Write the PR description by talking.
What GitHub offers for voice today
We checked GitHub's and Visual Studio Code's documentation on October 18, 2026.
- On github.com: the GitHub docs we checked describe Markdown editing, saved replies and Copilot features for issues and pull requests, but no microphone or dictation control in those text boxes. Your operating system's dictation or a desktop app like Whispek fills that gap.
- Copilot CLI: GitHub's docs say Copilot CLI has voice input. Run /voice to download a local speech model, then hold Space or press Ctrl+X followed by V to dictate. Transcription runs on your machine and the text lands in the prompt for you to edit. GitHub lists Windows, Linux x64 and Apple silicon Macs, not Intel Macs. If you only dictate Copilot CLI prompts, start there.
- GitHub Copilot app: GitHub's docs describe voice dictation in agent sessions, set up in the app's Voice dictation settings with a local transcription model.
- VS Code: Microsoft's VS Code docs describe experimental built-in dictation in chat, editors and terminals, and say editor dictation also works in pull request comment fields inside VS Code.
- Copilot pull request summaries: GitHub's docs say Copilot can draft a pull request summary in the description field, on paid Copilot plans. It summarizes the changes; it does not know the reasons and trade-offs you can say out loud.
When a separate dictation app makes sense
- You write most of your GitHub text in the browser. Issue bodies, pull request descriptions, review comments and discussion replies are ordinary text boxes, and the same Whispek shortcut works in all of them.
- You want one shortcut across GitHub, your editor, AI chats, Slack and email, instead of a different voice feature in each tool.
- You want to read the text before anyone else does. Nothing is posted until you click Comment, Create pull request or Submit review.
- You want names to come out right. Add repo, service and library names to Dictionary once and they're in your dictionary on every computer you sign in to, so Whispek writes them your way.
GitHub's built-in voice options vs Whispek
| GitHub and VS Code, per their docs | Whispek | |
|---|---|---|
| Issue and PR text boxes on github.com | No dictation control described in the docs we checked | Dictate, then paste or insert into the text box |
| Copilot CLI prompt | /voice with a local model; hold Space or Ctrl+X then V | Dictate into a notes field or chat box, then copy and paste into the terminal |
| Pull request comments in VS Code | Experimental built-in editor dictation, per Microsoft's docs | Dictate, then paste into the comment field |
| Platforms | Copilot CLI voice: Windows, Linux x64, Apple silicon Macs | Windows, macOS (beta), Linux (beta) |
| Rewrite and reuse | Saved replies on GitHub for reusable responses | Transforms, Snippets, History and Dictionary |
How to dictate a pull request description with Whispek
- Click into the description box on the Open a pull request page, or into any issue or comment box on github.com.
- Press your Voice input shortcut and explain the change. Right Alt on Windows, Right Option on macOS, Left Ctrl + Left Alt + Space on Linux; change the Voice input shortcut on the Home screen. Turn on Settings > Voice & translation > Live transcription display to see the transcript above the capsule while you talk.
- Tidy it if needed. Select the text and run a transform, for example a New transform whose Instructions turn spoken notes into Summary, Changes and Testing sections. Choose Original in Style > Cleanup when you want your exact words.
- Add what speech is bad at, then post. Type @mentions, issue numbers, code and links yourself, check the Preview tab, then click Create pull request. If the text did not appear in the box, copy it from History and paste it.
Three things worth saying out loud on GitHub
- Pull request description: "This moves retry handling out of the webhook controller into a small queue worker. The controller was timing out under load. I kept the public API the same. To test it, replay the fixture events and check that each one is processed once."
- Bug report: "Steps: sign in, open Settings, change the email address and save. Expected: the new address is shown. Actual: the page reloads with the old one. It happens in Firefox and Chrome, not in the desktop app."
- Review comment: "This works, but the loop makes a database call for every item. Could we load them in one query before the loop? Not blocking if you'd rather do it in a follow-up."
Say the reasoning; paste stack traces, file paths and code yourself.
Using gh or Copilot CLI in a terminal?
For gh commands, or for Copilot CLI if you don't use its /voice, dictate into a notes field or chat box, review it, then copy and paste the text into the terminal. Do not dictate shell commands to run directly. For agent prompts in a terminal, see voice dictation for Claude Code and voice dictation for Codex.
Keep names right and reuse what works
- Add repository, service and library names to Dictionary. To add many at once, use Dictionary > More actions > Import terms.
- Save text you repeat as Snippets, such as a testing checklist or a sign-off line. GitHub's own saved replies work well for full canned responses.
- Find an earlier description or comment in History and copy it again.
- Create a New transform with your own Instructions, for example to turn a spoken bug report into Steps, Expected and Actual.
What Whispek does not do on GitHub
- It is not a GitHub product, app or browser add-on.
- It does not read your code or diffs, open pull requests, post comments or submit reviews.
- It does not add Markdown, @mentions or issue links for you unless you ask a transform to format the text; check the Preview tab before you post.
Pricing
Whispek is free every 7 days. Pro is unlimited for $8.25/month, billed yearly ($99), or $12 month to month. Prices exclude tax.