The Real-Time Voice Pipeline
Mic → STT → LLM → TTS → playback, in under 300ms. Wired to the Lovable AI gateway and Gemini 3 Flash.
Speech-to-Text
Sub-300ms streaming STT via the Lovable AI gateway captures every word of the caller in real time.
LLM Reasoning
Gemini 3 Flash handles multi-turn intent, tool calls, and grounded answers from the tenant's knowledge base.
Text-to-Speech
Natural, interruptible voice output with per-agent voice selection (Alloy, Coral, Ash, Echo and more).
Telemetry
Live frequency bars and pipeline status (idle / listening / thinking / speaking) with turn-level transcript rows.
Realtime Transport
Production targets OpenAI Realtime + Twilio Media Streams for true human-speed conversation.
CRM Write-Back
Every turn's outcome, sentiment, action items and next-step date flow into the contact record automatically.










