kdbai-knowledge

Use when building vector search, RAG pipelines, hybrid search, time-series pattern matching, or managing tables in KDB.AI. Also use when asked about kdbai_client, similarity search, reranking, KDB.AI filters, or CAGRA GPU indexes.

Guide KDB.AI sizing, capacity planning, and server resource configuration by collecting dataset and usage requirements, recommending suitable vector indexes and starting configurations, and estimating RAM, GPU VRAM, and persisted disk. Use for KDB.AI hardware requirements, capacity planning, memory-fit questions, row/vector limits, sizing from a data-volume figure (size per day or total size) rather than a row count, index selection, whether and how to partition a table (partition-key selection, the partition_column, and how partition scope works), qHnsw mmap sizing, CAGRA GPU sizing, and CPU worker/thread configuration — including standalone NUM_WRK and THREADS worker/thread-splitting or core-count questions and partition-key mechanics questions that carry no rows/dims sizing hook. These CPU and partition topics belong here, not in the sibling kdbai (schema/search) skill. Do not invent throughput, latency, cost, IOPS, or distributed topology from dataset shape alone.