Voice coding with SayDart
Voice coding here means dictating prose and prompts, not using a voice-command programming language. Use SayDart to draft prompts for Cursor, Claude Code and Codex, along with commit messages, PR descriptions and code review comments in compatible text fields.
When coding by voice, dictate, then review, then send. SayDart does not execute dictated code or commands. For terminal agents, draft in a notes field first, check it, then paste it into the agent's prompt.
Why developers talk to coding agents instead of typing
- The useful part of a prompt is the context. "Fix the menu bug" is one line. The steps to reproduce, the edge case you ruled out and the behaviour to keep are what gets skipped when you type.
- Requirements are easier to say than to write. Talk through a feature as you would at a whiteboard, then tidy it before you send.
- Iteration adds up. Every "good, but keep the old API" is another message to type.
Where it fits: Cursor, VS Code chat, Codex and browser AI chats
SayDart is not an IDE plugin. One desktop shortcut covers an editor's AI chat panel, a browser coding assistant, a pull request description or a Slack thread.
In most chat boxes, the text lands right where your cursor is. Read it once and press Enter. If it ever doesn't show up, it's waiting in History.
For agents that run in a terminal, such as Claude Code or Codex in a shell, dictate into a notes field or chat box instead, review the text, then paste it into the agent's prompt. A terminal can treat unexpected text as a command.
Using Cursor most of the day? See voice dictation for Cursor.
Built-in voice (Claude Code /voice, Cursor Ctrl+M) vs a system-wide app
- Claude Code has a /voice command. Its documentation describes holding Space to record in the CLI and says it needs a Claude.ai account sign-in, not an Anthropic API key or a provider such as Amazon Bedrock. An optional tap mode sends the prompt automatically when you tap to stop, unless the transcript is very short.
- Cursor added voice mode in version 2.0, and its 3.1 changelog describes pressing and holding Ctrl+M to speak in the Agents Window.
If you only prompt one tool, its built-in voice may be enough. SayDart gives you one shortcut for your editor, browser chats, GitHub and Slack, with auto cleanup, Dictionary for your code terms, and Create an AI prompt to tidy a ramble.
Comparing tools? See the best dictation apps for vibe coding.
Using Claude Code? See voice dictation for Claude Code.
Using Codex CLI? See voice dictation for Codex.
Writing pull requests and issues? See voice dictation for GitHub.
Claude Code /voice not working? See the Claude Code voice mode troubleshooting guide.
A spoken structure for a coding request
Saying four things in order makes agent output easier to review:
- Context: "On the settings page, the account menu stays open after I click outside it."
- Goal: "Find the click-outside handler and make the smallest change that closes it."
- Constraints: "Keep the keyboard shortcut behaviour and don't touch the public API."
- Done when: "Explain which files you will change before editing, then add a test for this case."
Paste the file path and error, then send. For a five-part framework, read AI prompt dictation.
Typical developer workflows
- Dictate a coding-agent prompt. Click into Cursor's Agent box, or a notes field for Claude Code or Codex in a terminal. Hold Right Alt (Right Option on Mac, Left Ctrl + Left Alt + Space on Linux) and say the context, goal, constraints and done-when check. Filler words disappear. Review it, paste it if you drafted elsewhere, then press Enter yourself.
- Speak, then structure the request. Dictate a rough need, select the text, open the Transforms window and choose Create an AI prompt. Review the structured prompt before using it.
- Teach it your stack. Put it in Dictionary once, for example Kubernetes, PostgreSQL, Tauri, useEffect and pnpm, and they're in your dictionary on every computer you sign in to, so SayDart writes them your way.
- Reuse your rules with Snippets. Bind a short spoken trigger to a project-rule paragraph or PR template, such as "write tests the project way, don't touch the public API". Say the trigger and the exact paragraph drops in.
- Ask the SayDart Assistant about an error. Select the log, press Right Alt + Space (Right Option + Space on Mac, Left Ctrl + Left Alt + A on Linux) and ask, "What could cause this?" It answers in its own window; with web search on, answers include sources.
- Speak Chinese, get English. Press your Translation shortcut and say the prompt or PR description in your own language; English lands in the box. 30 languages in, 303 out: recognition can detect the language automatically, or you choose up to 4 of 30.
Keep identifiers right: Dictionary and pasted code
Add repo names, libraries and acronyms to Dictionary, so SayDart writes them your way. For paths, stack traces and shell commands, paste them; use your voice for the explanation. In Style > Cleanup, choose Original when you want your words as spoken.
Review before you send: Live transcription display, Transforms, History
- Watch the transcript as you speak: turn on Settings > Voice & translation > Live transcription display to show it above the capsule.
- Create an AI prompt: select a rambling request and turn it into a clean, structured prompt in one step. You can write your own transforms too.
- History: dictations stay on your computer with the original transcript and saved result versions, so you can reuse a prompt that worked.
What SayDart does not do
SayDart writes text. It does not tag IDE files, read your repository, edit code, run commands or send requests to a coding service.
Wispr Flow's help center says it offers file tagging in Cursor and Windsurf chat panels and variable recognition in Cursor, Windsurf and VS Code. If those matter, see SayDart vs Wispr Flow. Considering Typeless? See SayDart vs Typeless.
Platforms and pricing
Default voice shortcuts are Right Alt on Windows, Right Option on Mac and Left Ctrl + Left Alt + Space on Linux; change them on the Home screen. Speech is processed in the cloud, so check your team's rules before dictating private code details.
SayDart is free every 7 days. Pro is unlimited for $8.25/month, billed yearly ($99), or $12 month to month. Prices exclude tax.
Your next prompt, out loud
Open your agent's chat, hold your voice shortcut and say the task you've been putting off typing.