Every context-aware AI assistant needs a vector store. Most reach for cloud-hosted solutions because vector search at scale is genuinely hard. But at Maple, cloud-hosted is a non-starter. Our privacy architecture demands that every piece of user data, including every embedding vector, stays on the user's local machine.
"The best vector store for a local-first application is the one that doesn't require explaining to the user why their laptop fan just turned on."
We built a vector store that is fast enough for real-time retrieval, small enough to run alongside a full desktop workload, and simple enough to ship as a single embedded library with zero external dependencies using scalar-quantized HNSW and memory-mapped pages.