In this video: MiniMax launched an agent platform that aims to combine an OpenClaw-style set of functionalities with a Claude-style platform. I'll take you through the features (skills, experts, agent setup, MaxClaw, etc.), how to use the platform, and strong/weak points versus OpenClaw and/or Claude.
CHAPTERS: 0:00 OpenClaw versus Claude 0:22 Minimax Agent Launch 0:58 What is the MiniMax Agent Platform? 2:00 M2.7 Model Launch 2:50 Follow Along for Free 3:09 Navigating the Platform 3:48 Coding With MiniMax Agent 6:10 Setting Up and Using MaxClaw 9:47 Desktop Platform Version 10:18 Interacting With Desktop Files 10:45 Using MiniMax Skills 12:00 Using MiniMax Experts 13:20 Building an Expert in MiniMax Agent 15:50 Use Cases & Getting Started
------------------- Clips channel @jonlawclips
I do NOT sell any products of my own. Watch out for scams.
There's a little bit of a debate going on in the AI world.
We've had both OpenClaw and Claw Desktop out for a while now,
and they're both really cool systems.
But some people love how OpenClaw can run 24-7 on your messaging systems.
It can act like your personal assistant.
And other people love the polished, very professional,
quality experience with the Clawed models.
But there is a third horse in the race that you probably haven't heard of
that actually beat the Clawed models with a recent release,
and which was intended to basically pull together the best of both worlds
from OpenClaw and Claw Desktop.
So the editor is going to put all the benefits on the screen
that you can expect from what we're about to look into.
And it's called the Minimax Agent.
It's a really cool system.
In this video, I'm going to show you everything
and how to become more productive as an individual
for your company with this platform and set of tools.
It's going to be a good one. Let's go.
So first, what is the Minimax Agent?
So their tagline is an AI native workspace where everyone is an agent designer.
I like agent designer.
That feels like a very cool new age job title to me.
But I think that tagline covers it pretty well.
It's a platform built from the ground up around AI agents.
So AI that builds things, deploys things, runs autonomously,
and can complete tasks potentially continuously.
So 24-7 if you want them to.
And Minimax, the company behind it, is one of the leading AI labs or companies globally.
They're publicly listed on the Hong Kong Stock Exchange.
And they build their own foundation models.
And they made a lot of news and headlines just a few weeks ago
when they released a model that was Opus Clawed level.
So insanely good, like frontier good, but for a tiny fraction of the cost of Opus.
So this is a legit player and a legit stab at the current throne of Clawed,
Cowork, and OpenClaw.
And they build their own foundation models.
And that also ties into why I'm making this video now.
So their M2.7 model just launched recently.
And this is a self-evolving model,
which means it actually taught itself in part during its training phase.
And on SWE Pro, which is one of the world's hardest software engineering benchmarks,
it scored around 56 percent.
That matches the performance of models like Opus,
so the best of the best at a fraction of the cost.
It's also really good at real-world tasks across platforms like Excel, PowerPoint, and Word.
And again, it does that at like one-tenth to one-twentieth the cost of alternatives.
And that's the trademark of Minimax as a lab.
They make these really good models at a tiny fraction of the cost of competitors.
And so that new model is what is going to drive
all the updates and the features I want to show you throughout these demos.
And that in mind, we're going to dive into my computer now.
I'm going to give you the rundown on the Minimax Agent.
Now, if you want to follow along with this video and learn better,
just copy what I'm doing on the screen.
Then you can head to this link on the screen or in the description.
It takes about 30 seconds to set up, and you do help support the channel.
If you do just want to watch and learn, that is completely fine as well.
And that in mind, let's jump into my computer.
Okay, so here we are inside the Minimax platform.
This is at agent.minimax.io.
And there is a lot to cover here.
This is the main portal or interface to interact with the meta,
like the high-level Minimax agents.
There's two modes.
We have Air, which is faster response time,
thinks through things less, and then Max for the highest level of intelligence.
We can access different skills here, all kinds of different experts.
We're going to get into all of this.
And in terms of the higher level automations, the agentic work,
right here we can access sub-agents and MCT capabilities.
Again, there is a lot to get in here, and we are going to do a deep dive into all of it.
And I want to start off here with one of the primary use cases for agents,
and that is just their ability to code and program.
So that could be web apps, that could be animations,
anything that you need to exist, to organize data,
to visualize, to perform a given function.
You can do that with programming.
So I want to show you just a few web apps that we can build with the Minimax agent.
And then we'll dive into the main story of today's video, which is Max Claw,
and we're going to get into that in a few minutes.
Now for these app builds, I'm going to give two prompts to the agents.
First is going to be build a real-time investment research dashboard
for tracking AI companies.
So like a Bloomberg terminal style meets a modern SaaS product.
As far as the model being used, this will be Minimax M2.7.
We won't add any skills or experts.
I will leave this on Max and let's go.
Now in a second tab, I'm going to do the same thing
just so we can contrast and compare.
And this will be a premium freelance agency website
for a one-person AI consulting studio.
So let's say you're starting a new business
and you want a website for that.
Same settings, and let's go ahead and start.
And we can see the agent thinking and going ahead and building the app.
And as far as the architecture goes on this model and platform,
basically the way it works is the one agent kind of manages the project,
and then it pulls from all these other specialized skills,
which go and do specific things,
like let's say the visuals, the UI UX, the logos, the design,
whatever it may be.
And so that's how it's able to, as you're about to see,
create these extremely developed web apps from a single prompt.
And then there's a really special feature
on the web dev side of things that you'll see as well
once we have these demos up and running.
Okay, we can go ahead and see both results.
Let's jump into the first one right here.
We'll open it at a deployed URL hosted by Minimax.
And this is that AI consulting premium site
with all kinds of interactions, project status, and so on.
Basically, it's exactly what we want.
Contact form, premium site, lots of interactions.
And then we'll jump over here into the more complex build of the two.
And we can see that stock AI and private company watch list
and the various research tools along with a newsfeed.
So that is completely one shot, one prompt.
We get pretty complete builds here.
And I show you this as a demonstration of the inherent quality
of the Minimax agent.
But that in mind, I want to jump into the big story of this platform,
which is MaxClaw.
And so let's go there now.
You may have heard about that in the news.
It has been blowing up over the past few months.
And I really do want to stress,
OpenClaw and MaxClaw are not like a chatbot.
Instead, they run 24-7.
They run everywhere.
And they can interact and control all different kinds of apps and screens.
And finally, the way you interact with them isn't through a website or an app.
It can be through anything or anywhere.
You can communicate with your agent on WhatsApp.
You can talk to them in Slack.
This thing is like a 24-7 assistant or AI employee that you can set up.
You can give them their own social media accounts, their own email.
And they can just keep running, keep learning, and keep improving over time.
And now the problem with the standard OpenClaw is that you need Docker.
You need API keys.
You need to go through some kind of effort to set this thing up.
And for those who are not as technical, that can be a bit of an impediment.
But MaxClaw is basically the user-friendly version of that.
You can deploy MaxClaw in about 10 seconds like we're about to do.
There's no API keys, no terminal, no nothing required to do that.
And then it will run 24-7 across Discord, Slack, any app literally.
And then everything will sync as well to the Minimax platform here.
So let's jump into it.
I'm going to go to the MaxClaw tab.
And we can see a fun interpretation.
If you are familiar with the original branding around OpenClaw, I'm going to click start now.
Now I can select a specific configuration.
I'm just going to go for default for this demo.
And then I will click I'm ready.
And MaxClaw is waking up for us.
Okay, there we go.
And our assistant is set up.
Now the first thing you want to do is just connect it.
Connect it to your social apps so you can talk to it.
Connect it to apps that it needs to do valuable things for you.
And I think it's summed up well here from cron jobs that are scheduled.
So it runs autonomously all the time.
Automations, you name it.
This thing can actually do valuable work for you and engage autonomously.
So I'll go click try on Minimax app.
And so right here on the app, I can talk to my agent.
So I can control my desktop or all kinds of things or apps that it engages with on the
computer from my phone just by texting it.
Or what we'll do now in the computer is connect it, for example, with Slack so that we can
talk to our agent on that platform as well and just ask it to do tasks for us.
So I'll say set up Slack.
And then here are the instructions to set up that Slack app.
If I want to do the same for Telegram, I'll go ahead and say connect Telegram, and then
I can rinse and repeat that process there as well.
So that's how we can actually talk to this agent to start to get it to do things for
us.
I can also click skills right here.
I can click manage skills.
I can click create with Maxclaw.
And then let's just say, what else can I do with Maxclaw?
And then what I suggest you do is the second step is start connecting it again with the
actual apps that it will do work in.
So Google Docs, social media accounts, etc.
And from that point onward, just ask it to complete tasks for you to check on things,
to pull data.
Or you can set up what are called crong jobs, which are basically scheduled tasks on your
computer.
And then you can say every 24 hours do this or just every five minutes do this.
Whatever that task may be, you can ask the agent to do it as often and as autonomously
as needed.
And that overall is probably the most user-friendly version of OpenClaw I've seen, period.
OK, now before we get into the whole set of expert and skill features, I do want to touch
on the desktop version of the Minimax agent.
You can download that at this URL.
That's agent.minimax.io forward slash download.
And it is functionally everything you get on the web version, except it can also interact
in your computer with your files.
And what it looks like is just like this.
So we can see the same lab elements, OpenClaw skills on the left here, as well as experts.
To give you an example of how the agent can interact with your files, I'll click under
projects here and then I can select the folder I want to work out of.
So for example, I have this folder with YouTube scripts.
I could select that.
I'll say trust workspace.
And then I can say something like extract a list of tools used for these scripts and
videos.
So in sum, you get all of the features you get on the web and then that extra layer of
interactivity.
And if you like that, then you'll want to use the desktop platform.
Now, one quick note as well on skills as delineated from experts.
Experts are trained on a custom knowledge base like you saw me just upload, whereas
skills detail a certain protocol that you can plug in to your agents.
So, for example, let's say generally speaking, I want to generate a research paper in a certain
format that normal agents are not that good at.
What I can do right here is just click add to my skills.
And then now my agent is going to be really good at generating research papers in that
specific format.
Same thing with here for deep research, poly market expert.
These are just protocols that are compressed into a skill that you can use for your agents.
And the way you want to materially use that when you're going about and working on the
Minimax platform is just search for whatever you're working on.
For example, that's poly market.
Then I'd go ahead.
I'd just by default add this to my skill.
If I was doing something related to YouTube, I would just go ahead before I start the work
and I would install all of these skills for my agents if it comes up, has access to those.
Also, guys, a quick reminder, if you still do at this point want to follow along with
everything that I'm doing, you can go to agent dot minimax dot IO.
There's a completely free, no risk trial to get you started there.
So you learn better while you are spending time to watch this video.
Again, that is completely free.
And if you don't want to give it a test right now, no worries at all.
And let's get back to it.
OK, now I want to show you the last major set of features inside Minimax.
And this is the Minimax agent in a sense.
But it takes that even farther.
And it is derivative experts from the Minimax agent.
So it's a Minimax agent or the same underlying model that is trained on other information
to become an expert, a one specific thing or workflow.
And then you can use that kind of trained version of the agent to do that specific task for you.
And not only can you use generally available experts, quote unquote,
you can also create your own.
And that's the last demo that I want to show you here.
So you've seen this explore experts tab on the left.
Let's go ahead and press on that.
And this is where we can see the community of different experts that are taken from
or based upon the underlying Minimax agent.
So, for example, let's say we want to create PowerPoints in the same exact style as McKinsey.
All we have to do is create a chat with this expert,
which has been trained on that specific PowerPoint style.
And then that agent is just automatically going to be able to create
super high quality presentations just like McKinsey would.
So that's half of it.
That's the community portion of the experts.
And then we can go into my experts here.
And in a previous video about how I actually use AI to materially save me
or make me money on a month by month basis,
I mentioned a script writer, which functions in a very similar way to this.
So I want to rebuild that kind of a workflow as a Minimax agent.
Let's go ahead and click create.
And then what I'll say here is I want an expert that writes YouTube scripts in my script style
based on previous scripts that I have written as a human.
Let's just give it that.
And here's the response from the meta agent.
It's going to go ahead and help us create this script.
And this is where I'm going to upload some examples of prior scripts that I,
as a human, have written that I want the agent to base scripts off of.
So these are real scripts.
I'll go ahead.
I'll copy all of those into here.
So I'll say here are the sample scripts.
And then it had a few other questions here as well.
So I'll say educational only.
Tone is professional and engaging.
Medium length with call to actions.
And now it's going to go ahead and create that expert for us.
Okay.
And here we can see the configured agent.
So we have the name, the model, a description.
We have a set of instructions based on the provided context,
conversation starters, capabilities, and conversation starters as well.
And so basically this agent will now write scripts in my voice.
And I'll go ahead and click create.
I'll create it for only me.
And then I'll click chat with expert.
So now what I can do is I can say,
write a YouTube script in my style based on this information.
And then I would upload, for example, a brief or my notes or an idea.
And then it would flesh that out into different segments and an outline.
And that is an example of just how you can very easily
automate work in your life by taking the prior work you've done,
basically telling an expert like this, how exactly to do it,
and then just letting it do some of the legwork.
It doesn't mean these experts can do everything,
but there is a really wide range of capabilities
that you can unlock by just creating these yourself.
And even if you don't, there's this whole community
right here across every white collar domain.
And that in mind, that is everything I wanted to show you about minimax.
This has been a pretty cool video,
whether the open class side, the desktop platform experts, you name it.
I think this is 100% one of the leading platforms
for doing work autonomously and dealing with agents.
And I'm very excited to see everything coming out of this lab.
And there you go.
That is all for demos and pretty much all of what I wanted to cover about minimax.
Like I said, I think this presents a really strong set of use cases for white collar work.
So if you work in finance, e-commerce,
any kind of Excel, PowerPoint, Word, kind of document work, freelancing, etc.
I think there's a ton of tasks there that something like the minimax agent
can essentially do for you.
And hopefully this video helps you learn how you can go ahead
and actually do that for your own work.
So again, if you want to try any of that yourself
and you want to support the channel, you can head to agents.minimax.io.
That link is on the screen or in the description.
And that in mind, I hope you found this one useful.
Appreciate your time and being here.
And I will see you in the next one.