10 things you didn't know about cloud code, not tricks, but fundamentals that you should have
been taught before installing cloud code in the first place. If you want to get paid for AI
consulting or services, what I'm going to cover in this slide, especially point nine and 10,
is what clients actually ask. Number one, cloud code was built for code base, and code base is
just a fancy word for a folder with files. Cloud code or agentic coding assistance, they actually
work in that folder with files. And they can read your files, they can edit, they can run scripts and
commands on your computer, and breakthrough, it can loop. More on looping just in a second. When
people say like agents need tools or agents use tools or Cloud Code tools, Cloud Code comes with
a basic set of tools. Read, edit, write, bash, glob, grab, agent, web search, web fetch, and task.
When you work with Cloud Code, you will start seeing these tools, and you need to kind of understand what they are.
Read is Cloud Code's ability to read the content of a file.
Edit means that it can edit the content of a file.
Write actually creates a new file.
Bash is a tool that allows Cloud Code to run actual commands on your computer.
And what is the command?
Command, for example, install.
Install the package is a command.
Run scripts or checking system information.
Every time Cloud Code runs bash commands, you can actually see what it is.
Glob is kind of like a search.
It allows to find file by name.
Grip is different.
Don't confuse it with Glob.
Grip searches within a file.
So usually agents would go and use Glob and then use Grip.
Agent tool allows Cloud Code to launch another helper agent or start a side quest and a loop.
Keep in mind the word loop.
This is key.
Web search and web fetch is similar to glob and grab.
Web search allows agent to go and search information using keywords to fetch
lists of links.
Web fetch means that you go into the link and actually extract information.
In other words, read the page.
Task tool is an ability for cloud to track its process.
It's literally a to-do list file that it goes and references after it was done.
So now let's talk what is agent, what is agentic assistant, and how does cloud code actually work.
Let me introduce you to the loop.
Cloud code agents run on the loop.
That's what agents are.
People, it's just not sexy to call it loops.
First, it observes.
It uses all those tools that allows it to find information or read information.
Plan, it creates the file, the to-do list, and this is where model shines
because it can actually think what to do before touching or actually act.
Before sending any project, use plan mode.
Tell it to plan first before going and touching files.
In the act stage, this is where agent uses all the tools that modifies its environment.
So bash commands, writing, creating new files, spinning up agents.
And those agents have their own loop.
They need to, again, do observe, plan, act, verify.
verify though this is that agent is forced to go and look at the plan at its to-do list at
executions and ask like did we achieve the goal if not loop starts all over again it goes observes
plans again a new task acts and then again verifies my co-founder shirt put incredible
free resource for our community members all about loops simply explained that visual exercises
it's probably the most simplest best resource I've seen online for loops. Okay, Clot code can be run
to four places, but at its core it was designed for terminal. Another word for terminal is command
line interface. This is the little black box and to power up Clot, all you have to do after
installation is just type Clot. I search terminal in my Mac and I open the new window. If I type
cloud. Boom, cloud code is powered. As I mentioned, cloud code works in four different places. The
first one is terminal. This is where I live, for example. The second one is a terminal wrapper,
which is known as IDE, Integrated Development Environment, Antigravity, or Cursor are all
IDEs, and they all emerge and build on top of VS Code. So let me show you. This is VS Code,
or Visual Studios. It's a wrapper on top of terminals. We have our folders. We have our
files inside of folders. And we have our terminal, which is Cloud Code. It looks a little bit easier,
maybe approachable. Then Cloud edits the file. You immediately in one place can see the file.
Three is Agent is the King. It's just a fancy way of saying that we run Cloud Code in the clouds.
This way you can have cloud code abilities and all those tools that it can use, but you have it in virtual environment, in a cloud, embedded in the products that people access.
Four is desktop, mobile, and app.
You can also access cloud code.
If you have cloud desktop, this is how it looks.
We have home and then code.
And this is accessing cloud code.
And again, same thing.
You select a folder on your computer and work in the files.
The only limitation here is that you need to intentionally decide in which folder to work and not just let cloud code access all the folders and all the files.
So the functionality is limited.
I already explained agent SDK, but this is just visually for you to understand that if you have your app or solution or your business infrastructure and you want cloud code abilities, but you want it to work always 24-7 in the cloud, this is going to be agent SDK.
and agent SDK will fetch Entropic API. So it's a thing sitting in the middle. And this is what I
meant when I said that businesses do not go directly through Entropic API. They actually
have Cloud Code, agentic coding assistant sitting in the middle with all those tools and
functionalities. Another thing, when we say we use Cloud's subprocessors, it's actually another way
of saying that we run agent SDK functionality. I build OpenClaw replica inside of Cloud Code
cheap and safe. Cloud P is what allows me to control and talk with my Cloud Code on my telegram
or call it on a phone. Another thing to understand when people are confused like, oh, there's models
and everyone's chasing the best top model. The reality for businesses, it's really split in two.
There are models that you use in production in the products where speed and cost really matters.
The Frontier AI labs with the most expensive capable models are for building for hard problems.
But once you build and if that product has AI capabilities and it's going to be used by your teammates or other users,
it's going to use either cheap models or open source models.
In Cloud Code, you can access all the different models, which is Hyco, Sonnet, Opus, and NeoFable,
which I'm going to talk about in a second. However, when you are in a terminal and you do
slash command model, you have these options. However, there is a way if you don't see a
specific model, for example, you want to use Sonnet 4.6 and it's not an option here, there is a way to
set up it custom. You can ask Cloud Code to open ZSHRC file. Inside the file, there's all the
configuration. So it's different paths, different models, your API keys and stuff like that. And you
can ask it that on every launch use this model. This is what I did when they...
this is what i did when they launched opus 4.7 i didn't like it they made a default so i went
and in my settings in my configuration i changed it to opus 4.6 another thing you might not know
that you can launch clod code which is a shell sitting on different models like haiku sonnet or
opus but this shell can sit on any other model as well so for example if i type clod deep this is my
custom setup that means it's going through open router accessing deep seek model and i can use
now clod code with deep seek i can also use clod code with a completely local model on my computer
multiple reasons to do this it's different projects with the sensitive information that's one another
one is that services are not reliable and you can't access antraffic api you don't want your
work to stop and you can use another model option three is limitations if you run into your
subscription limitations and you want to continue working but antraffic opus 4.8 api is expensive
Maybe you just route to GLM models, which are open source, a fraction of the cost and capability on the same part as Opus 4.8.
We call it hybrid approach or fallback models.
Also in the same setup file that I told you, create that in case there is an outage on traffic, it will automatically default to open router.
And then if OpenRouter is also not working, I can still access my files and information and work with limited capabilities, but I can use local models.
So this is known as fallbacks.
Now we have a new model, which is MIFOS and Fable.
Fable is just MIFOS with a guardrails, but you can't work on biology type of projects and cybersecurity.
It's really sad because if you want to protect your development in your apps, you can't.
You can't hack others. You can't also safeguard yourself. It's also the most expensive model out
there. The thing to know about this next level models, remember what I said, cheap models for
production, expensive models for work. Same principle applies. You never use Fable for
everything. You use Fable for two steps in the loop. It is for planning and verifying. You never
use Fable for observe or act. Here you can have way cheaper models running those commands that
are not destructive. Another thing that people don't really know or think about is the effort.
Before you switch from one model to another, that is a sudden jump and huge price increase. Use
efforts to max out with your effort on one model. And if their result's not good, this is when you
graduate to the next model. Slash command effort, low, medium, high, x high, max, and ultra code.
Next, this is for consultants and AI service providers. Seven places where CloudCode can run
and only five is where you will do it for business. Antropic API is really for developers,
but not businesses. Businesses do not go directly to Antropic API. In Antropic console,
So you can run cloud code instance, which is called managed agents.
They're just running on Anthrofix cloud, not on your premises or not on your machine.
Cloud teams, you buy seats for that team.
This is for the very small teams who do not need a central secure login space and admin controls.
If you don't need full infrastructure with IT departments and compliance, you can use cloud teams.
Now, enterprise.
This is where real money is really. You need admin controls, but also higher limits and rate limits.
Enterprises can't afford that. Hey, you run into rate limits. Oopsie. And rate limits is basically
how often you can use this product and how fast. For the enterprises and businesses,
when you talk about cloud code, the number one question is, do they run their business in cloud
or on a premise. If it's on premise, meaning they have their own servers, you use cloud enterprise.
Next question is that if they run on cloud, do they use Amazon, Google, or Microsoft? And depending
on their answer, you will have to use Amazon Bedrock and access cloud code there, Google
Vertex AI, or Microsoft Azure AI Foundry. To finish up, I want you to walk you through the five things
that any business or enterprise is going to ask you if it is on premise or on the cloud. And these
are the words that you need to know. SSO stands for single sign on. Very simply explained is that
you have your company's domain and every person who logs in into central place can use the same
domain and get access. So it could be Mark at Antraffic or could be Steve at Antraffic. Antraffic
domain is approved in one central place. Skim provisioning allows you to add people, update,
or remove automatically, so you don't have to manually manage those access and control in case,
for example, you don't have an employee anymore. RBAC, role-based access control, so now your
overall team can log in and use cloud code, but maybe not everyone has the same access controls
and not everyone can access the same information.
For some teams, Cloud Code can access the folder for marketing.
For another team, Cloud Code can access the folder finance.
Androfi provides compliance API,
and this is a way for compliance and security and IT department
to be able to fetch instantly what happens.
See full traceability of who used Cloud Code,
what they did, what happened, what tools have been used,
so they can trace back if there is any problems.
manage policy settings it's additional guardrails that you put on cloud code and this is usually
it department will reinforce policies of how it should be used for example maybe using cloud code
you can't go to certain domains and find information this is part of a cloud code training that we do
in our community we are getting also cloud certified and i'm going through cloud code badge
if you're interested to learn more we have a full dedicated classroom on cloud code beginner to pro
We have seven day free trials, so we would love to get to know you.
If not, I'm going to see you in the next video.
Ciao!