Q&A With Mem.ai Co Founder Dennis Xu

Mem.ai co-founder Dennis Xu traces the company’s origin to a long-running frustration: knowledge workers generate massive amounts of information, yet existing tools make it hard to retrieve what matters when it’s needed. The core pitch is that “data ownership” and “information flow” aren’t served by dumping content into rigid structures like folders. Instead, Mem is built around a self-organizing workspace where users can capture notes freely and rely on semantic search, contextual understanding, and emergent organization to surface relevant material later—without forcing people to pre-plan how they’ll organize it.

Xu connects that thesis to his relationship with co-founder Kevin, whom he met in college. Their shared habit of building prototypes—ranging from an exploration of Wikipedia’s “shortest path” between pages to experiments in incentives and knowledge management—eventually narrowed into a single question: why does modern technology, despite producing more data than ever, still feel “useless” for everyday recall? That question sharpened into Mem’s founding focus in 2019, after years of experimenting with how people store and find information. Xu describes his personal failure mode as writing notes that later become unfindable—ending up in places like Apple Notes or email drafts—until the time comes to use them.

A key differentiator, Xu argues, is Mem’s rejection of folders as the dominant organizing mechanism. Folders, he says, are a digital carryover from physical filing systems—an interface metaphor that doesn’t match how people naturally think. When someone recalls “high school prom,” the mind doesn’t ask which folder it belongs in; it triggers associations. Mem’s approach aims to mirror that associative process by converting text into semantic vectors and then measuring “distance” between meanings. In practical terms, Mem can treat different sentences with related intent as close in meaning—so search becomes less keyword-driven and more intent-driven.

The conversation also clarifies how Mem handles tags and links. Tags are framed as a limited tool: users should only tag when they know exactly how they’ll want to retrieve something later (for example, saving an “employee contract” for future contract searches). The longer-term goal is an end state where users don’t need to tag at all, because the system will infer organization automatically through “emergent” topics. Bidirectional links are positioned as a way to connect ideas in the flow of writing—creating context that helps later recall and reduces writer’s block.

Collaboration came up as a practical challenge, especially around notifications and task granularity. Xu acknowledges that Mem notifications can be “all or none” at the workspace level, and that tasks are not yet a first-class collaboration feature. Still, the broader collaboration strategy is to avoid creating another silo: Mem supports imports and an API so teams can bring information in and have it immediately connected to the rest of a user’s knowledge graph.

Finally, Mem is described as free in beta for now, with plans for pro and teams tiers later. Xu also previews “similar mems,” intended to surface related notes automatically even without explicit links or tags—aimed at strengthening the “past-to-present” retrieval loop that Mem is designed to make effortless.