The interactions API is now generally available.

Now, this is exciting because it's

a unified interface for all the Gemini models and agents,

allowing you easy access through a single API.

And this is really, really powerful

because you can see here you can create

an interaction with a model.

So this will be synchronous.

You make this call and you'll get back the response.

Or you can do this where you can have

an anti-gravity agent perform something really,

really difficult. For example, in a remote environment,

multi-step installation, setup all that single API call

on the interactions API.

So there's tons of things that you can do now.

Managed agents is the most recent release

that's available on the interactions API.

You have background execution tool improvements.

You can mix in build tools with your own tools as well.

You have Deep Research built into this

as well because you can run it in the background media

generation.

You can generate images with Nano banana.

You can generate music with lyric 3,

you can generate speech with TTS.

So tons of good things that are in here,

and it really makes the development so much easier

across all the Gemini models.

And also it is an agent first ecosystem.

So you can simply grab the Gemini interactions API skill

and get running with your favorite coding agent.

So let's do exactly that.

So here in my anti-gravity Ide, I can simply go ahead

and I can install the Gemini interactions API skills.

I will install it here in the project so that you can see it.

And so now we can also install it globally of course.

So here we have our reference.

We have our skill.

And so our coding agent will know all the good things

to do when building with the interactions API.

And what's also really exciting is that it is a stateful API.

So if you're building something that

builds upon the previous interaction.

You can very easily just pass in the previous interaction ID.

So let's do exactly that.

Please build a simple application

that allows me to edit my photos using the Gemini interactions

API.

I want to be able to put in a photo of mine,

as well as a text or a speech prompt,

and then I want to iteratively get image generations

and just iterate on them with the previous interactions ID.

All right, there we go.

And now, anti-gravity obviously also

has audio understanding built in, which is really cool.

And so you can now see that we're looking at our skills

here.

Our Gemini interactions API skills and anti-gravity

is going off and building our application here.

So yes let's allow this.

And so first of all we get our integration plan here.

So we can see yeah we'll need an API key env file.

We want Gemini 3.1 Flash image.

Yep That's good 3.5 Flash as well.

Yeah we'll just use simple NPM.

Yep That's good.

Server JS fantastic.

Yep that looks good.

So let's proceed with that.

OK go.

So we got a server JS file here.

And we can see we have an API backend here

that's being built for us.

And that has our previous interaction ID.

We'll see here our model that we'll use.

And yeah.

And then we can see our interactions create here

with our payload.

Fantastic our model.

Yeah that looks good.

OK fantastic.

So let's accept all these changes.

We will need to set up our env variable here.

So what we can do is we just can simply go to ai.dev.

So in ai.dev we can go to our API keys

and we can grab maybe one of our API keys here.

Great So I put in the API key.

Let's do a quick NPM install.

And then NPM run dev.

OK cool.

So here we go.

Anti-gravity Canvas.

We can drag or drop our photo in.

We can select the aspect ratio the resolution and then

our prompt OK sweet.

So let's drop my photo in here.

And then we're just going to say change the color of the cap

to be green.

And so now we are going to generate our hour edit.

So in the first step, we're transcribing the voice note

and then we are doing our image edit.

And so if we look into the logs here,

we can see OK we've got the input.

So that's our image input.

And then we have our transcribed text.

So in the first step, you can see here

where having an A transcribe API endpoint.

So that is also using the interactions API.

Just Gemini 3.5 Flash here to transcribe what we said.

And then we're passing that into our image editing transaction.

So Nano banana does currently not support audio as an input.

So we'll first need to transcribe it.

And then we can edit our image using the text prompt.

So you can see here change the color of the cap to be green.

And so now this is nice.

It gave us this little slider to see the outcome.

And so now here the real powerful thing

of the interactions API.

Now change the shirt to be pink.

And so now this time we actually don't need to pass in our image

anymore.

But we can simply just pass in the previous interaction ID

here.

And we will see, in combination with the prompt we have access

to the cached tokens.

And now you can see the cap is still green.

We only wanted to change the shirt to be pink.

And this is really cool Yeah Really awesome.

So you can see here and then obviously you

could go back to this, actually didn't like this.

And you can branch from here.

So you could say.

Actually change the color of the shirt to be green as well.

And so now I'd hope because it's the cash tokens,

it would use the same green as the cap.

Let's see if that is the case.

There we go.

No, it is using a different green.

OK, maybe that's a bit too much green.

But yeah, this is really, really cool.

Here the interactions API.

Try it out now.

Really recommend the coding agent skill.

It's a really, really good way to get started

with the interactions API.

Awesome let me know if this was helpful

and I can't wait to see what you will build with the interactions

API.

Cheers