swingby.ai/ Just Speak
Developer voice workflow
Talk through prompts, bug reports, code reviews, and documentation. Just Speak pastes the transcript at your active cursor.
Developer work contains a surprising amount of explanation. Dictation is strongest when the thought is longer than the command.
Describe a bug while the reproduction steps are still fresh. Explain a refactor to an AI coding agent. Draft a review comment without breaking concentration to type every clause.
Voice is not the fastest input for symbols, exact identifiers, or short shell commands. A useful developer workflow switches between speaking and typing instead of forcing one method everywhere.
This loop is especially useful with AI coding tools. A good agent prompt often needs more context than a developer wants to type: what changed, which files are in scope, what must stay unchanged, how to verify the result, and where uncertainty remains. Speaking that structure can make a complete prompt easier to produce.
Just Speak does not run commands, modify files, or decide what your coding agent should do. It turns speech into text at the cursor. That boundary keeps the product useful across tools and makes the final submit action visible to you.
In a terminal, compatibility depends on whether the current input accepts standard pasted text. Always inspect commands before running them. Dictation can mishear flags, paths, punctuation, and package names, so shell commands are usually better typed directly.
Local Whisper is the default. After the model is downloaded, speech recognition can work offline and the recording does not need to leave your Mac. That can be valuable when a prompt contains unreleased product details, internal code names, or incident context.
You can also explicitly select OpenAI or Groq and provide your own API key. In that mode, audio goes directly to the chosen provider under its terms, not through Swingby servers. Just Speak never silently moves a local session to cloud transcription.
Accuracy varies with the model, microphone, language, speaking style, and background noise. The most reliable way to evaluate it is to use the free trial with your actual technical vocabulary.
Use dictation for semantic density. Use the keyboard for syntactic precision.
| Task | Voice fit | Recommended approach |
|---|---|---|
| AI agent prompt | Strong | Speak context and constraints, then review before submitting |
| Bug report | Strong | Narrate the sequence while it is fresh, then add exact versions |
| Code review | Strong | Explain reasoning by voice, then verify names and line references |
| Documentation | Strong | Draft paragraphs by voice and edit for structure |
| Shell command | Weak | Type directly and inspect before execution |
| Source code | Mixed | Speak comments or intent, type exact syntax |
Yes. Just Speak inserts transcription into the active Mac text field, so it can work in editors, AI coding tools, issue trackers, documentation apps, and compatible terminal input fields.
No. Just Speak is a dictation layer. It transcribes what you say and pastes the result at your cursor. Your editor, terminal, or coding agent handles the next action.
Yes. Local Whisper is the default and works offline after the selected model has downloaded. Optional cloud providers are used only when you explicitly select one.
No speech recognizer is perfect with every library, identifier, and acronym. Test your own vocabulary during the trial and review exact technical tokens before you submit or execute anything.
Try Just Speak for 7 days or 50 dictations. No card or account required.