Overview

What is this?

A ready-to-deploy solution that processes and generates embeddings for your images automatically in your AWS account. No need to build custom infrastructure or handle connections to AI models yourself.

What You'll Get

After deployment, you'll have a complete system that can:

• Store images in the cloud
• Generate AI embeddings for those images
• Search for similar images using text or other images
• Query your image database

Infrastructure Overview

Frame deploys several key components in your AWS account using AWS CDK (Cloud Development Kit).

Full details of the infrastructure to be deployed for each individual stack can be found in the docs folder of this repository. Please ensure you review these documents before deploying the infrastructure.

Prerequisites

Before starting, you'll need:

• An active AWS account with admin permissions
• Node.js v14.15.0+ and npm (download here)
• AWS CLI configured with your credentials (installation guide)
• AWS CDK (installation guide)
• TypeScript 3.8+ (npm -g install typescript)
• Git (installation guide)
• Sufficient AWS quota limits for the services being deployed

Deployment Process

Step 1: Get and Install the Code

Clone the repository and move into the top-level directory:

git clone https://github.com/Frame-Platform/Frame.git
cd Frame

Step 2: Build the Project Environment

Run the below command:

npm run build:all

This will run the build-all.js script to:

• Set up the project environment by installing the required dependencies for the CLI and the CDK.
• Build the CLI by compiling the TypeScript files into JavaScript.
• Link the CLI globally on your system (to enable the use of the document-embedding commands referenced below).
• Build the Lambda functions referenced in the CDK (packaging them up into zip files ready for deployment).

Step 3: Configure Your Settings

Run the below command:

document-embedding init

This command will guide you through the initialization process. In particular, you will be prompted to provide the following information:

• AWS Access Key ID (can be obtained in the AWS console)
• AWS Secret Access Key (can be obtained in the AWS console)
• AWS Region
• Database Credentials (used for setup)
  • PostgreSQL Username
  • PostgreSQL Password
  • PostgreSQL Database Name

Step 4: Deploy the Infrastructure to AWS

Run the below command to deploy the image-embedding pipeline infrastructure to your AWS account:

document-embedding deploy

During the deployment:

• The CDK will create all necessary AWS resources. This process can take 10-15 minutes, so please be patient.
• Upon completion, you'll receive confirmation of successful deployment along with important endpoints or access information for the created resources.
• In particular, you will receive confirmation of the API Gateway URL and your specific API Access Key ID details. You will need to locate the API Key associated with the API Key ID (in the AWS Management Console) in order to connect to our API endpoints via our SDK client.

Step 5: Set Up The Database

Run the below command:

npm run database:setup

This creates an optimized database table for storing vector embeddings.

Step 6: Enable Amazon Bedrock Access

For the AI models to work, you need to enable Bedrock access.

Go to the Amazon Bedrock console. Details of how to enable access can be found here.

At the time of writing, the steps are:

1. In the left menu, select "Model access" under "Bedrock configurations".
2. Select "Enable Specific Models".
3. Select the check box next to "Amazon Titan Multimodal Embeddings G1" model.
4. Click "Next" → "Submit".
5. Wait for "Access granted" status (may take a few minutes)

Step 7: Verify the Deployment

Upload a test image using our SDK client to verify everything works correctly.

Step 8 (if relevant and necessary): Destroy the Infrastructure

When you no longer need the image-embedding pipeline, you can remove all deployed resources with:

document-embedding destroy

Important:

• This action will permanently delete all resources created by the deployment
• All stored data, including images and embeddings, will be lost
• This operation cannot be undone, so proceed with caution

Usage

With your deployment complete, you can:

• Upload images for automatic embedding
• Search for similar images using text descriptions
• Find visually similar images
• Query, update, or delete your image data

Refer to the SDK documentation for integration with your applications.

Monitoring and Troubleshooting

Your deployment includes monitoring through AWS CloudWatch:

• Check Lambda function logs for errors
• Monitor database connections
• Track S3 storage operations

To access logs:

• Go to the AWS CloudWatch console
• Select the log group for the component you're investigating
• Review logs for errors or performance issues