# Mu Li outlines scaling the Higgs audio stack for real-time voice agents

> Original title: Voice-native AI needs more than a text serving stack.

- Source: SGLang
- Published: 2026-11-09T02:26:00.000Z
- Bad Influence: https://badinfluence.news/items/ywz4rwld7mytibwgys8gluuu7
- Original: https://x.com/sgl_project/status/2119617469707100247

## Summary

Mu Li of Boson AI discusses scaling the Higgs audio stack for real-time voice agents. The talk covers continuous streaming, interruption handling and low-latency serving. He argues that voice-native AI needs more than a text serving stack.

## Post

Voice-native AI needs more than a text serving stack.

Mu Li from @boson_ai on scaling the Higgs audio stack: continuous streaming, interruption handling and low-latency serving for real-time voice agents.

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHWpg5m6aMAAiiIx.jpg&mode=thumb&exp=1794528000&sig=260f4f787d2860d4)