Run Java / Spring AI Agents on AgentCore

Key Takeaways

So, lots of people are building AI agents, but where do we run those? There's the new Amazon Bedrock Agent Core runtime, which allows us to take a container built with any agent framework and get that up and running super quick on the cloud. So, let's take a look at how. First, I'm using a Java project with Spring AI, and I've used Spring AI to create my agent. So, to do that, I have a chat client builder, and I just create a chat client. That's going to be connected up to Bedrock. I do that through a configuration parameter, so we can go into my source, resources, and see where I've actually set up the model to point to the Amazon Nova Light model. And then I've set a few more parameters in here. To run on Agent Core runtime, we need to have a request handler for {slash} invocations, and we need to have another request handler for {slash} ping, which will determine if the application is healthy, if the agent can respond to requests. So, I've set those parameters here in the configuration. Now, let's go look at the source code. I've got the chat client, and I've plugged those into a post mapping handler in Spring. So, whenever I get an HTTP post into my application, that'll be handled by this function. It's going to take a prompt and then call Bedrock and return the response. I've then containerized this application. I've uploaded the container up to AWS, and now I'm going to go create a new Agent Core runtime environment for this thing to run it. So, to do that, I'm going to say host agent. I can give it a name. I can specify my Docker container, which is this one. And then I can assign the IAM permissions and hit host agent. That'll take a couple minutes to provision and get all up and running, but once it's up and running, what we'll see is, like this one, here's the information about this agent running on Agent Core runtime. And this thing is secured through IAM. And so, the easiest way to actually test this thing is just within the console, I can go to the agent sandbox, select which agent and which version I want to use, and then put in the JSON that I want to send to that invocations endpoint, which will go to my agent, go to Bedrock, and I'm going to say, all right, my question is, tell me a joke, and I hit run. So, now Agent Core runtime will spin up my container, send the request to it. That container will then make the request to Bedrock and return the response, and there we go. Sure enough, I see my response from my agent. Joke is, why did the scarecrow win an award? Because he was outstanding in his field. So, great joke, Bedrock, thank you, and hope you try out Agent Core with Java or with whatever agent framework you want to use.