Show HN: VectorLiteDB – a vector DB for local dev, like SQLite but for vectors (github.com)

VectorLiteDB is a new open-source, single-file vector database for local development — think "SQLite for embeddings." Released as a small Python package (pip install vectorlitedb), it stores embeddings + metadata in one file and exposes a simple CRUD/search API (db = VectorLiteDB("file.db", dimension=384); db.insert(...); results = db.search(query, top_k=5)). It supports cosine/l2/dot distance metrics, metadata filtering, returns similarity scores and metadata, persists automatically to disk, and works fully offline. The project is v0.1.0 (Apache 2.0) and aims for simplicity and readability over performance. This is significant for AI/ML workflows that need fast iteration, privacy, or edge/offline capabilities: local RAG prototyping, personal assistant memory, Raspberry Pi semantic search, or trying ideas before switching to a managed service. Key technical trade-offs: VectorLiteDB uses brute-force search (roughly ~100ms for 10K vectors, not optimized beyond ~100K vectors), does not generate embeddings, lacks concurrent-write handling, and isn’t tuned for production speed or scale. The API is intentionally lightweight and swappable with cloud indices (e.g., Pinecone) so you can prototype locally and migrate to scalable vector services when needed.