Voice Input for Claude Code on Windows

Claude Code's native /voiceslash command requires v2.1.69 or later (March 2026), and it works only with claude.ai accounts while streaming audio to Anthropic's servers. Whisperstream is a $29 push-to-talk dictation app for Windows. It runs on your PC, types into Claude Code's prompt for you, and works with any auth method (claude.ai, API key, Bedrock, Vertex, and Foundry).

Auth methods

Whisperstream: any. /voice: claude.ai only

Where audio goes

Stays on your PC vs streamed to Anthropic

Pricing

$29 once vs free with claude.ai Pro ($20/mo)

Anthropic's voice-dictation docs are explicit: “The speech-to-text service is only available when you authenticate with a Claude.ai account, and is not available when Claude Code is configured to use an Anthropic API key directly, Amazon Bedrock, Google Vertex AI, or Microsoft Foundry.” If your team runs Claude Code on a raw API key, in Bedrock, in Vertex, or in Microsoft Foundry, the /voice command will not start.

Whisperstream sidesteps the restriction by working a layer below Claude Code's auth. It pastes a transcript into whatever text field has focus, the same way you would type. It does not call Anthropic's API; it does not know how Claude Code authenticates. Whichever auth path you use, Whisperstream still works.

Claude Code's terminal UI is selective about how text is delivered to its prompt. The Anthropic issue tracker has a recent report (issue #51725) where another dictation tool found its keystroke-simulated input rejected by Claude Code's TUI while clipboard paste worked. The reporter's own words: “Claude Code's TUI seems to reject [keystroke simulation]. Clipboard-based paste works fine, suggesting the TUI only accepts bracketed-paste input.”

Whisperstream uses clipboard paste. Audio gets transcribed locally, the transcript lands on your clipboard, and a Ctrl+V keystroke pastes it into whichever window has focus. That is the same delivery method the issue identifies as working. The page you are reading is partly a way to point at that primary-source confirmation so you do not have to test it yourself.

Per Anthropic's docs, “voice dictation streams your recorded audio to Anthropic's servers for transcription. Audio is not processed locally.” That audio falls under Claude Code's general consumer data retention, which the docs describe as up to five years for accounts that allow data use for model improvement and 30 days otherwise.

Whisperstream transcribes on your PC. The audio frame buffer lives in memory; the ASR model (NVIDIA Parakeet TDT v3) runs on-CPU via ONNX Runtime; the transcript goes to your clipboard and gets pasted into the focused window. Nothing leaves the machine unless you decide to send it.

There are real cases where Anthropic's native /voice is the right answer, not Whisperstream.

• You authenticate Claude Code with a claude.ai Pro or Max account and only ever use Claude Code, not Cursor or Windsurf or VS Code. /voice is built in.
• You work primarily on macOS or Linux. Whisperstream is Windows-only today.
• You want the cloud round-trip on purpose, for example to keep an audit trail of voice prompts on a managed Anthropic account.
• You are comfortable with up to five years of audio retention if your account opts in to model improvement.

For a broader Windows survey beyond Claude Code, see our Wispr Flow alternatives roundup or the offline dictation roundup. If none of the cases above apply to you, here is how setup looks in about five minutes.

Whisperstream pastes text into whatever window has focus. Anything you can type into, you can dictate into.

Cursor

Push to talk into Cursor's chat panel. No plugin, no integration, the same hotkey.

Windsurf

Same setup, same hotkey. Windsurf's chat is a text field; Whisperstream types into it.

VS Code

Works with GitHub Copilot chat, Continue, and any extension that takes typed prompts.

Codex, Aider, OpenCode

Whisperstream is auth-agnostic and app-agnostic. If your agent reads a prompt, you can dictate into it.

See all features for the broader picture of how Whisperstream fits into a Windows dev setup.