Architecture Overview

Architecture Overview

Loquent follows a clean architecture pattern with clear separation between domain logic, infrastructure, and presentation layers.

High-Level Flow

Incoming Call (Twilio)
  → Twilio Stream Handler
  → Realtime Session (OpenAI / Gemini)
  → Agent Tool Execution
  → Post-Call Pipeline
    → Transcription (GPT-4o)
    → Analysis (Configurable Analyzers)
    → Reports (Email)
    → To-Do Extraction
    → Contact Resolution

Key Design Decisions

• Multi-provider realtime — Agents can use either OpenAI Realtime API or Gemini Live API, configurable per-agent
• Per-organization configuration — Client lookup, agent settings, and reporting are scoped per org
• Core configuration table — Runtime settings (API keys, voice selection, feature toggles) stored in core_conf rather than environment variables
• Graceful shutdown — Clean session teardown on process termination
• Panic safety — Critical paths are panic-safe with proper error propagation