Programming Agent Server

Description

Your Programming Agent allows you to discuss with an AI chatbot what sort of custom web app you'd like it to create for you. Within a few seconds you'll see the results!

This repo contains the main backend code.

For the frontend repo, see the Programming Agent UI. For the contents of the user app Docker container (editted live by the AI), see the User React App.

Features

• Talk with a ChatGPT-like chat bot
• Discuss changes you'd like to make to the app at each stage
• Drag and drop your own images to be included in the app

Example

Architecture

The main front and back-end servers are AWS EC2 instances. The AI-created app is served from an ECS Fargate container. This container makes periodic checks to see if the code in the Amazon S3 bucket has changed. This is changed on-demand by the Golang backend, which comprises an OpenAI API GPT-4o invocation with access to tool calls.

The main frontend is an Nginx-hosted server hosting a React build, served at https://stephencowley.com (note that the setup isn't in production at the moment). Users login with AWS Cognito. From then on, the username is included as a header in requests to the backend http server, https://api.stephencowley.com. The user's website creation is an npm development server (for quick rebuild speeds) in a Docker container at https://username.stephencowley.com.

The current user state is stored in DynamoDB, and the uploaded images are stored in another S3 bucket.

Through the use of Application Load Balancers, serverless user-data storage, and Amazon Cognito, the architecture is scalable, allowing for multiple users. With some tweaks, it could possibly handle millions of global users.

Installation

Prerequisites

• An AWS account
• Basic AWS knowledge
• EC2 server running Ubuntu 24
• EC2 role IAM permissions to access S3, DynamoDB, Secrets Manager
• Go v1.22.5

To run a development http server:

go run main.go

For user data to be correctly stored in DynamoDB and S3, you will need to change the constants to match your own table and bucket names in awsHandlers/dynamoDBHandler.go and awsHandlers/s3Handler.go. Then the http server will be able to change S3 files.

To create the container that will listen to changes to S3, run a Docker container based on the container User React App. Follow the installation guide for that repo.

The CORS headers added in apiAgent/apiAgent.go will need to be changed to allow localhost if running the frontend locally, or removed entirely if just a testing of the endpoints is wanted.

Note also that currently, the OpenAI API key is securely extracted from Amazon Secrets manager. You may need to change the code to allow your method of extracting an OpenAI API key.

This project contains functions in awsHandlers/ecrHandler.go and awsHandlers/ecsHandler.go that automate the creation of the container, but this is not required.

You could utilise the YAML spec files to create an automated CI/CD pipeline with AWS CodePipeline, but this is not required.