Alexander SidhuMay 27, 2026
Getting Started with Axleclaw: A Complete Walkthrough 🤖
16views
Getting Started with Axleclaw: A Complete Walkthrough 🤖
Alexander Sidhu
Learn how to navigate the Axlclaw platform, from initial setup and bot onboarding to managing knowledge files, integrations, and AI model settings. This guide covers everything you need to get started with your autonomous agent.
Getting Started with Axleclaw: A Complete Walkthrough 🤖
Alexander Sidhu
Learn how to navigate the Axlclaw platform, from initial setup and bot onboarding to managing knowledge files, integrations, and AI model settings. This guide covers everything you need to get started with your autonomous agent.
Chapters
00:00
Getting Started
Introduction to the platform and initial setup process.
01:10
Platform Integrations
02:43
Knowledge Files Overview
04:42
Skills and Automation
06:33
Composure and API Access
07:38
Model Selection and Billing
09:30
System Health and Settings
Transcript
00:00
Thank you for signing up to axlclaw today.
00:02
I'm just going to show you around how it works and what the platform is for and isn't for.
00:07
So when you first sign up, what you'll see is a loading screen and not this chat interface.
00:12
It'll take you through what is happening in the background.
00:15
Basically, we are going out and provisioning you a separate computer for your instance.
00:20
So this takes a few minutes.
00:22
It can take up to six minutes.
00:23
If you do see an error, feel free to reach out to support.
00:27
but if you don't and it works, that's great.
00:30
And my suggestion is you just send it a first message to get the bot working.
00:35
Now, it's going to be a bit slow to start with.
00:36
The reason is it needs to take all that context that we provided it and start it off.
00:42
Then you can work with the bot to basically onboard yourself and onboard the agent as a actual executive assistant or whatever type of bot you're looking for.
00:52
Now, the way I would treat this platform is more of a management side.
00:56
You can use it to have these conversations, and as a testing ground and especially as at the starting point.
01:02
But in the long run, you wanted to move it to the platforms that you're most comfortable with.
01:07
So to get there, first off, we only have access to creating Telegram.
01:12
If you need Slack, Discord or something else to actually have conversations on, then feel free to reach out to support.
01:19
We'll also be working on building these integrations in the future.
01:23
These are still possible to connect to, and you can either work with support or with your bot itself to go through the process of adding Slack, Telegram, whatever.
01:33
Now, if you don't know what Telegram is, I find Telegram is probably the easiest way to use Axel Claw on the go.
01:41
Telegram is a messaging platform like WhatsApp, but because, because it has such great use of, creating bots, you can really easily connect Telegram with Axacool.
01:52
Now, Slack and Discord, very similarly, they enable you to create bots.
01:57
The only thing I would consider is WhatsApp.
02:00
The issue with connecting WhatsApp is that you actually need a separate phone to be able to chat to it.
02:05
we may add the ability to use, use one of our WhatsApp phones to connect to it.
02:11
But this doesn't mean that the bot will have access to your WhatsApp.
02:15
And I'd highly recommend not giving it access through this way.
02:19
To access WhatsApp in the future, we're going to add an integration where you can connect your WhatsApp so the bot can Also manage your WhatsApp DMs.
02:27
Now this is where the conversations and the channels will be.
02:31
The next thing I want to cover, is the knowledge files, basically because Axl Core connects to an actual computer.
02:38
The way the bot works is that it has all these files, these files in the root directory.
02:43
So all these MD files, which is short for markdown which is just a text format, is what gives the bot the ability and understanding of who you are and what it does.
02:54
You don't really need to understand a lot of this.
02:56
The bot should be proactively adding to these markdown files.
03:00
But to give you a brief overview, Agents is basically the starter starting place, which tells the bot all about the file system and what it can and can't do.
03:09
Bootstrap is the first file that it will use to onboard you and should delete it after the onboarding process is done.
03:17
And Heartbeat file is basically what keeps this bot alive.
03:21
So if you need something to be done every 5, 30 minutes, you can get it to prompt it in the heartbeat.
03:26
So every 30 minutes, whenever your heartbeat is set to, it will read this file and report back anything it needs to do.
03:33
Identity.
03:34
Obviously that is what dictates who the bot is and how the bot acts.
03:39
Memory is a key one.
03:41
The big thing I see and the big issues we see is that the memory file gets too bloated.
03:46
So here the memory file should contain the core things it needs to know on a day to weekly basis.
03:52
What will then happen is that it will take the daily memory and put it into a file in the metamary folder, which it then can call in the future.
04:01
The soul is sort of the rules and the Persona that you want it to take.
04:07
Very similar to the identity file.
04:09
Not a huge difference, but, but it basically tells it how to act.
04:13
And finally, the tools file.
04:17
This is where the bot knows how to use certain tools and we'll save it here for key integrations.
04:24
Other things you need to consider is skills so you can tell your bot to go and find and install skills.
04:29
We have a preset, number of skills that every Axle core instance starts with.
04:36
These help it act in a way that we find best, especially around improving itself memory and actually acting within the platform.
04:44
You can go to Agent Team to add multiple new agents.
04:47
My suggestion is that you start with just one and then once you feel confident that you have another agent that has a specific role, you can then create it.
04:56
Adding multiple agents creates complexity and actually uses a Lot more tokens than just having one agent.
05:02
So I'd suggest that for advanced users, after you get comfortable using just one agent.
05:08
Now, scheduled jobs.
05:10
Your bot will start with a few scheduled jobs including daily memory catalog as well as weekly self improvement.
05:16
You can add many scheduled jobs and we recommend you do that.
05:19
And the first one I suggest you start with to test it is basically a daily executive summary.
05:25
This is something that I find super useful.
05:27
If you connect your meeting, transcribers and your calendar, it will be able to give you a good idea of what your day looks like with the meetings and with the notes from the past meetings.
05:38
In technical terms a scheduled job is actually referred to as a cron job.
05:41
But we've simplified it here and your bot should be able to understand that the scheduled job is a cron job.
05:46
Now canvas is a way for you to interact with bot files so your bot can create a HTML file that you can see here.
05:54
Things like a Dashboard is cool.
05:57
You can add here.
05:58
We're also adding in the future the ability to deploy those HTML files to an actual website.
06:03
Now we have added Composure as the integration layer.
06:08
Composure allows us to connect with a bunch of apps without actually having to do the the work of creating those apps with these businesses.
06:17
Now you'll see that we have access to all of the usual things that you'd want, Outlook, Gmail, etc.
06:23
but there may be things that we don't currently have here.
06:27
There's two options for you.
06:29
It's to talk to your bot to get, to figure out how you can get access to those softwares.
06:34
If the software provides an API key, you can tell your bot to read the documentation of this, of this software and also give it the API key.
06:42
It will then store it for you.
06:44
If there is an integration and you want to know and be pointing the best way, just reach out to us and we will show you the best way to integrate that software.
06:51
If there is a really popular software that's not on here, we will go and work towards building that out into the platform.
06:58
So one we're currently working on is meta ads.
07:01
Right now there isn't a good way to connect to it.
07:05
Now looking at AI model choice, we've given a few models here.
07:09
You can actually go and further add more models.
07:12
And this is pay as you go.
07:14
So you will be charged based on a credit system that we'll cover shortly.
07:18
For the pay as you go models we've also added the ability to add codecs.
07:24
Now internally this is what we use you can pay for their Pro plan for 100 USD per month and we find it that it hits basically all the points we want and allows us to get access to the best models.
07:38
of OpenAI with a cheaper price.
07:42
We default to GLM5 but we highly recommend you use Untest Codex for your instance.
07:48
This will basically limit the amount of tokens and spends you make on an LLM.
07:55
And because openclaw is a big and useful thing, it will be take.
08:00
It will use a lot of tokens.
08:02
now you can view your usage of the pay as you go credits.
08:06
In the credits and usage, this will log all the costs for you.
08:10
We set it to GLM5 because we find that's the good balance of a good model and also cost effective.
08:16
You can switch it to something like Claude Sonnet that you're probably used to, but the cost of that is very high in comparison.
08:23
So you can see that the input tokens are three times more expensive and the output tokens are almost five times more expensive.
08:30
So that's why we Recommend Starting with GLM5.
08:32
It will be able to do most of the things you want it to do and I find that it works for the basic tasks.
08:38
Another thing you don't really need to worry about, but it's good to know is that this is where you can look at your workspace health.
08:45
This isn't super important, but it is a good overview of usage.
08:51
So if you're seeing errors and it being super slow, you can go into here and see that potentially memory or CPU usage is very high right now.
08:59
You can see that all of these are quite low for me.
09:02
The next thing is the terminal.
09:04
Now this is a technical part that I wouldn't really recommend.
09:08
You need to start with your bot has access to the terminal, so it can actually act and do these operations for you.
09:15
But for the more technical people you can actually connect to the terminal and then interact with it for the most part.
09:22
I'd suggest leaving this alone unless you want to integrate something else that the bot can't do.
09:28
Now obviously this is a platform that is going to be built upon and we're going to continually update it with the things that we find that customers want the most.
09:36
So your feedback is really appreciated.
09:38
And if you have issues feel free to reach out.
Comments
Comments
No comments yet