Repository Issue Activity (beta)

#414 MLX tool-call follow-up fails when assistant tool message omits content

#412 Whisper-family STT handlers ignore `mode`, causing premature turn finalization mid-utterance in `--mode realtime`

#444 The model has finished downloading,Does it support full offline operation?

#438 Harden Realtime response output ordering and terminal lifecycle

#447 Keep Realtime audio delta content_index stable across chunks

#439 Align Realtime transcript events and client compatibility

#436 Introduce a backend registry for STT, LLM, and TTS construction

#440 Add lifecycle support for process-shared backend resources

#296 Feature Request: AMD GPU support via ROCm/Vulkan for STT/TTS models

#308 Refactor speculative turn tracking around global conversation order

#423 faster-whisper backend never reports a language, so --enable_lang_prompt and per-turn TTS voice switching cannot work

#386 Consider relaxing the mlx==0.31.1 pin — MLX gained multithreaded support in 0.31.2

#433 Reduce barge-in latency with reversible pre-confirmation audio ducking

#134 Does it handle interruption ?

#421 Use Realtime as the only pipeline engine

#431 Expired OAuth token leaves Space signed in and silently stops the orb

#404 whisper-mlx: auto mode silently drops the user's turn when the detected language is outside SUPPORTED_LANGUAGES

#334 Add missing faster-qwen3-tts features for the ggml backend

#394 --enable_lang_prompt silently emits no instruction for 17 of the default STT's 25 languages

#385 LightningWhisperSTTHandler does not acquire the global MLX lock — SIGSEGV in mlx::core::metal::CommandEncoder (the second path missed by #377/#379)

#315 Idea: add short sentences when LLM responses are taking too long

#377 MLXAudioWhisperSTTHandler bypasses the global MLX lock → Metal assertion crash (SIGABRT)

#375 Whisper STT: language detection reads a text token, silently mistranscribing non-English speech (German → Italian)

#367 Add cancellation-aware OpenAI-compatible STT/TTS client backends

#363 Final governing design: shared STT/TTS inference coordination