Infinity is a vector database built for LLM applications, providing fast hybrid search across dense vectors, sparse vectors, tensors, and full-text data.
The tool addresses the need for unified search across multiple data modalities in retrieval-augmented generation and other AI applications. It combines approximate nearest neighbor search with full-text search capabilities, supporting dense embeddings, sparse embeddings, multi-vector tensors, and BM25-based text retrieval in a single query engine. The architecture uses HNSW indexing for vector search and integrates reranking strategies including reciprocal rank fusion, weighted sum, and ColBERT-based reranking.
Infinity suits teams building RAG systems, semantic search, recommendation engines, or conversational AI that need to query multiple data types simultaneously. The tool emphasizes low latency and high throughput, claiming sub-millisecond query times on million-scale datasets and thousands of queries per second on full-text indexes. It offers a single-binary deployment model with no external dependencies, making it straightforward to deploy. The Python API is designed for AI developers, and the database can be embedded directly as a Python module. The README does not compare it to other vector databases.
The project maintains active development with regular commits across its C++ codebase and Python bindings. Documentation is comprehensive, covering benchmarks, API references, and deployment guides. The team engages with the community through Discord and maintains a public roadmap. Development appears focused on performance optimization and expanding search capabilities rather than rapid feature proliferation.
