Voice Mode: dictate tasks out loud

Talk to Claude Code

Voice Mode turns your microphone into an input interface. Hold the spacebar, speak, release — your dictation becomes a prompt sent to Claude Code. This is useful for tasks that are long to describe, contexts where typing is uncomfortable, or simply to save time on complex requests.

Enable Voice Mode

# In an interactive Claude Code session, type:
/voice

Claude Code confirms voice mode activation and shows you how to use it.

Usage: push-to-talk

Voice Mode works as push-to-talk: hold the spacebar while speaking, then release to send the prompt.

Disable Voice Mode

# Disable in the current session
/voice

# Voice Mode also deactivates automatically
# at the end of each Claude Code session

Use cases

Describing complex tasks

Long, detailed prompts are often more natural to dictate than type. Voice lets you describe constraints, context, and requirements more fluidly:

[Voice dictation]
"Refactor the authentication service to support OAuth2
with Google and GitHub in addition to the current email login.
Maintain backwards compatibility with existing sessions,
add tests for each provider, and document the required
environment variables in the README."

Debugging out loud

Describing a bug while observing it is more natural than typing while searching:

"I have a memory leak on the dashboard page. The profiler
shows the ChartWidget component isn't unmounting properly.
Event listeners are probably not being cleaned up.
Analyze the component and fix it."

Code review sessions

Comment on code verbally during a review:

"Look at the functions in utils slash validation dot ts.
Add strict TypeScript types wherever annotations are missing,
and suggest unit tests for edge cases."

Supported languages

Voice Mode supports transcription in English and French natively, as well as most major European languages. Language detection is automatic.

Known limitations

Working around transcription errors

For highly technical terms:

# 1. Spell out critical names
"Create a variable named — spelling it out: A-U-T-H underscore T-O-K-E-N"

# 2. Combine voice and typing
# Dictate the general structure, then manually complete
# exact variable names in the terminal

# 3. Use descriptions
"The variable that stores the JWT authentication token"

Combining with other features

Voice Mode + Extended Thinking

Dictate complex architecture questions with Extended Thinking enabled for deep analysis without typing long prompts:

# Enable both
/voice    # Enable voice mode
Alt+T     # Enable Extended Thinking (Linux/Windows)
          # Option+T on macOS

# Then dictate your architecture question

Voice Mode + Plan Mode

Request a full plan verbally before starting an implementation:

[Dictation]
"Before coding anything, give me a detailed plan
for migrating our Express API to Fastify. I need to know
the migration order, risks at each step, and points of
no return. Wait for my approval before starting."

Next steps

• Cheatsheet: Quick reference — All slash commands at a glance
• Extended Thinking and Plan Mode — Combine voice with deep reasoning
• CLI: Full reference — All flags and subcommands
• settings.json: Configuration — Configure Claude Code to your needs