Feature request: JARVIS-style voice-to-voice hands-free mode for Claude Code

Problem / use case

I run a field-service CRM project (HVAC dispatcher, 903+ clients, 2 crews). I spend a lot of my working time away from my desk — driving between sites, walking around, hands full. My laptop stays at home and I connect to Claude Code via Remote Desktop from my iPhone, wearing AirPods.

Right now the workflow on the phone is painful:

• I have to tap the mic on the iOS keyboard to dictate
• Then tap the send button
• Then read Claude's response by hand — or swipe twice with two fingers to trigger iOS "Speak Screen"
• Then tap mic again
• ...repeat

What I actually want is a continuous voice-to-voice conversation — the same experience ChatGPT Advanced Voice Mode / "JARVIS" gives you. I talk, pause, Claude Code hears me, reasons, answers out loud, I can interrupt, I answer back, no taps in between.

The closest thing today is the Claude mobile app Voice Mode, but that app doesn't see my project files / git / codebase — so it can't actually help me ship features. Claude Code can ship features but has no voice layer at all.

Proposed feature

A built-in Voice Mode in Claude Code:

• Toggleable per session: /voice on or a dedicated hotkey
• Uses a realtime STT → model → TTS pipeline (Anthropic-hosted, ideally using Claude with native audio I/O if/when available)
• Continuous listening with barge-in (interrupt the assistant mid-sentence)
• Streams Claude's response to TTS so the user hears it as it's generated, not after
• For tool calls and long file operations, speaks a short status ("reading calendar-settings.ts, running typecheck…") so the user knows work is happening
• Keeps the full text transcript visible in the terminal / web UI exactly as today — voice is additive, not replacing
• Works in the web session too (claude.ai/code/session_…) so remote-desktop / mobile use-cases work

Why this is specifically valuable in Claude Code (not just chat)

Developers and small-business operators like me aren't only coding. We're reasoning about real work while in motion — triage, planning, reviewing what an agent did, deciding what to build next. Today I have to choose: tap-typing a complex instruction on a phone screen, or losing the project context by switching to the Claude chat app. A voice layer closes that gap.

Languages

Please make sure non-English locales are first-class — in my case Russian. Both STT and TTS should respect system locale / a --voice-lang flag.

Not what I'm asking for

• Not "transcribe my dictation and paste it" — that exists everywhere
• Not "read the last message aloud on command" — iOS Speak Screen already does that
• What's missing is the continuous loop with barge-in

Alternatives considered

• iOS Shortcuts → Claude Code CLI over SSH: fragile, can't barge in, no streaming
• Custom pipeline with Whisper + Anthropic API + ElevenLabs: doable but out of scope for 99% of users, and loses the Claude Code tooling (git, file ops, hooks)
• ChatGPT Advanced Voice Mode: great voice UX, no access to my project / files / codebase

Would love to see this land. Happy to beta-test on Russian + a real CRM codebase.

Thanks!