Vector search in C++ with CPU-specific builds and a branding clause

What it does Endee is a C++ vector database targeting AI search workloads like RAG, semantic search, and hybrid retrieval. It stores dense vectors, supports sparse vectors for hybrid queries, and applies payload filtering to narrow results. The project runs as an HTTP server and targets modern CPU instruction sets including AVX2, AVX512, NEON, and SVE2.

The interesting bit While most vector databases chase distributed scale, Endee bets on single-node performance, with repository metadata claiming capacity for up to one billion vectors per node. Less common for an Apache 2.0 project is its trademark clause: if you offer a hosted version, you must rebrand and cannot imply official affiliation.

Key highlights

• Targets single-node deployments with CPU-optimized builds for AVX2, AVX512, NEON, and SVE2
• Supports hybrid search combining dense vectors, sparse retrieval, and metadata filtering
• Includes operational features like backup APIs and runtime instrumentation
• Ships Docker images and local build paths for deployment flexibility
• Repository description claims billion-vector capacity on a single node

Caveats

• Benchmarks and a hosted cloud tier are marked “Coming Soon” in the README badges, so comparative performance data is not yet published
• The trademark policy explicitly prohibits using the “Endee” name for third-party hosted services, which may complicate commercial forks or managed offerings
• Client libraries and SDKs are not highlighted; current documentation focuses on direct HTTP API usage

Verdict Consider Endee if you want a self-hosted, CPU-tuned vector database for retrieval workloads and can tolerate an early tooling ecosystem. Pass if you require published benchmarks, managed cloud convenience, or unrestricted branding freedom.