClaw space channel space ad. I'm
going to select the default primary account and follow the instructions here to create a Telegram bot, which I can utilize for OpenClaw. On your phone, head to Telegram. You'll want to create
a new message to the following contact, which is atbotfather.
Select the first one with the verify tag. And in here, you'll want to set up
tag. And in here, you'll want to set up a new bot for openclaw to use. Add the
following command, which is for/ newbot.
Give your telegram bot a custom name that's unique. I'm going to call mine
that's unique. I'm going to call mine Adrian Tuarog's clawbot. You'll be given a token that you'll need to provide to open claw. So, make sure you copy this
open claw. So, make sure you copy this and then paste it here in the terminal.
With that done, we've got Telegram now set up. Scroll down to finished. And
set up. Scroll down to finished. And
this should complete our setup for OpenClaw. This will let you interact
OpenClaw. This will let you interact with OpenClaw directly from your phone as you're out and about. And these chat sessions will be directly in sync with your own PC back at home. If you use a different provider like WhatsApp or
Discord, it'll function similarly. Now,
let's connect up some tools and apps for OpenClaw to use. I'm going to see if it can connect up to my Zap year MCP server since I already have lots of apps preconfigured and connect it through
there. It's also a little bit more
there. It's also a little bit more secure since I can select exactly what I want OpenClaw to be able to access.
OpenClaw already has given me directions to set it up. All I have to do is head to the Zapier MCP server, create a new one with a type as other, and then connect up the tools which I want to
use. Right now, I just want to do some
use. Right now, I just want to do some testing. So, I'll only connect it up to
testing. So, I'll only connect it up to Gmail to find emails and maybe also to create drafts. I don't want it to start
create drafts. I don't want it to start deleting or sending emails on my behalf.
This is the benefit of being able to have a little bit more granular control over the types of apps you give OpenClaw access to. Now, to establish the
access to. Now, to establish the connection, I'm going to generate a token. And rather than me manually
token. And rather than me manually configuring this like a programmer, I'm just going to copy paste all of this content straight into OpenClaw and hope that it sets it up all for me. And after
a few seconds, here's the response. It's
all configured and it's also got the available tools I've given it access to, such as getting emails and creating graphs. Let me test that this is working
graphs. Let me test that this is working by getting the latest five emails from my inbox. And here they're presented in
my inbox. And here they're presented in a nice chat layout. Before we go further, I definitely think it's worth mentioning how important security as well as risk is when you're using OpenClaw. For example, the very first
OpenClaw. For example, the very first thing that you need to do when you install OpenClaw is to accept those risks. Now, when it comes to installing
risks. Now, when it comes to installing skills or using things like tools, then I've read up to 17% of them are already honeybots made by hackers to leak information. Pretty scary if you think
information. Pretty scary if you think about it. It's why I'm using things like
about it. It's why I'm using things like Zapier MCP and other MCP tools to act as a man in the middle as a trusted source rather than connecting it directly. Even
just using keywords like help me set up my Gmail might ask you to pass that information in and it might not even be Gmail that you're connecting to. So
definitely make sure that you are very specific about what you give it access to. Additionally, be aware that OpenClaw
to. Additionally, be aware that OpenClaw will start using up a lot of API requests and this can amount to a lot of money. Even day-to-day use just for me
money. Even day-to-day use just for me testing has amounted to a few dollars and this could be vastly more if you're running it 24/7. This is where running local models on your own PC might be
beneficial through something like Olama, which has already added a configuration file here for OpenClaw. If you haven't got Olama installed, now's the time to do that. Let me show you how to quickly
do that. Let me show you how to quickly configure this with a good model to utilize with OpenClaw. Copy the Olama launch Open Claw configuration into your
terminal. The first recommended model
terminal. The first recommended model will actually be a new one here called GLM47 Flash. And this one's around 25
GLM47 Flash. And this one's around 25 gigs. I'm going to select to utilize it
gigs. I'm going to select to utilize it and download it, which is going to happen in the background. This model
might be a little bit big and this is why a lot of people are buying Mac minis to run their open claw in isolation on their own local PC. Of course, you don't have to do it that way. You can always run it on a VPS connected to something
like open AI or claude. If you want to know more about how to customize this, there's great documentation on the open claw website. But now that we have it
claw website. But now that we have it all set up, if you check console by running open claw space models, you'll see that the lama model is now the default model just over here with some
fallback models for anthropic over here.
The next thing we'll do is open up the default folder of open claw so we can view it in VS code or cursor. To do
this, just copy over the main workspace which is normally open claw just as I've presented here inside of cursor. This
workspace will give you access to everything from the agents and their configuration sessions and logs as well as things like automatic chron jobs, our MCP connection to Zapier and much more.
What's useful about this workspace is that you can now sync it to GitHub and that way you always have a copy of it in case you want to copy it over to a different PC or run very similar versions of it. Open Claw looks like a
really interesting project. I'm going to use it for 30 days building multiple agents and seeing how they actually help me on a day-to-day basis. I hope you guys enjoyed this video. A big thanks to Zapier for sponsoring it.
Loading video analysis...