康宏文 Henry

康宏文 Henry is the founder and CEO of Clipto AI and the guest in 为什么硅谷开始重新定义「AI 记忆」｜ S10E20. In the episode, he argues that cloud foundation models are strong at public world knowledge, but personal AI assistants need a separate memory layer for private files, long-term experience, and precise recall.

Henry’s core distinction is Data-to-Memory Transformation: data becomes memory only when it is understood, structured, searchable, reusable, and actionable. He also frames local-first memory as a systems challenge that pulled Clipto into model optimization, post-training, edge scheduling, chip/operator work, and end-user product design.

Source Position

• Henry treats Local-First Memory Layer as an independent platform opportunity rather than a small feature inside a single foundation model.
• He argues that Multimodal Personal Memory matters because users’ valuable context often lives in audio, video, images, meetings, and archives rather than only in text notes.
• He sees Mem0 and NotebookLM as validating nearby parts of the market, while distinguishing Clipto by local execution, multimodality, and personal-memory scope.
• His claims about Apple, Google, and other large platforms are best read as startup positioning: the source records the argument but does not independently verify that large companies cannot build similar memory products.

Connections

• Clipto AI — company/product Henry leads.
• Persistent Agent Memory, Context Engineering, and AI Data Memory Infrastructure — larger memory context for his argument.
• On-Device AI, Edge-Cloud AI Boundary, and On-Device Memory Scheduling — edge systems constraints he emphasizes.
• NotebookLM, Mem0, OpenAI, Apple, Google, and Nvidia — comparison and platform actors in the episode.