OpenClaw is a tool that everybody is using right now. It’s literally everywhere. But let me throw a question out there: is it actually useful? Because seeing cool demos on Twitter is one thing, and having it actually help you in your day-to-day is a completely different story.
Well, today I’m going to show you how I actually use it and what it does for me. I’m going to walk you through an automation I’m running on my channel to handle a complete video creation process. Although you can adapt this to your business or your own use case however you want. I’m also going to show you how to talk to it with voice messages and have it understand you, without paying for any API, completely free. At the end of the day, I’m going to show you how I use it for my work and how you can optimize it for yours.
Where to Run OpenClaw
As you probably know, OpenClaw can run on any machine. But since I like having it active all the time - it’s my 24/7 assistant - I need it running on a server. And this time I’ve chosen HostGator for hosting.
HostGator is a pretty solid hosting platform that offers many types of hosting, but we’re interested in their VPS. From their website you can check out the different options they have. Honestly, you can pick whichever server you want. I went with the cheapest one because it costs less and it’s more than enough for what I’m doing. But you can pick whatever power you need. They have plans at different prices and especially if you’re from Latin America, it’s worth checking out because their main support is there.
I already have my machine set up with OpenClaw preinstalled, which the folks at HostGator kindly set up for me. They give you all the relevant info: the IP and the port to access via terminal. You can also open a terminal right in the browser, and they show you the root password, which obviously you need to change. In my case I got two gigs of RAM and 50 gigs of storage, more than enough.
If you’re interested and need to host your own OpenClaw or any other application (we already have a ton of tutorials on the channel showing how to host many other things), then HostGator is your solution. Check out the link I’m leaving there where you can see everything they offer and the pricing.
Initial Setup and Security
Once connected to the server via SSH, since I already have OpenClaw installed, it’s time to configure it. I won’t go too deep into this because we already have plenty of videos about installation and setup, but I’ll just say that as a model I’m using Qwen 2.5. That famous model that’s been blowing up lately, although I don’t really like that word.
I’ve also set up a bunch of security policies that I talked about in a previous video. And honestly, since I couldn’t remember, I just checked my own blog, copied and pasted what needed to be done, and that was it. Didn’t take me much longer.
A reminder here: among other things, we restrict the ports we can connect to. If you’re using HostGator, pay attention because the SSH port they give you might not be the standard 22. In my case they gave me 22022, so be careful - don’t restrict the wrong port or you’ll brick the machine and have to format it.
Another thing I do is use Tailscale to connect securely without having to expose our service. This is also super important. And well, a couple more settings that I already covered in other posts.
Free Voice Messages with Groq
Now let’s get to the good stuff. First, let’s look at how I set up the voice message feature - where I send an audio and it transcribes and understands it. Because by default it doesn’t do this. OpenClaw suggests using the Whisper API from OpenAI, but that costs money. You have to load credits, credits expire… it’s a hassle we don’t need just to be able to send a voice message.
How are we going to do it? Well, in case you didn’t know, there are providers that offer access to AI APIs for free. Why free? Well, to get you using their services, they offer a pretty generous free tier. For this case we’re going to use one of these services called Groq. And no, I’m not talking about Elon Musk’s AI. No, Grok with a K… its dumb cousin.
With Groq we get a very generous free tier for Whisper that we can use without entering a credit card or anything. We go to their website, register, and we can generate an API key and start using it.
Skills: The Key to Everything
So how are we going to make OpenClaw understand us when we send voice messages? Pretty simple, and this is what the entire post is built on. To make OpenClaw learn something, you just have to tell it. But sometimes when you just tell it, the AI can hallucinate and not do things right, make too many retries, or simply not be able to figure it out.
So to go in with a sure shot, we’re going to use skills. Skills are basically markdown text files that document the process of how to do certain things. This way, the AI won’t make stuff up because it has a reference on how to do it. It’s basically like accessing documentation, but it’s more structured, more condensed, summarized, so it can understand it quickly and know how to use it without much fuss.
I went looking for skills on skills.sh, which is a site that collects open source skills. But heads up, be careful because you can find all sorts of things here, including malicious skills that might try to steal your APIs. So please be careful. Even though we’ve configured the bot with a security layer, there can always be something that slips through. Always carefully review the skills you use.
That said, I found a Groq skill. What I told OpenClaw was basically: “learn this Groq skill and from now on when I talk to you by audio, you’re going to transcribe it and understand me.” That’s it. Nothing more.
And sure enough, I tell it and it says to save the API key in the environment variables file. Thanks to our security setup, it shows me how to do it safely. I open the file, save the API key, reload the file, and with that it directly accesses Groq. I honestly prefer doing it this way because the API key doesn’t get sent through the chat or anywhere else.
And as you can see, after a bit (it’s not the fastest thing in the world with Qwen 2.5, but there it is), it works. I send an audio and it understands me. It returns what I said in the audio, which means it understood me perfectly.
Adding a Model for Programming
After setting up audio, what I do is add one more model, because I want a model for when I need to code and do heavier stuff. And this time I went with Codex. To add it I go to the terminal and type the command openclaw models login, point it to OpenAI Codex, and with that it gives me a link to do OAuth with my GPT subscription and get Codex access.
With this part done, we now have access to an extra model besides Qwen 2.5 that’s powerful for coding, plus the ability to talk to it by audio. Perfect. Now with these quality-of-life improvements, let’s see what this automation I want to build looks like.
The Complete Automation: From My Twitter Bookmarks to a Script in Notion
I mentioned something similar in a previous video, only now it’s more sophisticated and more focused on my business. Let me explain.
I have tons of bookmarks on Twitter. I’d love to be able to gather them and analyze them because I have a lot of saved stuff and I think many of them are great for channel videos. So I wanted to build an automation that I can trigger whenever I feel like it, and it does several things.
First it builds a profile of me - who I am, what I like. This is thanks to all the videos I’ve been uploading over time. Since they’re stored in Notion with script titles, it already knows what videos I’ve made. Based on that, it goes to my Twitter bookmarks, which also reflect my interests. But obviously there’s a ton of stuff saved there and it needs to pick what to suggest.
To decide what to suggest as my next video, it relies on one thing: it knows how to search for trends. I tell it “from the things I’ve given you, check what’s hottest, what’s being talked about the most on Reddit, on Twitter.” I think it even has some access to Google Trends. And with this, knowing the trends for each topic, it can decide based on what’s trending and what I like which ideas could work for upcoming videos.
And it doesn’t stop there, because it suggests three ideas (or however many you want - I set it to three). On top of explaining why each one is good, when I want to develop a specific idea, I just tell it. “Let’s develop this one.” And it starts looking for information because it can search Google too. The important thing is that it generates a script in Notion, and here’s the kicker - based on what it’s learned about me. Because since it has all my scripts, it knows how I talk, and with that it generates a draft script that obviously I need to review and tweak. Sometimes it’s too short. I don’t write my scripts with AI, but it does give me an idea of how I can approach the content. They’re more like storytelling steps than an actual script, but that’s where the value is.
I now have a complete process from validating a video idea all the way to creating its script. And all of this in a moment and super easily. So let me show you how I built it so you can grab ideas for your own automations.
Step 1: Connect the Twitter API
First and most important: the Twitter API. I think it used to be restricted or paid, but they recently updated everything and now it’s very accessible. I haven’t had to pay anything so far.
To use it, we go to the Twitter developer console. They have an admin console there where we’re going to create and manage our API keys. You need to create all of them, every single one. Don’t worry if you get a bit lost, just copy all the APIs it generates.
And what we’re also going to use is - guess what - a skill. In this case I found a Twitter skill that I liked. There are tons, but I’ll share the ones I used in case they’re useful.
Same drill: I pass it the skill URL, tell it to learn it, and then pass it the tokens. I had to ask it for help because I didn’t know which token was which (there are like four or five), but it explained them to me, I put them in the environment variables file, it picked them up, and after a little while, it has access to Twitter. It can see my bookmarks now. Awesome.
Step 2: Trends
I found a trends skill that doesn’t need any API and searches Reddit, Twitter, everything. I don’t really know how it works without an API, but honestly it works and it’s fantastic. So this one’s easy: you tell OpenClaw to install it and learn it, and that’s it.
Step 3: Connect Notion
Next up is Notion, so it has access to all my scripts and can also create new ones. This is also super easy. I found a specific skill, it learns it, and it needs the Notion API key.
We’ll go to the Notion developer area. Here we create an integration. There are two types, we can create the simple one - no need to build a full application. Once you create it, the API key appears. As always: copy it, paste it where OpenClaw tells you (which already knows that these APIs shouldn’t go through the chat and must go directly into the system), and it now has access to Notion.
The Prompt That Orchestrates Everything
One less thing to worry about. What’s left? Well, we actually don’t have any more skills to configure. What’s left is to give it the big prompt so it executes this whole cycle I’ve explained.
And this part - what I’ve explained to you - I went to ChatGPT (yes, I still use it, and sometimes I do, look at that) and explained everything. I gave it details about where in Notion my scripts are, explained the whole thing as best I could, and told it to generate a good prompt for OpenClaw. To make it do exactly what I’ve been telling you about.
And it generated a text that - heads up - you can’t pass that much text through the Telegram chat because it doesn’t allow it. So I put it in a txt file and told OpenClaw to read that txt. That’s it, a plain text file. It read it and in principle it learned it.
Testing the Automation
So I go ahead and test it. I tell it to run. Sure enough, it starts searching, goes through my bookmarks, cross-references with the trends API we mentioned earlier, and once it has everything, it makes its suggestions.
And as you can see, it suggested three things. The one that interests me, and will probably be a video soon (I hope, if I have time), is the Hytale server tutorial. And yes, it’s been on my to-do list. I need to make a video about setting up a Hytale server. I know, I know, and this thing reminded me. It’ll be one of the next things I do, just give me time.
So perfect, this fits exactly what I wanted and I have it. So I tell it to write the script in Notion. And well, I have to say I was expecting something more complete. Like I said, it seems more like a storytelling guide for the video than anything else. I imagine for these things I could now tell it to use Codex, which has a larger context window, and even though it’s built for coding, I could suggest it use it for writing this script.
And well, if I really wanted to spend money (because all of this is coming out relatively free or cheap), I could ask Opus 4.6 to do it, which is a much more powerful model. But I don’t have the Claude subscription and I didn’t see it as necessary.
This Doesn’t End Here
And with that we’d have the complete cycle. But it doesn’t stop here - I can go deeper. I can give it permissions on GitHub and teach it how my blog works. And based on what’s already written, once I’ve corrected the whole script and recorded (because I like to record first), I’d tell it to generate the blog post. That’s right, I do the blog after recording.
I haven’t implemented it yet, but I’m thinking that taking the script and the video transcript, I can tell it to write the blog for me. But do it automatically, with GitHub permissions, and make a pull request with the changes. Because if my blog is on GitHub, where else would it be? It’s another thing I’ll probably end up implementing.
Wrapping Up
Well, I don’t want to ramble any longer. Take whatever ideas you want for your own businesses. Let me know in the comments what automations you’re building - I want to know because maybe I’ll learn something I don’t know. Even if you think this can be improved or done differently and better, tell me, I want to know, I want to hear from you.
If you want a specific video about another automation, same thing, drop it in the comments and I’ll read it.
And if you need a VPS to set all this up, you already know - check out HostGator, they have great pricing and you’ll have more than enough power to run OpenClaw and whatever else you want to throw at it.
On my end, thanks for sticking around this far. That’s all folks, thanks a lot and see you in the next one.
What do you think?
Leave your opinion, question or suggestion. Comments are synced with GitHub Discussions .