Nozzle is a lightweight TypeScript library for handling LLM text streams, buffering only where necessary and emitting everything else immediately, such as holding a potential regex match until confirmed. It is zero-dependency with a small footprint and offers streaming conversion APIs like consume and toReadableStream, suited to low-latency streaming UIs and parsing pipelines.
Streaming Output
流式输出
Technology that returns results chunk-by-chunk during generation rather than waiting for completion
Curated Intelligence
Meta has launched real-time voice conversation and expressive avatar features for its personal AI assistant Muse, rolling out first to US users across mobile, web and Mac. Two models, Muse Realtime Voice and Muse Realtime Avatar, drive the experience, streaming 448×768 portrait video at 25 fps with roughly 870 ms latency to first audio-video response, plus interruptible turn-taking and swappable avatars and voice styles.
The Qwen team released Qwen3.5-LiveTranslate-Flash, a multimodal simultaneous interpretation model built on Qwen3.5-Omni. It expands language coverage to 60 languages (with voice output for 29), achieves an end-to-end latency of 2.8 seconds per word via new 'Readable Unit' streaming technology, and introduces real-time voice cloning and hotword enhancement for improved accuracy in professional scenarios.