Built a Grok bot that autonomously calls restaurants and books reservations via voice. Zero human phone interaction required.
Tech stack: Grok API + ClawCall telephony service (free tier = 30 calls/30 mins). Bot parses natural language input like "book Gourmet Italia in Temecula, 4 people, tonight, around 7", autonomously looks up the restaurant's phone number, confirms booking details (time flexibility, name, callback number), then dials.
Post-call: Bot transcribes the restaurant's response verbatim, offers calendar integration (exports .ics file if no connected calendar). Guardrails: no credit card data transmission, no autonomous dialing, no voicemails unless explicitly requested, single-call limit per task.
This isn't restaurant-specific. Same architecture works for:
- Appointment scheduling (dentist, haircut, massage)
- Service bookings (tee times, court reservations, car service)
- Quote gathering (plumber, electrician, caterer)
- Hotel availability checks
The pattern: natural language input → autonomous lookup → voice call execution → structured output. Turns 10-minute phone calls into 30-second text commands.
Shared bot template available for replication. ClawCall integration is plug-and-play but any SIP trunk works. Grok needs native telephony to scale this properly.
This is what agentic AI should look like: zero prompt engineering, zero UI friction, just task completion in the physical world via legacy infrastructure (phone networks).
Tech stack: Grok API + ClawCall telephony service (free tier = 30 calls/30 mins). Bot parses natural language input like "book Gourmet Italia in Temecula, 4 people, tonight, around 7", autonomously looks up the restaurant's phone number, confirms booking details (time flexibility, name, callback number), then dials.
Post-call: Bot transcribes the restaurant's response verbatim, offers calendar integration (exports .ics file if no connected calendar). Guardrails: no credit card data transmission, no autonomous dialing, no voicemails unless explicitly requested, single-call limit per task.
This isn't restaurant-specific. Same architecture works for:
- Appointment scheduling (dentist, haircut, massage)
- Service bookings (tee times, court reservations, car service)
- Quote gathering (plumber, electrician, caterer)
- Hotel availability checks
The pattern: natural language input → autonomous lookup → voice call execution → structured output. Turns 10-minute phone calls into 30-second text commands.
Shared bot template available for replication. ClawCall integration is plug-and-play but any SIP trunk works. Grok needs native telephony to scale this properly.
This is what agentic AI should look like: zero prompt engineering, zero UI friction, just task completion in the physical world via legacy infrastructure (phone networks).