Batch indexing is the process of adding or updating many vectors in a single bulk operation, rather than inserting them one at a time. When you have a large corpus to load — say, embedding an entire document archive — batching the inserts is dramatically more efficient than issuing individual write requests.
The efficiency comes from amortising overhead. Building or updating an index structure like HNSW or IVF involves per-operation costs that are far cheaper when spread across thousands of vectors at once. Batch operations also allow the database to optimise disk writes, parallelise work, and defer expensive index maintenance until the whole batch is loaded.
Batch indexing contrasts with real-time indexing, where vectors become searchable immediately as they arrive. Many systems support both: a fast batch load for the initial corpus, then incremental real-time inserts for new data. The trade-off is freshness versus throughput — batch loads maximise ingestion speed, while real-time inserts minimise the delay before new data is queryable.