v1.4.0
This release adds a choice before your text is pasted — send the transcription through an LLM for polish first, then paste. The provider lineup expanded too, and custom endpoints can now carry extra parameters. Special thanks to vfasky for requesting multi-provider polish.
The transcribed sentence pauses first in the "Edit before paste" window, where you can paste it out as-is; this release adds a choice — press Polish ⌘B at the bottom of the window (the button in the red box in the screenshot) and the text gets run through an LLM to clean it up, removing verbal tics and fixing punctuation. It's a manual trigger — skip it and everything works exactly as before.
Provider lineup expanded: This release expands the available LLM providers to OpenAI / Anthropic / Gemini / Groq, plus a "Custom OpenAI-compatible endpoint" — any proxy, self-hosted vLLM / Ollama, or internal company gateway can be plugged in. Different users can pick whatever suits them: Claude writes nuanced Chinese, Groq is cheap and fast, Gemini has a long context window — switch based on the scenario.
Extra JSON for custom providers: The request body for a custom provider can now include an Extra JSON block, sending along whatever extra fields the endpoint needs (reasoning_effort, enable_thinking, top_k, …). Built-in templates grouped by vendor let you fill in common configs with one click, no digging through docs required.
Also fixed a few positioning issues on multi-monitor setups: the edit window opened with E after recording on a secondary display no longer jumps back to the main display; the overlay now re-corrects its position automatically when screens or Spaces change, with a dragged position now bound to a specific physical display — unplugging or swapping displays sends it back to the bottom-center of that display; and the preview window in settings can be dragged around without polluting the real position or disappearing on its own.