Tether Ships QVAC SDK v0.18.0 and v0.18.1 Within 13 Minutes Today, Adding Continuous Batching to On-Device Inference
tetherto/qvac published sdk-v0.18.0 at 10:14 UTC and sdk-v0.18.1 at 10:27 UTC on August 24, adding continuous batching so one loaded LLM serves several completions at once, plus VisionPsy Nano multimodal constants, Audio8 and CosyVoice3 TTS, Indic Conformer transcription and ACE-Step cover generation. It also removed dynamic toolsMode configuration, so tools now always prepend after the system message, which is a breaking change for anyone already integrating. The repo is only at 535 stars but this is Tether shipping a cross-platform local AI SDK (Linux, macOS, Windows, Android, iOS) with no cloud and no API keys, and the release cadence since v0.17.0 on August 10 is four tagged versions in two weeks.
Source
↳ Follow the thread