Edge vector search means running similarity search on devices close to where the data is generated — phones, browsers, IoT hardware, or on-premises servers — rather than sending every query to a centralised cloud database. The goal is to keep retrieval and inference physically near the data.
The main motivation is latency and independence from the network. By performing search locally, an application avoids the round trip to a remote data centre, enabling real-time responses even on slow or intermittent connections. It also keeps sensitive data on the device, which helps with privacy and data-residency requirements since nothing needs to leave the local environment.
Edge vector search typically relies on lightweight or embedded vector databases optimised for small footprints and limited compute. The trade-off is scale: an edge device can hold and search far fewer vectors than a cloud cluster, so this approach suits focused, on-device knowledge — a personal assistant’s memory, a local product catalogue — rather than massive shared corpora.