Voicebox

A voice-first workspace with a reusable live-model harness, shared tools and explicit execution boundaries.

Pluggable live models — Gemini 3.8 Live, Thinking and Flash; OpenAI Realtime; provider contract, audio rates and tool loop.
Architectural overview — browser, host and portable core.
Complete documentation guide — operations, security, tools and coding-agent delegation.
Source and quick start

Live model layers, audio rates and bounded tools

Claude’s live adapter is currently a placeholder; Claude coding-agent delegation is separate.