About This Architecture

Agentic AI workflow architecture on AWS combines ECS-hosted Go APIs, Temporal Cloud orchestration, and external LLM providers (Vertex AI, OpenAI) for durable, event-driven agent execution. CloudFront with WAF fronts an ALB routing synchronous requests to Public API containers, while Workflow Workers poll Temporal for long-running agent tasks, caching state in ElastiCache (Valkey/Redis) and persisting results to RDS PostgreSQL Multi-AZ. Async event flows use Redis Streams with Server-Sent Events (SSE) to stream agent progress back to users in real time, decoupling LLM inference latency from API response times. This pattern demonstrates production-grade agentic system design for AWS architects needing fault-tolerant, observable AI orchestration at scale. Fork this diagram on Diagrams.so to customize worker scaling policies, swap LLM providers, or add Step Functions for hybrid orchestration.

People also ask

How do I architect a production agentic AI workflow on AWS with Temporal and external LLMs?

Use ECS to host Go APIs and Temporal Workflow Workers, ElastiCache (Valkey/Redis) for agent state caching, RDS PostgreSQL Multi-AZ for persistence, and Redis Streams with SSE for async event delivery. This diagram shows CloudFront + WAF → ALB → ECS API → Temporal Cloud orchestration → LLM providers (Vertex AI, OpenAI) with real-time progress streaming to users.