An open source and enterprise-grade implementation of the orchestrator-worker pattern from Anthropic's paper, "How we built our multi-agent research system," using the swarms framework.

The agent processes Pub/Sub trigger events (such as emails or transaction alerts), normalizes the payload, scrubs sensitive data (SSN and Credit Cards), defends against LLM prompt injections, evaluates transaction risk using an LLM for expenses $\ge 100 , and suspends execution for manager approval using Human-in-the-Loop (HITL) breakpoints.