All pull requests

Pull requests

• fix(voice): preserve tool outputs across concurrent realtime context updates

  #7905·Epsilon006 opened Oct 17, 2026·

  ·
  1

  1

• fix(voice): keep overlapping realtime chat context pushes from deleting each other's items

  #7904·Rehansanjay opened Oct 17, 2026Contributor·

  ·
  1

  2

• fix(voice): vectorize the speaking-rate spectral flux

  #7903·omerekin opened Oct 17, 2026·

  ·
  1

  1

• fix(ipc): unblock thread executor aclose() when initialize is cancelled

  #7899·BlueX888 opened Oct 17, 2026Contributor·

  ·
  5

• fix: reuse in-hand final transcript in commit_user_turn (manual turns)

  #7898·iamdgarcia opened Oct 17, 2026·

  ·
  1

  5

• fix(soniox): emit END_OF_SPEECH when an endpoint follows a fully retracted segment

  #7897·iamdgarcia opened Oct 17, 2026·

  ·
  1

• fix(google): report realtime usage metrics once per generation

  #7896·BlueX888 opened Oct 17, 2026Contributor·

  ·
  6

• fix(voice): resolve the aligned-transcript wait when TTS fails before first yield

  #7895·BlueX888 opened Oct 17, 2026Contributor·

  ·

• fix(llm): drop nonexistent name from __all__

  #7894·BlueX888 opened Oct 17, 2026Contributor·

  ·
  2

• fix(ipc): fail pending inference requests when the inference process dies

  #7893·BlueX888 opened Oct 17, 2026Contributor·

  ·
  3

• fix(cli): exit non-zero when the worker fails to start

  #7892·BlueX888 opened Oct 17, 2026Contributor·

  ·
  3

• fix(google): count Gemini thinking tokens in LLM completion_tokens

  #7891·ninglinLiu opened Oct 17, 2026·

  ·
  1

  1

• fix(meta): don't delay Muse STT audio by the websocket connect time

  #7890·DEVJHAWAR11 opened Oct 17, 2026·

  ·
  1

  7

• fix(anthropic): don't add message_delta output tokens to the message_start count

  #7888·pcbeingused333 opened Oct 17, 2026·

  ·
  1

• fix(inference): emit every parallel tool call sent in one chunk

  #7887·pcbeingused333 opened Oct 17, 2026·

  ·
  1

• fix(boson-avatar): follow the Boson API's paths and object names

  #7886·warlockee opened Oct 17, 2026Contributor·

  ·
  3

• feat(azure): use the Azure Speech SDK for TTS

  #7884·tinalenguyen opened Oct 16, 2026Member·

  ·
  3

• feat(microsoft-ai): add DecisionModel for Microsoft-Decision-1

  #7883·tinalenguyen opened Oct 16, 2026Member·

  ·
  2

• feat(openai): add DecisionModel for OpenAI's Decisions API

  #7882·tinalenguyen opened Oct 16, 2026Member·

  ·
  2

• fix(voice): place expressive instructions after the agent instructions

  #7881·tinalenguyen opened Oct 16, 2026Member·

  ·
  2

• fix(stt): preserve the utterance when batch recognition falls back

  #7879·antonsoo opened Oct 16, 2026·

  ·
  1

• fix(bey): use the current session endpoint and send 16 kHz audio

  #7877·niqodea opened Oct 16, 2026Contributor·

  ·

• fix(mcp): detect a dead server connection instead of reporting it as initialized

  #7875·levalencia opened Oct 16, 2026·

  ·
  3

• feat(gradium): default the TTS voice to Everly (2/2)

  #7874·hhugo-gradium opened Oct 16, 2026Contributor·

  ·

• fix(gradium): STT cumulative interims, vad_bucket=0, connect timeout, websocket hardening (2/2)

  #7871·hhugo-gradium opened Oct 16, 2026Contributor·

  ·