What is an IVF index and when should you use it?

Inverted File Index, or IVF, partitions vectors into clusters using k-means, then searches only the few clusters closest to the query. The two key parameters are nlist, the number of clusters, typically the square root of the dataset size, and nprobe, the number of clusters to search at query time. IVF uses less memory than HNSW and indexes faster on billion-scale datasets, but HNSW usually beats it on recall at equivalent latency. Use IVF when memory is tight or when you need fast batch indexing.