Collections and indexing

Automatic creation defines Anvia ID, logical document ID, serialized document, and float-vector fields. It creates HNSW with M: 16, efConstruction: 256, and the configured metric, then loads the collection.

Those values are a development baseline, not a universal production configuration. Provision collections and indexes through infrastructure when shard count, consistency, replicas, partitions, or build parameters matter, then call validate() at startup.

The configured dimensions must match the embedding model and existing field dimension. Metadata names beginning with __anvia_ are reserved.

Ingestion is a replacement upsert: upsert() deletes rows for each supplied document ID before inserting, so re-ingesting a document replaces it instead of duplicating it.