FWAI Codex AI Call

Production-ready AI call agent service that integrates n8n -> application -> Plivo -> user. It supports configurable, per-client question flows, low-latency scripted prompts, and end-of-call summaries.

Features

• Config-driven question flow per client
• Preloaded greeting and scripted prompts
• Optional short acknowledgements before questions (configurable)
• Plivo call control with speech input
• Objection handling and interest scoring
• End-of-call summary + transcript delivery to n8n
• Optional per-turn transcript webhook to n8n
• Gemini audio generation for all prompts (no Plivo TTS)
• Optional Gemini live streaming mode (barge-in)
• Minimal logging at key integration points

Quick start

1. python -m venv .venv
2. source .venv/bin/activate
3. pip install -r requirements.txt
4. cp .env.example .env
5. edit .env
6. uvicorn app.main:app --host 0.0.0.0 --port 8000

n8n flow

• Import workflows from n8n_flows/
• Start call webhook: /webhook/fwai_start_call
• End webhook: /webhook/fwai_call_end
• Turn webhook (optional): /webhook/fwai_call_turn
• Webhook URLs can be set per-client in the client config (preferred), or in .env as defaults.

Plivo setup

• Set PUBLIC_BASE_URL in .env to your public HTTPS URL (e.g., ngrok).
• The app uses PUBLIC_BASE_URL for Plivo callbacks.
• Supported callback paths: /plivo/callback and /plivo/hangup.

Gemini audio modes

• AUDIO_MODE=gemini: pre-generated Gemini audio played with <Play> (no latency during call).
• AUDIO_MODE=gemini-live: Gemini Live streaming over WebSocket with barge-in.
• AUDIO_STRICT=true ensures all prompt audio is generated before the call starts (zero latency during call).
• USE_GEMINI_AUDIO_CACHE=true and PREWARM_ON_STARTUP=true are recommended for low latency.

Acknowledgements (optional)

• Set ack_enabled=true and add acknowledgements / ack_rules in the client config.
• The agent will say a short acknowledgement before the next question while staying in the scripted flow.
• In strict audio mode, acknowledgement + question combinations are pre-generated for zero latency.

Notes

• API keys are loaded via environment variables only. Do not commit secrets.
• For ultra-low latency, keep prompts scripted and preloaded in config.