Respondo-Vapi

AI phone support demo using Vapi. Callers dial a real phone number, talk to a voice agent, and the agent uses tools to look up orders, check hours, search policies, and escalate to humans.

How it works

Caller dials phone number
    │
    ▼
Vapi (voice platform)
    ├── Deepgram STT (speech → text)
    ├── GPT-4.1 LLM (decides what to say / which tool to call)
    ├── Vapi TTS (text → speech)
    │
    └── Tool calls ──POST──▶ This server (FastAPI)
                                ├── lookup_order
                                ├── business_hours
                                ├── search_kb
                                └── escalate_to_human

Vapi handles the entire voice pipeline. This server is a stateless tool server — it only gets called when the LLM needs data.

Setup

python -m venv venv && source venv/bin/activate
pip install -e ".[dev]"
python scripts/index_kb.py

# Start the server
uvicorn respondo.main:app --reload --port 8000

# Expose publicly for Vapi
cloudflared tunnel --url http://localhost:8000
# or: ngrok http 8000

# Fill in keys
cp .env.example .env
# Set VAPI_API_KEY, VAPI_PUBLIC_KEY, PUBLIC_WEBHOOK_URL

# Create the Vapi assistant
source .env && python scripts/provision_vapi.py

# Attach the assistant to a phone number in the Vapi dashboard
# Call your number!

Test orders

Order	Zip	Status
NW-1001	97201	Shipped
NW-1002	10001	Processing
NW-1003	94110	Delivered
NW-1004	60614	Delayed (weather)
NW-1005	30301	Out for delivery
NW-1006	02139	Lost
NW-1007	98101	Refunded

Project structure

respondo-vapi/
├── agent/
│   ├── config.yaml          # Vapi assistant config
│   └── system_prompt.md     # Voice agent prompt
├── kb/*.md                  # Knowledge base (5 files)
├── server/respondo/
│   ├── main.py              # FastAPI webhook server
│   ├── db.py                # SQLite call logging
│   ├── tools/
│   │   ├── schemas.py       # Tool definitions (Pydantic + Vapi format)
│   │   └── handlers.py      # Tool implementations
│   └── mocks/orders.py      # Mock order data
├── scripts/
│   ├── provision_vapi.py    # Create/update Vapi assistant
│   └── index_kb.py          # Index KB → SQLite FTS5
└── docker-compose.yaml

See also: respondo-ollama — same tools running locally with Ollama (no API keys needed).