← All cases

Voice and agents

Voice agents: from HUD to a voice-driven terminal

Three voice systems in production: a WhatsApp assistant that talks, a JARVIS-style panel with camera and tools, and a bridge that lets you give a code agent voice commands and get a spoken summary back.

30
voices in the gallery, cached
72
tests on the voice → terminal bridge
3
systems live
Client
In-house products and client demos
Period
2026
Stack
Gemini Live API (native audio) · Cloudflare Workers + KV · Node (zero dependencies) · WhatsApp

What exists

  • Clara — a WhatsApp-style chat with text and voice, also connected to a real WhatsApp number.
  • JARVIS — the same engine with a HUD interface, camera (“eyes”), driving mode, and tools the agent calls on its own (change protocol, send email).
  • Voice bridge — you speak, a code agent runs on the server and replies with a spoken summary and on-screen text. Persisted queue, per-conversation session, cost-based model routing, under a watchdog. Phase 1 is read-only, by design.

What was learned (and applies to any client)

  • The voice gallery is cached text-to-speech, never streaming — streaming doesn’t cache and blows through quota.
  • A mirrored camera image destroys the model’s text reading; unmirror before sending.
  • A voice agent never publishes anything on its own: it dispatches and notifies.

What it’s good for in your business

Voice-based support, document reading via camera, an internal assistant that queries systems — with quota and cost limits already mapped out.