<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>KAON Group · LLMobile.news</title><link>https://llmobile.news/companies/kaon-group/</link><description>A concise news ticker covering AI on mobile devices, local models, apps and hardware.</description><language>en-GB</language><atom:link href="https://llmobile.news/companies/kaon-group/index.xml" rel="self" type="application/rss+xml"/><item><title>ENERZAi runs smart home agents and traffic classification on Broadcom NPUs in CPE</title><link>https://llmobile.news/ticker/enerzai-broadcom-cpe-demos/</link><guid isPermaLink="true">https://llmobile.news/ticker/enerzai-broadcom-cpe-demos/</guid><pubDate>Sun, 01 Nov 2026 13:12:00 +0100</pubDate><description>ENERZAi, an edge AI software company based in Seoul, is widening its work with Broadcom across customer-premises equipment and showed two demos running on Broadcom NPUs at Network X in Vienna, the company said in a press release. Both ran at the booths of the CPE manufacturers Sagemcom and KAON Group, on commercial devices built on Broadcom chips. ENERZAi says its on-device AI is already deployed on more than 3 million smart home devices, including set-top boxes.
The Smart Home AI Agent, shown at the Sagemcom booth, combines speech recognition, a small language model for understanding intent and generating responses, and speech synthesis in a single pipeline that runs on the device. Users control appliances and check their status by voice, and according to the company the agent carries context from earlier in the conversation and answers a vague request such as &amp;amp;ldquo;I&amp;amp;rsquo;m cold&amp;amp;rdquo; by proposing an adjustment. The second demo ran on a Wi-Fi router at the KAON booth and identifies what each connected device uses the network for, so operators can give priority bandwidth to critical services. ENERZAi says that classification previously ran in the cloud and now runs directly on the Broadcom NPU from minimal data.
Underneath both demos sits the company&amp;amp;rsquo;s own quantization and compiler stack. Its website describes the models as built on quantization down to 1.58 bits, with Optimium as its AI inference optimization engine for deployment across hardware platforms. At that format each weight is stored as one of three values, which cuts memory use at the cost of precision. Beyond Broadcom, ENERZAi says it contributes to Torq, the open AI compiler in the Synaptics edge AI software stack, which lets it run its models on Synaptics Astra parts including the new SR200 series, whose NPU Synaptics developed with Google Research.
The Korean deployments run on LG Uplus set-top boxes, where ENERZAi supplies voice-command AI. Outside Korea the company says it sells on-device models to CPE manufacturers and is in commercialization talks with global operators that have completed proofs of concept. Hanhim Chang, CEO of ENERZAi, named what the company wants next from its chip partnerships.
The next step is turning our technical work with Broadcom, Synaptics, and other semiconductor partners into deployments in real customer products.
Source: https://www.einpresswire.com/article/952529566/enerzai-expands-broadcom-partnership-with-on-device-ai-demos-at-network-x-2026
Read the article: https://llmobile.news/ticker/enerzai-broadcom-cpe-demos/</description><category>Demos</category><category>NPU</category><category>Speech</category><category>Agents</category></item></channel></rss>