All pull requests

Pull requests

• Daily transport: send DTMF on SIP dial-in calls without an explicit session id

  #6165·jamsea opened Oct 14, 2026Contributor·

  ·
  4

  

• Check backchannels and mid-thought pauses in scripted evals

  #6164·aconchillo opened Oct 14, 2026Contributor·

  ·
  2

• fish: support the drama-3 model

  #6163·cshape opened Oct 14, 2026Contributor·

  ·
  3

• Resample Whisper STT input to 16 kHz

  #6161·ANDROIDHASSAN opened Oct 13, 2026·

  ·

• Import WhisperSTTService on Apple Silicon without mlx-whisper

  #6160·ANDROIDHASSAN opened Oct 13, 2026·

  ·

• Run an LLMWithBackend backend at its provider's default reasoning

  #6159·markbackman opened Oct 13, 2026Contributor·

  ·
  6

• fix(turns): keep frames queued behind a turn-opening transcript

  #6155·OlliTapio opened Oct 13, 2026Contributor·

  ·
  2

• feat(smallest): add itn_normalize and no_format_delay to SmallestSTTService

  #6154·harshitajain165 opened Oct 13, 2026Contributor·2 tasks·

  ·
  1

• feat: add Sarvam SageMaker speech services

  #6153·wirjo opened Oct 13, 2026Contributor·

  ·
  2

• Skip the FastAPI websocket close when the app side is already disconnected

  #6152·omer-insait opened Oct 13, 2026·

  ·

• Let the TTS service hold a speculative response

  #6151·aconchillo opened Oct 13, 2026Contributor·

  ·
  6

• fix: keep user idle detection suppressed while tools are pending

  #6150·Luke-salinero opened Oct 12, 2026·

  ·
  1

• fix(inworld): create TTS context with the first sentence

  #6149·ianbbqzy opened Oct 12, 2026Contributor·

  ·
  1

• Keep the LLMWithBackend frontend from describing a stop as a handoff

  #6148·kompfner opened Oct 12, 2026Contributor·

  ·
  2

• Close the stt span when a transcript is pushed upstream (Gemini Live, OpenAI Realtime)

  #6121·jamsea opened Oct 12, 2026Contributor·

  ·
  7

  

• Play NVIDIA SageMaker TTS audio through its audio context

  #6120·aconchillo opened Oct 12, 2026Contributor·

  ·
  1

• End Deepgram SageMaker TTS audio contexts on Flushed

  #6119·aconchillo opened Oct 12, 2026Contributor·

  ·
  1

• Simplify TTSService and its audio context handling

  #6118·aconchillo opened Oct 12, 2026Contributor·

  ·
  3

• fix(mcp): deliver a successful call's filter output as-is

  #6116·BlueX888 opened Oct 11, 2026·

  ·
  1

• Keep a tool result placeholder as a thinking reply saw it

  #6114·souravendra opened Oct 11, 2026Contributor·

  ·
  5

• Leave out Anthropic thinking blocks the conversation no longer matches

  #6113·souravendra opened Oct 11, 2026Contributor·

  ·
  1

  2

• feat(serializers): add a jambonz serializer for the listen verb

  #6112·Rehansanjay opened Oct 11, 2026Contributor·

  ·
  1

• buffer audio during receive-loop reconnects

  #6109·pradneshfernandez opened Oct 10, 2026Contributor·

  ·
  1

• Send each LLM model the lowest reasoning setting its provider documents

  #6108·markbackman opened Oct 10, 2026Contributor·

  ·
  1

  19

• Add support for model parameter for Azure Speech Services (needed to use mai-transcribe-2-streaming model)

  #6105·seanrobbins opened Oct 10, 2026·

  ·