Conway Research unveils Underdog, a fully local AI assistant
Conway Research, a US AI startup founded about eight months ago, has shown Underdog, its first personal AI assistant, which runs entirely on the user's device and can work offline; it is still invite-only, with beta builds for Mac and Windows. The company said the project is backed by a16z, the Anthropic-Menlo Ventures investment group Anthology, and individual investors from Google, OpenAI, Hugging Face and elsewhere, without disclosing the amount. According to 智东西, Underdog uses a Saluki 27B model fine-tuned from Qwen3.8 27B plus a proprietary Husky inference engine, and Conway claims Woof 4B handles up to 730 tokens per second on a MacBook, 4.5x faster than Apple's MLX — figures that come from the company itself.
It represents a local, offline approach to personal agents that differs from cloud-virtual-machine designs such as Meta's Muse and OpenAI's Dot, though the report says its performance does not yet exceed those products and the key numbers are not independently verified.
智东西 作者 | 程茜 编辑 | 李水青
智东西10月19日报道，近日，一家刚成立8个月的美国AI Agent创企Conway Research，亮出首款个人AI助手Underdog，同时官宣获得Anthropic旗下投资联合体、硅谷知名风投公司a16z，以及来自谷歌、OpenAI、Hugging Face的AI大牛等个人及机构的投资。
▲Conway Research创始人Sigil Wen官宣融资的推文（图源：X）
该公司将Underdog称作“迄今为止隐私性最强的Muse竞品”。
#Conway Research#Underdog#Sigil Wen#a16z#Anthropic#Menlo Ventures#Qwen#Stripe#Meta#Muse