What you are building
A personal AI voice assistant you summon from WhatsApp. It holds natural conversations, places real phone calls on your behalf, summarizes them back to you in the same chat, reads your inbox when you ask, and manages your calendar. The voice is yours to choose. The personality is yours to design.
By the end of this guide you will have an agent with a name (you pick it), a personality (you tune it), a voice (your choice), and the ability to call a restaurant, negotiate a reservation with a fallback, send an email follow-up, add the dinner to your calendar, and text you the summary, all from a single WhatsApp message you wrote in five seconds.
What it costs
There is no flat subscription. You pay for what you use, plus two small fixed costs. The math is simple.
Two small fixed costs you pay regardless of usage: a Twilio phone number at $1.15 a month, and a one-time Twilio top-up of $20 to fund outbound calling. Cloudflare Workers, the WhatsApp Sandbox, the D1 database, and Google APIs are free at this volume. Everything else scales with calls placed.
| Calls per month | Total minutes | Variable cost | Plus base | Monthly total |
|---|---|---|---|---|
| 10 calls | 30 min | $3.45 | $1.15 | $4.60 |
| 30 calls | 90 min | $10.35 | $1.15 | $11.50 |
| 60 calls | 180 min | $20.70 | $1.15 | $21.85 |
| 100 calls | 300 min | $34.50 | $1.15 | $35.65 |
For most personal users this lands between five and fifteen dollars a month. Email reads and calendar reads are free at Gmail and Google's API limits. You only pay for what your agent speaks.
The two routes
Wherever the build gets technical, you have two ways to do it.
Browser route. Open Chrome, visit each website in turn, click through their dashboard, copy values, paste them, save. No commands. No terminal. No code editor.
Terminal route. Open Terminal or PowerShell, run a few commands, edit a few files. Faster if you are comfortable with a shell.
Pick one and stick with it. The toggle on each page remembers your choice. We default to the browser route because it works for everyone.
The seven steps
Each step is its own page. Work through them in order. The progress widget bottom-right keeps your place across pages and tracks what you have finished.
- Sign up everywhere. ElevenLabs, Twilio, Cloudflare, Google, Google Cloud. Five tabs, thirty minutes.
- Build your agent. Pick a voice. Pick a name. Tune the personality. Write the system prompt. The fun part.
- Get a phone number. Buy a Twilio number, plug it into ElevenLabs. Test that your phone rings.
- Wire up WhatsApp. Connect Twilio Sandbox, deploy a small Cloudflare Worker that bridges them.
- Add memory. Cloudflare D1 database so your agent remembers prior calls. Optional but recommended.
- Email and calendar. Google OAuth, Gmail API, Calendar API. The killer-feature unlock.
- Test and ship. Three escalating tests, then send your agent into the world.
Before you click next
Have these in front of you. Each takes one to ten minutes to gather. The checklist on the next page goes deeper.
- A computer with Chrome installed (Mac, Windows, or Linux).
- A phone with WhatsApp.
- A credit or debit card. Total first-month spend: under $25.
- An existing Google account (you almost certainly have one).
- About three hours, broken across one or two sittings.