nOps' Clara FinOps agent ships 75% faster with Amazon Bedrock AgentCore

nOps, an AI-powered cloud optimization platform, rearchitected its Clara FinOps agent around Amazon Bedrock AgentCore to overcome the limits of API-centric infrastructure. Previously built on Kubernetes, LangChain/LangGraph orchestration, and web API tool wrappers, Clara suffered from slow response times, inconsistent answers, and heavy operational overhead. Long context messages and a data path tied to API responses rather than a dedicated semantic layer made iteration difficult. By moving to Bedrock AgentCore—an AWS service for building, connecting, and optimizing agents at scale—nOps gained a managed agent runtime with built-in memory and orchestration, while retaining the freedom to use any framework or model. This allowed the team to ship 75% faster and focus on domain logic instead of infrastructure maintenance.

The new end-to-end solution combines a Vercel-hosted Next.js frontend with streaming Server-Sent Events, a single Strands-based agent running on Bedrock AgentCore, and a data layer built on Databricks Lakehouse Metric Views for governed analytics semantics. Databricks Lakebase (serverless PostgreSQL) stores durable application state like sessions and canvas widgets, while an async workflow layer uses DynamoDB for job tracking, SNS/SQS for notifications, and API Gateway WebSocket for long-running tasks. This purpose-built architecture reduces multi-agent routing overhead, improves response quality, and scales with customer demand. Clara already supports over $4 billion in cloud spend under management, delivering commitment optimization across AWS, GCP, and Azure, and the migration demonstrates how managed agent infrastructure can accelerate production AI features significantly.