Guides › How to Use Voice Correction in OpenCode
How to Use Voice Correction in OpenCode
Voice correction is the core workflow: the AI is speaking, you interrupt with your voice, the model catches what it was saying and pivots to address your correction.
1
Let the AI start responding
Ask a question and let the AI begin its response. TTS will start playing as text arrives.
2
Interrupt by speaking
The plugin detects your voice, stops TTS, and aborts the current generation. The model remembers what it was saying so far.
3
Speak your correction
Say your correction naturally: "Actually, make it recursive" or "Use a hash map instead." Your speech is captured by the mic monitor.
4
The model pivots
The next AI response includes both the partial content it was generating and your correction. It pivots seamlessly without needing context restatement.
Tips
• You can also use /ptt for a more deliberate interruption. /ptt gives you explicit control over when recording starts and stops.
• The correction window (Pro config: timingWindowMs) controls how long after interruption the correction is accepted. Default 5 seconds.
• The plugin injects both the partial AI content and your correction as system context — the model sees the full picture.
Troubleshooting
If corrections are not being applied, check that the plugin's chat.message handler is firing. Enable debug mode and look for "injected" messages in the logs.