AWS DynamoDB ChatMemoryRepository for Spring AI — optimized for AWS (1 read + 1 write per turn)

[docwho2]

🚀 DynamoDB Chat Memory for Spring AI

High-Performance · Low-Latency · AWS-Native · Drop-In Replacement for Spring AI ChatMemoryRepository

This sub-project provides a DynamoDB-backed Chat Memory implementation for Spring AI, engineered for the fastest possible performance when running on AWS infrastructure (Lambda, ECS, EKS, EC2, Fargate, etc.).

It replaces Spring AI’s default storage patterns—which can produce multiple reads and writes per turn—with a fully optimized design that performs:

• ✅ Exactly 1 DynamoDB read + 1 DynamoDB write (batch) per full chat turn
• ✅ Zero redundant I/O calls (thanks to per-JVM caching)
• ✅ Append-only persistence (tail-only writes)
• ✅ Built-in TTL cleanup via DynamoDB Time To Live (never execute delete operations on turns)
• ✅ AWS CRT-powered ultra-fast HTTP client for DynamoDB

This makes DynamoDB the ideal choice for chat memory when your application runs inside AWS and latency is crucial.

ChatMemoryRepository link

The top-level project is a realtime AI voice IVR built with Spring AI, AWS Lambda, and AWS Chime SMA. Because callers expect instant, conversational responses, every millisecond matters. Traditional persistence layers added too much latency per turn, so a DynamoDB-backed chat memory—optimized for single-digit-ms reads/writes inside AWS—became essential. This ensures Spring AI can maintain conversational context without introducing delays audible to the caller, enabling truly voice-grade, production-ready AI interactions.

Full Voice IVR application based on Spring AI link