OpenCode voice input · Windows and Mac
Talk to OpenCode instead of typing every prompt
OpenCode's current public documentation describes a typed prompt in its terminal UI and does not document a built-in microphone-to-text action. A separate desktop dictation layer can put speech into the focused OpenCode prompt without changing OpenCode itself.
Current status, checked 26 August 2026: OpenCode's official TUI guide and keybind reference document text input but no speech-to-text binding. An open feature request asks for microphone input, while proposed voice implementations such as PR #11345 should not be treated as shipped merely because their code is public.
How the external-input approach works
VocalCode is not an OpenCode plugin. It is a push-to-talk desktop input utility. It records while you hold a configured key or mouse button, recognizes speech locally, then inserts the transcript into the field that still has focus.
- Start OpenCode in your normal terminal and place the cursor in its prompt.
- Hold your VocalCode talk binding and speak the task in ordinary prose.
- Release the binding. Review the text that appears in the OpenCode prompt.
- Submit with your keyboard, or configure VocalCode's separate send binding if that fits your workflow.
Because the two programs are separate, you can use the same talk binding in OpenCode, Claude Code, Cursor, a browser or a code-review field. Recognition audio and transcripts stay on your computer after the selected model has downloaded.
What this does not do
- It does not add a microphone button to OpenCode. The interaction remains an external global binding.
- It does not understand OpenCode commands or permissions. VocalCode only supplies text and, when configured, an Enter key.
- It does not read OpenCode's response aloud. This is speech-to-text input, not a voice conversation layer.
- It is not a Linux solution. The current builds support Windows 10/11 and macOS 11 or later on Apple silicon.
- Terminal compatibility is not universal. Higher-privilege processes, secure controls and terminals that reject synthetic input can refuse insertion. Test the exact terminal and privilege level you use.
Why focus safety matters in a terminal workflow
A speech tool should not finish a long transcription in the wrong window. VocalCode records the foreground target and focused UI element when recording starts, then checks them again before insertion. If you switch the focused field or foreground window while speaking, it refuses the write instead of guessing where the text belongs.
That safeguard means the intended workflow is simple: focus the OpenCode prompt, hold, speak and release without changing targets during the recording.
Local recognition and the remaining network use
The speech-recognition model runs on your own CPU, and VocalCode does not upload recognition audio or transcripts. Initial trial setup, model and update downloads, checkout, activation and periodic licence validation still use the internet. After model download, speech recognition itself works without a network connection.
The full-function trial lasts 30 days and does not ask for a payment card. The licence is $4.99 one time, plus applicable tax, and activates one device. No account is required.
Should you use this or wait for native OpenCode voice?
Use a native OpenCode feature when one ships if you want the microphone, prompt context and agent controls to be one integrated system. Use an external dictation layer when you want one talk binding across several coding tools, local recognition, or a solution that does not depend on OpenCode's release schedule.
This guide is written by VocalCode, so the recommendation is interested. The OpenCode status links above go to OpenCode's own documentation and repository; check them before deciding.