ClusterPOS empowers its support team with an AI agent on Amazon Bedrock

Cluster Systems’ retail point-of-sale SaaS supports restaurant and retail operators whose questions split into two very different classes: “how does this feature work?” — answerable from documentation — and “what is happening in my store right now?” — answerable only from live operational data. A retrieval-only chatbot handles the first class and fails the second. Serving both from one assistant meant building a system that could decide, per question, which data source to consult, call it safely with the right credentials, and ground its answer in what came back — the defining problems of a production tool-using agent.

VeUP deployed OpenWebUI as the agent framework on Amazon ECS Fargate, fronted by an Application Load Balancer with AWS Certificate Manager TLS, and wired it to Amazon Bedrock with Amazon Nova Pro as the primary reasoning model — model choice kept toggleable so alternates can be routed per task. The agent’s tool surface has two members: a Zoho knowledge-base retrieval tool over the curated product corpus, and a live POS API tool that fetches operational data on demand — the addition that took the system from RAG chatbot to tool-using agent when it shipped to production. Responses are grounded in retrieved content and cite their sources. Amazon RDS for PostgreSQL and Amazon EFS carry state, AWS Secrets Manager holds the tool credentials, and the estate is instrumented with Amazon CloudWatch and protected by AWS Backup and KMS encryption — agent infrastructure run with the same operational rigor as the POS platform it serves.