4 Essential Checks Before Using AI Agents 🤖

So before you let an AI agents real business, run these four checks just in case.

This example, I've made this diagram for a very curious client to explain how the Hermes agent setup would work.

I'm not going to go through every detail of the diagram.

You can pause the video, but we're going to focus on.

The control, what the agent or can actually do on the server.

can already tell you that there will be four pillars in this video.

Key, gates, box and pose.

Let's start with gate, quite simple.

You want to set the approval mode, to human, because the default in Hermes will be smart.

And I tell you this as especially if you're getting started with this, because when you have smart, it means the AI will determine, if the operation is risky or not, if it needs your approval or not.

in the beginning you want to get a, touch of it, get an understanding.

I would say it's beneficial for you to start with some human, approval in the loop.

And that's how you do it.

Just change the setup.

And then you start to receive a lot of allow or deny questions.

And this is what we call the gate.

The second one is what I'm, calling the box, it's a container, a Docker container you set up inside your server.

So the AI agent won't have full unlimited access to your entire server or if you're running on your local machine or your entire local machine.

And that's how you prevent, leak your API keys or other sensitive information.

The number three is the key, which is the most probably intuitive one.

it's the scope.

What are you allowing your apps or the tools that you are connecting to your agent, to be able to do For instance, on a Gmail, you can give like, access to read your, emails, draft responses, but not to send emails.

And again, in the beginning you might want to have this limitation just so you get a sense of the kind of drafts your agent is creating potentially avoid any.

Email being sent that should not be sent.

And the fourth one is a pulse.

Just because a dead agent cannot tell you it's dead.

So the, the server will ping an outside, service.

In this, I'm using healthchecks IO.

It's free.

Every minute when the ping stops, you get an email saying, silence is an alarm.

In this case, you might want to check out.

Again.

Those are the four checks that I would recommend you to go through if you're getting started with, AI agent.

In this case, I'm using Hermes as an example, but that's true for if you are setting up for your business, but also for a client.

Follow for more videos like this.