Performant Isolation for Agentic AI

DO NOT RUN THIS CODE IN PRODUCTION. This is a demo repository to evaluate performance characteristics. It is not production code.

This repository contains demo code for a research project evaluating how AI agents perform on the Edera platform. For more details about this research, see the report at TODO.

Using this demo

There are 2 elements: agent-in-docker and agent-in-edera. As the names suggest, they implement the same agent structure in Docker and Edera. Here are instructions for running these agents

agent-in-docker

Add your OpenAI API key to the environment variable OPENAI_API_KEY.

Then run:

docker compose -f compose.yaml -f compose.openai.yaml up

agent-in-edera

This will only work on a machine with the Edera runtime installed.

Add your OpenAI API key to the deployment files where indicated by ADD_ME.

Inside the folder run:

kubectl apply -f .

The results will be available in the logs for the supervisor agent. To view them run

kubectl get pods

Note the name of the pod that begins with supervisor-agent. Then run

kubectl logs <supervisor agent name>

The timing for individual components can be seen in the logs for that component. If the run does not finish the first time, the services may not have been provisioned in the correct order. To fix this run

kubectl delete pod <supervisor agent name>