# PomaiDB

A lightweight, embeddable vector database and multimodal AI memory engine built for edge devices such as Raspberry Pi and NVIDIA Jetson, combining vector search, graph relationships, and offline RAG in one process.

PomaiDB is an embeddable, single-process vector database and multimodal AI memory engine designed to run directly inside an application on resource-constrained edge hardware, including Raspberry Pi, Orange Pi, NVIDIA Jetson, NAS devices, gateways, and cameras, rather than as a cloud-first distributed service. It combines vector search, semantic retrieval, offline RAG, graph relationships, mesh and time-series data, and key-value storage into one native, C++20 database.

Design priorities are predictable latency, low memory use, sequential I/O for flash-storage longevity, deterministic execution, offline operation, native ARM64 support, and resource management aimed at avoiding out-of-memory failures. Vector search supports approximate nearest-neighbor methods including IVF and HNSW, batch and point queries, metadata filtering, and scalar quantization, making it suited to semantic search, recommendation, visual similarity, and local AI memory. It also includes an embedded offline-first RAG pipeline.

Website: https://pomaidb-web.vercel.app