infiniflow/infinity is an AI-native database designed for LLM applications, and its latest release is v0.7.4. It provides hybrid search that combines dense vector, sparse vector, tensor, and full-text search, together with result re-ranking. Tensor search is a way of splitting a single item into several vectors for comparison, while full-text search finds documents by keyword. However, the v0.7.4 changelog is not included in this research material, so what changed in this version cannot be confirmed.
Retrieval-augmented generation (RAG) is a way of combining the evidence gathered through search with the model's response, and Infinity can be applied to RAG applications such as search, recommendation, question answering, conversational AI, copilot, and content generation. It can add filtering on top of hybrid search, and it provides multiple data types and an intuitive Python API. Because it is a single dependency-free binary, it can be deployed with Docker or as the binary itself; the implementation language is C++ and the license is Apache-2.0. The repository has 4740 GitHub stars and was created on July 18, 2022.
That said, it is hard to confirm how far this repository's performance has been validated. This material contains no performance benchmark figures and no comparison results against other databases; of the evaluation score items, only Momentum's 0.0 is filled in and the rest are empty. It is not possible to judge what level it reaches in read latency or write throughput. At this stage, rather than claiming the upper hand, it is better to treat it as a candidate for attaching an embedding store to a RAG prototype.