v1.0.0

A macOS menu-bar tool: hold a hotkey to speak, let go to paste. Mixed Chinese/English/French/Japanese input, with 5 STT engines to choose from — Soniox, ElevenLabs, Volcengine, Groq, Apple Speech — pick whichever suits your scenario and network.

AI polish that's actually useful: strips verbal tics, fixes word choice, keeps the meaning intact. Terminal paste got its own path around CGEvent dropping keys during heavy rendering — Ghostty, iTerm2, Kitty and the like now go through the Accessibility API to trigger the Edit → Paste menu directly. Word corrections support custom term mappings, jieba-based Chinese word segmentation, and filler-word filtering. Soniox keeps the connection alive with a keepalive after finalizing, so the next recording starts with zero cold start. Bluetooth HFP auto-detects the A2DP transition state and waits for negotiation to finish before recording. Sparkle auto-updates are signed with EdDSA.

Later builds ironed out a few rough edges: bursty CoreAudio device-plug callbacks are now coalesced into a single handling pass, keeping logs clean; the pre-warm retry delay dropped from 0.5s to 0.2s, cutting total cold-start time by roughly 300ms; the hidden Soniox strict-language-mode toggle is back in the UI, with a one-time migration resetting anyone's leftover "strict language mode = on" back to off; stt-rt-v3 was removed from the Soniox model picker (the server already auto-routes to v4, so keeping it around just invited mis-picks); and the clipboard-restore delay dropped from up to 1.5s to 50ms, so pressing Cmd+V right after a transcription pastes your original clipboard content again immediately.