Couchbase built a multi-model AI architecture for Capella iQ with Amazon Bedrock

Couchbase has adopted Amazon Bedrock to power its Capella iQ AI assistant, using Anthropic's Claude Sonnet 4.5 as the initial production model. The architecture is hosted on Amazon EKS across two AWS Regions (us-east-1 and us-west-2) for high availability, with a VPC interface endpoint providing private connectivity to Bedrock. Cross-Region Inference automatically routes requests to the optimal US Region for failover and load distribution. The team evaluated multiple models on benchmarks covering SQL++ generation, index recommendations, query explanations, and multi-turn conversations, with Claude Sonnet 4.5 achieving ~76% accuracy. The provider-agnostic design allows model changes via configuration without code changes. Future plans include cost optimization via fine-tuned smaller models using Bedrock's Custom Model Import. Key challenges included cross-region failover testing and model benchmarking normalization.