[FEATURE] Voice Mode in Claude Code (Especially in extensions like for VS Code)

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

The core problem: there is a gap between "thinking with Claude" and "building with Claude Code". You can't riff vocally with Claude Code like you can Claude which is super helpful, especially in planning mode.

Current workflow friction: There's a disconnect between planning/thinking with Claude and building/executing with Claude Code. These feel like two separate activities that don't flow into each other.

When working on complex, multi-component projects, I need to think out loud—explore options, get feedback, change direction, and iteratively refine an approach before committing to code. Right now, that requires typing everything out in detail, reading through lengthy responses, and manually translating conversational decisions into actionable prompts for Claude Code.

What's missing is a fluid, conversational collaboration mode where I can naturally discuss architecture and tradeoffs with Claude, receive real-time pushback and suggestions, make decisions together through dialogue, and then transition directly into implementation—without losing the context of what we discussed or re-prompting from scratch.

Proposed Solution

I want to be able to talk to Claude Code the way I'd talk to a co-founder or lead developer—naturally, conversationally—and have that dialogue directly drive implementation. (à la Claude)

The ideal flow:

• I describe what I'm trying to build in plain language (ideally via voice, not just typing)
• Claude Code asks clarifying questions and suggests approaches
• We go back and forth, refining the plan together
• When we're aligned, I say "build it" and Claude Code executes—carrying the full context of our discussion
• While it's working, I can course-correct in real time: "skip that part," "change the approach here," "actually let's do X instead"

This means:

• Voice input support in Claude Code or seamless bridging between voice chat and Code
• Shared context between the planning conversation and implementation—no re-prompting
• No hard wall between "talk mode" and "build mode"—it's one continuous session
• Claude remembers decisions across sessions so I don't repeat myself when I come back to a project

Basically: make the conversation the interface for development, not just a precursor to it.

Alternative Solutions

Current workarounds I'm using:

• Voice chat in Claude.ai → manually re-prompting in Claude Code: I plan in conversation, then switch tools and re-explain everything. Context is lost in the handoff.
• Multi-platform brainstorming: Using ChatGPT, Grok, and Gemini alongside Claude for different perspectives, but this fragments context further.
• Voice-to-text dictation into Claude Code: Faster input, but doesn't solve the core issue—each message is still treated as an isolated prompt, not part of a collaborative dialogue.

None of these solve the fundamental disconnect between planning conversations and code execution inside Claude Code. They all require me to be the bridge between thinking mode and building mode.

Priority

Medium - Would be very helpful

Feature Category

Interactive mode (TUI)

Use Case Example

Real scenario: I'm building a new company by acquiring an existing agency's clients, team, and tech. I need to stand up everything—billing, CRM, client migration, onboarding—from scratch.

Here's how a session would work:

1. I open Claude Code and say: "I need to build the client onboarding flow. We're migrating 3,400 clients. Most can be automated, but some need manual review."
2. Claude asks: "That's complex, let's break this down piece by piece. For starters, what triggers the manual review?" I explain. We decide on a plan for that point, we move on to the next. We've covered all the points and take a birds eye view together to consider the dependencies, blockers. When we finally figure out the full plan...
3. I say: "Okay, build it." Claude starts coding—using the full context of our conversation, not a cold prompt.
4. While it's working, I redirect: "Skip notifications for now, focus on the core flow." Claude adjusts without losing context.
5. Two days later, I come back: "Let's continue the onboarding module." Claude remembers our decisions and picks up where we left off.

Additional Context

_No response_