Qualcomm is teaming up with AI startup Memories.ai to bring lightning-fast, on-device AI to something most of us have mountains of — photos and videos. The goal is to make it possible for users to search and interact with personal media using natural language, all while keeping the data stored and processed locally rather than in the cloud. The collaboration centers on Memories.ai’s upcoming Large Visual Memory Models 2.0 (LVMM 2.0), which are set to debut in 2026. At that point, both companies plan to pitch this new AI capability to device makers developing apps for smartphones, headsets, and PCs.
In concept, this could evolve into experiences like a Samsung Gallery app powered by Memories.ai, where users could search through photos and videos intuitively, simply asking questions like “When was I in Rome with friends?” or “Show me the dinner in Korea.” According to Memories.ai co-founder and CEO Shawn Shen, the system is inspired by how humans associate visuals with deeper memories — a single photo of a meal can trigger recollections of the entire experience surrounding it. Machines, he noted, are currently far better at linking words than they are at connecting visual information, and this is exactly what Memories.ai aims to change.
The company’s tech consists of two key elements: an encoder, which analyzes and compresses media into an information-rich format, and a search infrastructure, which allows AI to query that data intelligently. Rather than powering the image itself, the system captures its semantic meaning — what’s happening, who’s there, and where — creating dynamic tags on the fly. This means searches like “my friends at the concert last summer” or “the time my dog knocked over the vase” would instantly pull up the right clips.
Looking ahead, the technology could power constant visual encoding on compatible devices — such as XR headsets or home security cameras — allowing real-time event recognition and natural language interaction. Shen even described scenarios where users could simply ask their cameras, “What suspicious events happened around my house?” or “When was the pizza delivered last week?” Qualcomm’s involvement hints at the possibility of this technology running efficiently on Snapdragon-powered platforms, bringing personalized, private AI memory recall directly to everyday devices.
