Amazon has introduced a serverless image editing application built using Amazon Bedrock AgentCore, which allows users to upload photos and describe edits in natural language. The solution processes requests in seconds, leveraging three Stability AI models without requiring custom orchestration code. The application includes a React frontend, encrypted storage, and authentication, all deployed with a single command using AWS Cloud Development Kit (CDK). The agent runs in a stateful, isolated microVM with built-in memory and observability. It uses Amazon Bedrock to route prompts to the appropriate model and executes edits with minimal overhead.
The application supports prompts such as 'change the car color to blue' or 'extend the image 200 pixels to the right.' An agent powered by Claude Sonnet 4.6 breaks these requirements into steps, orchestrates tool calling, and executes edits. A watermark is added using a shell command on the microVM without consuming tokens. The solution demonstrates configuration-driven agent creation, per-invocation model switching, and persona override for industry-specific prompts. AgentCore Memory retains conversation history for 30 days, enabling the agent to reference prior edits without re-sending history from the frontend.
The architecture includes a React frontend hosted on AWS Amplify, an AWS Lambda proxy for security, an AgentCore harness with memory, and three tool Lambda functions calling Stability AI models through Amazon Bedrock. The agent is defined through API parameters, with no Python orchestration code or framework required. Tool declarations are made via the AgentCore Gateway, which routes prompts to the appropriate Lambda function. The harness handles tool selection, execution, and result integration automatically. The solution also supports model and persona switching based on user input, with the Lambda proxy routing basic chat to Claude Haiku 4.5 and edits to Claude Sonnet 4.6.
Source: awsml
