Introduction

What is VectorDB?

VectorDB is a lightweight, self-hosted vector database built for startups and small-scale AI applications. It provides a REST API for storing, searching, and managing vector embeddings with metadata — without the cost or complexity of managed cloud solutions.

Get started quickly:

• Quickstart — Get running in under 5 minutes
• API Reference — Full REST API reference
• Python SDK — Python client with sync and async support
• TypeScript SDK — TypeScript/Node.js client

Why VectorDB?

	VectorDB	Pinecone	Weaviate
Self-hosted	✅	❌	✅
Free tier	✅ Unlimited	Limited	Limited
Setup time	< 5 min	Managed	Complex
Python SDK	✅	✅	✅
TypeScript SDK	✅	✅	✅
CLI	✅	❌	❌
Hybrid search	✅	✅	✅

Key Features

• Collections — Namespace your vectors. Each collection has its own dimensionality and distance metric.
• HNSW indexing — Approximate nearest neighbor search using HNSWlib. Fast at any scale.
• Multiple distance metrics — Cosine similarity, L2 (Euclidean), and inner product.
• Metadata filtering — Attach JSON metadata to every vector and filter search results.
• Hybrid search — Combine vector similarity with keyword search using Reciprocal Rank Fusion (RRF).
• Recommendations — Find vectors similar to any stored vector.
• Reranking — Re-score a candidate set against a query vector.
• Multi-key auth — Role-based API keys (admin, readwrite, readonly).
• Redis caching — Optional caching layer for high-throughput search workloads.
• PostgreSQL backend — Swap SQLite for PostgreSQL + pgvector for production scale.
• OpenTelemetry — Distributed tracing, Prometheus metrics, structured logging.

Architecture

VectorDB is a single FastAPI service backed by SQLite (default) or PostgreSQL. Vector indexes are maintained in-memory using HNSWlib and persisted to disk on shutdown.

Client → REST API (FastAPI)
              ↓
         Auth & Rate Limiting
              ↓
         Vector Backend (SQLite+HNSW or PostgreSQL+pgvector)
              ↓
         Optional: Redis Cache