← Grok
Creator Magic
I Built an AI Radio Phone In Line (Grok Voice)
⏱ 234 min video · 2 min read7 Jul 2026Worth watching
TL;DR
Creator Mike Russell live-builds an AI phone screener using xAI Grok Voice and Twilio, aiming to let viewers call a real US number (424-500-8844), get screened by an AI agent, and be patched through live on-air during his stream. The build covers API setup, hosting decisions, and real-time troubleshooting across Twilio, Claude Code, Cloudflare Tunnels, and Proxmox.
Key points
1
xAI launched 21 new Grok Voice voices, enabling named personas like Luna and Iris with labeled use cases (e.g. sales, support)
2
The architecture chains Twilio (PSTN number) -> webhook -> Grok Voice agent (screens caller for name, location, topic, consent) -> browser dashboard -> OBS lower third
3
Mike provisioned a Santa Monica, CA Twilio number (424-500-8844) and configured a restricted API key scoped only to voice conferences and calls
4
Hosting decision shifted from Hetzner (12 euros/month) to a home Proxmox container behind a Cloudflare Tunnel, costing nothing and avoiding open ports
5
Claude Code on Claude Opus 4.8 was used as the primary coding agent throughout, scaffolding the project, validating Twilio credentials, and updating architecture docs in real time
Actionable insights
→
Use Cloudflare Tunnels instead of a paid VPS for low-traffic webhook apps you only run a few hours a day -- it costs nothing and exposes no open ports
→
When building with Twilio, scope API keys to least privilege (only the specific voice permissions your app needs) to limit blast radius if keys leak
→
Grok Voice retains full conversation transcripts with per-turn audio playback in the xAI console for 30 days, making it useful for post-call review without extra logging infrastructure
Notable quotes
“You can take the guy out of radio, you can't take the radio out of the guy.”
“What would I have done setting up Twilio without AI back in the day? Now I can just copy and paste screens into AI and say, what do I type here, bro?”
“Cloudflare workers are the wrong shape. AWS Lambda also wrong shape plus cold start latency.”
Worth watching?
✅
Worth watching the full video?
Watch if you want to see a realistic, unscripted walkthrough of wiring Grok Voice to a real phone number via Twilio -- all the key architecture decisions and credential setup gotchas are here, though the full build was still in progress when the transcript ends.
Saved you some time? The creator still deserves a like.
Found this by searching? That's the hard way.
summree watches the AI channels you follow and sends you the summary the moment a video drops. Nothing gets past you. From £4/month.
Try it free