v0.106.0

Features

• Know when recording really starts. The macOS overlay now shows an amber “Connecting…” indicator while capture starts, then a green waveform once audio arrives. If audio samples stop arriving, it switches to “Waiting for audio…”. Silence still counts as active recording. (#650)
• See elapsed recording time. A compact MM:SS timer sits beside the waveform. It starts with the first audio sample, excludes connection delay, and resets for each recording. (#650)

Fixes

• No more false success for empty transcriptions. Empty or whitespace-only results now produce a clear error with recovery guidance instead of “Transcription Finished” and an empty paste. The macOS check also covers older user-installed CLI versions. (#649)
• Keep recognized speech when cleanup fails. If LLM cleanup returns blank text, the CLI preserves the raw transcription on the clipboard and identifies cleanup as the failed step. Empty speech-recognition results leave the clipboard unchanged. (#649)

Maintenance

• Update Docker build actions and Codecov integration. (#645, #646, #647, #648)

Full changelog: v0.105.2...v0.106.0