OpenCode Voice Input¶
Status: Future project / nice-to-have
Goal¶
Enable voice input (speech-to-text) in the OpenCode CLI/TUI so I can dictate prompts instead of typing.
Options¶
1. First-party (not yet shipped in v1.17.9)¶
PR #11345 adds local Whisper transcription via @huggingface/transformers. Privacy-first, zero config, works offline. Still open as of May 2026.
- Enable in
/statusdialog - Press
\to record,\again to stop - Three model sizes: tiny (75MB), base (142MB, default), small (466MB)
2. Third-party plugin — renjfk/opencode-voice¶
Adds STT (local Whisper) + TTS (Piper). Available now for v1.17.x.
opencode plugin renjfk/opencode-voice
Requires sox and whisper-cli on system.
Features:
- /stt — record and transcribe
- /stt-submit — record, transcribe, auto-submit
- /tts-speak — read last response aloud
- /tts-mode — toggle auto TTS
- LLM normalises transcription (fixes homophones like "Jason" → "JSON")
- All processing can be local (Whisper + Piper)
3. Web UI¶
opencode web opens the browser interface which has had voice input support added in recent versions (browser-native SpeechRecognition API).
Notes¶
- Currently happy typing; revisit if RSI or typing fatigue becomes an issue
- Plugin approach is simplest path if/when wanted