Disclosure: this article is published by Relay. Author Raymond Zhu is a founding member. It is documentation-based, not a hands-on test. Features were checked against official documentation, changelogs and marketplace listings on October 29, 2026. Developer tools change quickly, so check each tool’s docs for the current version. Sources: Claude Code voice dictation, Cursor: Prompting, Cursor 3.1 changelog, VS Code Speech, Relay for macOS.
Most coding tools now have their own voice input.
Talking to a coding agent is a natural fit for voice: prompts are long, full of context and rarely need perfect spelling until you reach a file name. The tools have noticed. Before you add another app, check what your editor or agent already offers.
The question isn’t whether voice input exists in your tools; it does. It’s whether you want one per tool, or one that works everywhere you type.
Claude Code: /voice in the terminal.
Claude Code’s voice dictation is turned on with the /voice command. The default is push-to-talk: hold Space to speak. /voice tap switches to tap to start; the second tap sends prompts of three words or more automatically. The key can be rebound.
Audio streams to Anthropic’s servers for transcription; it isn’t processed locally. It needs you to sign in with a Claude.ai account rather than an API key or a cloud provider, and a local microphone, so it doesn’t work in cloud or SSH sessions. It works on macOS, Linux and Windows, and in the VS Code extension. On a Mac, your terminal app needs microphone access under System Settings > Privacy & Security > Microphone.
Sources: Claude Code voice dictation.
Cursor and VS Code.
In Cursor, you can click the microphone icon in the chat input to dictate a prompt. Cursor 3.1 added hold Ctrl+M to speak in the Agents Window; it records the whole clip, then transcribes it.
For VS Code, Microsoft’s VS Code Speech extension provides speech-to-text processed locally, with no internet needed. It adds a microphone to GitHub Copilot Chat, where Cmd+I starts voice chat (hold to speak, release to submit), and lets you dictate into the editor with Cmd+Alt+V. It runs on macOS, Windows and Linux.
Sources: Cursor: Prompting, Cursor 3.1 changelog, VS Code Speech.
Dictation apps with developer features.
General dictation apps have added features for code. Wispr Flow lists variable recognition for camelCase and snake_case, and file tagging in Cursor and Windsurf. Aqua Voice offers MCP, a CLI and an API, with pages for using it with Claude Code and Codex. Superwhisper’s documentation lists integrations with coding agents, including Claude Code, Codex and Cursor.
None of these pages measure how well each tool handles your codebase, so try your own project’s names.
Where a system-wide layer like Relay fits.
A system-wide voice layer earns its place when you don’t stay in one tool: a prompt to an agent, a reply in a pull request, a message to a teammate, a command in the Terminal. Relay for macOS works in every app where you type with a keyboard, so it’s the same key and the same habit everywhere.
Relay also uses the app you’re in to shape the output. In the Terminal, it can produce the command you meant rather than prose describing it. Skills are on Mac only for now; Relay for Windows is in public Beta without them.
Voice input itself isn’t unique to Relay, and the built-in options above may be all you need if you live in one editor.
Sources: Relay for macOS.
Choose by where you spend the day.
If nearly all your prompting happens in one tool, start with that tool’s own voice input: Claude Code’s /voice in the terminal, Cursor’s microphone, or VS Code Speech if local processing matters to you. There’s nothing extra to install or pay for, and the voice input is designed for that tool’s interface.
If you move between an editor, a terminal, a browser, an issue tracker and chat all day, a single system-wide layer means one shortcut and one set of habits. Compare those in the best dictation apps for Mac, and try each with a real task from your own project before deciding.
Review every command before it runs.
Speech recognition and AI rewriting can both get a command subtly wrong: a path, a flag, a branch name. Read any dictated command before you press Return. In Claude Code’s /voice tap mode, the second tap sends prompts of three words or more without a separate review step, so use hold-to-talk if you want to read a prompt first, and be especially careful with anything that deletes files, rewrites history, deploys or touches credentials.
Keep secrets out of speech entirely. API keys and passwords don’t belong in a dictation tool, a prompt or a transcript, whatever the tool’s privacy policy says. How to dictate better AI prompts has more on structuring what you say to an agent.