opencode-voice v0.2.3

Fixed

• Repaired unfinalized WAV headers before transcription so recordings interrupted by forced recorder termination remain usable (#4).
• Hardened RIFF chunk parsing and added focused regression tests.

Thanks to @clintonwanner for reporting the issue and contributing the fix in #4.

opencode-voice v0.2.2

OpenCode 1.18 compatibility

• Restored the TUI-only plugin export required by OpenCode 1.18.
• Includes the v0.2.0 Handy catalog and GigaAM models.

opencode-voice v0.2.1

OpenCode 1.18 compatibility

• Added the required no-op server export so the TUI voice plugin can load in OpenCode 1.18 and later.
• Includes all v0.2.0 managed transcription engines and Handy catalog models, including GigaAM.

opencode-voice v0.2.0

Managed transcription engines

• Added the managed Rust transcribe-cpp sidecar and 68 verified GGUF ASR models from Handy's catalog, including GigaAM, Parakeet, Moonshine, Canary, SenseVoice, Qwen, Voxtral, and Granite.
• Added install, status, import, remove, and diagnostics commands for managed whisper.cpp and transcribe-cpp engines.
• Added opencode-voice update for updating an installed plugin.

Voice workflow improvements

• Reorganized settings around one recording key and grouped transcription and system settings.
• Hold-to-talk is shown as unavailable until OpenCode exposes reliable terminal key-release events to TUI plugins.
• Expanded Windows recorder diagnostics with the exact ffmpeg command and captured stderr.
• Improved Windows microphone fallback handling and expanded the built-in Whisper model picker.

See CHANGELOG.md for full details.

v0.1.9

Fix Windows managed whisper.cpp engine install by shipping a static MinGW-built whisper-cli.exe that does not require Visual C++ runtime DLLs. Also fixes managed engine downloadDir handling and stream-stall timeout diagnostics.

v0.1.8

Fixed

• Added managed Windows recorder install: opencode-voice can now download, unpack, cache, and probe ffmpeg.exe itself before recording.
• Extended doctor diagnostics to report the managed Windows recorder path, manifest, install error, and probe result.
• Made managed native binary replacement retry-safe on Windows for recorder and whisper-cli engine installs.

v0.1.7

Fixed\n- Hardened Windows recorder command resolution for ffmpeg (absolute path candidates, extension fallbacks, local module fallback) so recordings use the resolved executable reliably.\n- Enriched doctor diagnostics with exact recorder paths and per-command probe results for faster troubleshooting.

v0.1.6

Fixed\n- Fixed recorder command execution on Windows to use resolved absolute paths (including bundled ffmpeg path), preventing startup failures even when ffmpeg is not on PATH.\n- Windows microphone listing now uses the same resolved ffmpeg path for reliable device probing.

v0.1.5

Fixed

• Recovered missing ffmpeg-static on Windows by installing it locally when absent.
• Recorded command resolution now logs exact recorder command path in diagnostics.

opencode-voice v0.1.4

Added

• Added bundled ffmpeg fallback on Windows via ffmpeg-static so voice recording works without user-installed ffmpeg.
• Added DirectShow mic discovery/input handling for Windows.
• Kept managed engine + model download flow unchanged.

Testing

• npm run check
• npm pack --dry-run