Couchbase Builds Multi-Model AI Architecture on Amazon Bedrock

Couchbase built a multi-model inference architecture for its AI developer assistant Capella iQ, using Anthropic Claude Sonnet 4.5 on Amazon Bedrock as the initial production model, achieving ~76% accuracy in internal BIRD evaluation. The architecture is based on Amazon EKS, deployed across us-east-1 and us-west-2, with private calls via VPC interface endpoints and Bedrock's cross-region inference (CRIS) automatically routing traffic between available regions without application logic intervention. This design allows model upgrades or vendor switches with only configuration updates, no code changes or downtime. For Capella iQ users, the entire inference process stays within the private network, meeting enterprise security and data residency requirements. Couchbase plans to use Bedrock's custom model import for fine-tuned lightweight models on high-frequency workloads like index recommendations to optimize costs.