QHV

The Load Bearing Empire team publicly disclosed a production-grade voice command center architecture called QHV, using Supabase as the backend hub, VAPI for the voice pipeline, and Vercel to host the management frontend.

The QHV architecture handles 8 concurrent AI agents, distributed across three business domains: valet parking, real estate brokerage, and demolition operations.

In the QHV architecture, every VAPI Webhook event is mapped to a state change in the database; Edge Functions keep thin logic, and complex business logic is deposited into the Postgres layer.

The QHV architecture designs 8 Postgres tables to track multi-agent state, including agent registration and configuration tables, call session tables, transcription and event tables, task and automation scenario tables, and state snapshot tables.

The QHV team deployed a 5-page management frontend on Vercel, covering the chain from voice routing configuration to task automation monitoring.

The QHV architecture is suitable for scenarios that require multiple voice agents to collaborate, have clear business flow rules, and where the team is already familiar with the Supabase ecosystem; it is not suitable for scenarios with extremely complex business logic or that require a large amount of compute-intensive processing.

For larger scales beyond 8 concurrent agents (such as dozens or hundreds), whether Postgres row locks and Edge Functions concurrency limits will become bottlenecks is not given data in the original text and is an unknown that needs to be considered when evaluating.