Limburgite
我们不追求面面俱到。同类产品里，大多数人天天在用的，其实只是其中五六成的功能——我们把这一部分做好、做到能在你自己的设备上运行，然后用极低的价格交给你。现在，是免费。 We don't try to do everything. Most people use maybe half of what an app offers, day in and day out. We build that half, make it run on your own device, and offer it at a very low price. Right now, that price is zero.
示意图，按同类产品的功能与价格对比。Illustration, relative to comparable apps.
本机推理On-device inference
基于 llama.cpp 的本机推理：有显卡自动用显卡（Vulkan，独显、核显都行），没有就用处理器；同一套做法，在电脑上跑得快，在 8 GB 内存的手机上也能跑。On-device inference built on llama.cpp: it uses your GPU when there is one (Vulkan, dedicated or integrated) and falls back to the CPU when there isn't. The same approach runs fast on PCs and still runs on phones with 8 GB of memory.
本机为主 · 云端可选On-device first · Cloud optional
响岩和月晤默认都在本机完成。想要更强的模型，可以接入你自己选的云端服务；只有你开启后才会联网，随时可以关掉。Phonolite and Lyrune work entirely on-device by default. Want a stronger model? Connect a cloud service you choose; nothing goes online until you turn it on, and you can turn it off any time.
