Beyond Kafka and Redis, Part 2: An AI Chat Backend on NATS 2.15

In the second part of his guide, developer Don Mon explores how to build a robust AI chat backend using NATS 2.15. Moving beyond traditional stacks like Kafka and Redis, the author demonstrates how NATS’s recent features—such as priority groups, per-message TTL, atomic batch publishing, and distributed counters—can handle complex requirements like GPU work queues, token streaming, and conversation history. By leveraging NATS as a unified cloud-native backbone, the architecture simplifies infrastructure by replacing multiple specialized services with a single, high-performance messaging system. The article provides practical code examples for implementing features like run deadlines without cron services, atomic conversation turns, and per-tenant usage metering. This approach highlights how NATS has evolved to close the gap with specialized systems, offering a more streamlined and resilient solution for modern, event-driven AI applications.