v1.40.0

Added

• OpenAI realtime transcription models — gpt-transcribe (recommended: accurate, fast, inexpensive) and gpt-live-transcribe (continuous live deltas, best OSD previews). Conversational gpt-realtime-2.1 and gpt-realtime-2.1-mini are now selectable in hyprwhspr setup, which derives realtime_mode from the model. (#226, #228)
• realtime_conversation_history — controls whether converse mode keeps prior turns server-side.
• Live transcript in the pill mic OSD — mic_osd_pill_transcript_enabled, with configurable word limit and idle timeout. Now works with any realtime provider, not just ElevenLabs. (#217)
• Post-transcription hooks can consume a transcription — exit status 77 marks it handled and skips the paste. Any other non-zero exit still passes the original text through. (#224)

Changed

• realtime_conversation_history now defaults to "turn". Completed converse turns are deleted from the provider, so retained audio is no longer re-sent and re-billed on every turn — previously this could amplify audio-token cost several times over a session (#227). Set "session" if you want a multi-turn assistant that remembers context.
• Realtime model capabilities (turn detection, language hints, delay support, preview behavior) now come from the provider registry instead of hardcoded model lists.
• Gemini native-audio model id updated to gemini-2.5-flash-native-audio-preview-12-2025.
• inject_text returns a tri-state outcome instead of a bool, so callers can distinguish consumed from failed. (#225)

Fixed

• Realtime responses are correlated per request, so stale or cancelled turns no longer land in the wrong transcript.
• Realtime preview callback is reapplied after initialization.
• Python dependency install handles more existing-environment states.
• Install snippet selection and copying on the website. (#221)

Contributors

Thanks to @jhsu, @nwoolmer, @wojciechsacewicz, and @spencer-thompson.

New PRs

• feat: add live transcript to pill mic OSD by @wojciechsacewicz in #217
• fix: improve install snippet selection and copying by @nwoolmer in #221
• feat: post transcription hook scripts can consume transcription text by @jhsu in #224
• feat: return tri-state outcome from inject_text instead of bool by @goodroot in #225
• Add GPT Live Transcribe support by @spencer-thompson in #226
• Add GPT Transcribe support by @spencer-thompson in #228

New Contributors

• @nwoolmer made their first contribution in #221 ❤️
• @spencer-thompson made their first contribution in #226

Full Changelog: v1.39.0...v1.40.0