Development experience, quickly implement prototypes with just a few lines of code
Generally speaking, if a database can realize the addition, deletion, modification and query of vectors, it meets the basic functions of a vector database. But is this enough?
In terms of queries alone, users will have queries about topk vectors, queries about vector distance within a certain range, and the need to filter data based on attributes. The calculation methods for vector distance include IP, Cosine and other calculations. requirements, and batch acquisition of data through iterators, etc. In addition, users have various demands for QPS, stability, index diversity, writing, etc. For example, for writing, there are demands for real-time writing, updating, batch import, and batch quick deletion.
Milvus has enhanced the API for the above scenarios in 2.3.x to meet users' rapid development needs in different scenarios. For example, with the upsert function, we can provide users with the insert or update function. When the user needs data coverage, only one line of code can meet the need, without the need for tedious steps such as querying, checking, and writing. Another example is the ScaNN index. We provide users with an index with more powerful performance outside of HNSW.
As a part of the infrastructure, Milvus has an indispensable ability to integrate data with other products. The community has open sourced Milvus/ES/Faiss to