Compute Integration with Lambda

In the previous lesson, you configured Amazon S3 prefixes and Amazon OpenSearch Service vector indexes as the storage foundation for Knowledge Bases for Amazon Bedrock. Now you’ll build the compute layer that coordinates those storage workflows with AWS Lambda. Most generative AI applications need a compute layer where business logic runs, validating user requests, preprocessing documents, and making calls to Amazon Bedrock APIs. AWS Lambda fits this role as a serverless compute service that runs code in response to events without requiring server provisioning or management.

AWS Lambda is a common compute layer for applications that use Amazon Bedrock. It can respond to events from Amazon S3, run business logic for action groups in Agents for Amazon Bedrock, and power API backends that call Amazon Bedrock from frontend applications. Think of Lambda as the nervous system connecting your AI application’s brain (Bedrock foundation models) to its body (databases, APIs, storage). This lesson covers six integration patterns that together form the complete compute architecture for production Bedrock applications: Lambda as agent action group handlers, function design best practices, event-driven document ingestion, API backends via API Gateway, streaming response architectures, and concurrency management. Lambda’s pay-per-invocation pricing model aligns naturally with the variable and bursty traffic patterns typical of conversational AI applications, where usage can spike from zero to hundreds of concurrent requests within seconds.

The following diagram illustrates some patterns for connecting Lambda to Amazon Bedrock across different architectural scenarios.

Lambda as Bedrock Agent action groups

When a Bedrock Agent reasons through a user’s request and determines it needs external data or must perform an action, it invokes a Lambda function as part of its action groupA named collection of API operations that a Bedrock Agent can call during its orchestration loop, where each operation maps to a Lambda function that executes the corresponding business logic.. The agent sends a structured payload containing the action group details, operation name, and parameters extracted from the user’s request. The Lambda function then executes the required business logic, such as querying a database, calling an external API, invoking a SageMaker endpoint, or performing a calculation, and returns a structured response. The agent incorporates that response into its subsequent reasoning step before generating the final answer for the user.

Event structure and response format

The event object your Lambda handler receives includes several critical fields. The apiPath field identifies which tool operation the agent wants to invoke (such as /lookup-order or /check-inventory). The httpMethod indicates the operation type. The parameters field contains an array of name-value pairs that the agent extracted from the ...