# Requirements and limitations
<a name="VectorSearch.Requirements"></a>

Keep the following requirements and limitations in mind when you work with vector indexes:
+ Vector indexes are supported only on tables that use on-demand capacity mode.
+ Vector embeddings are stored in the index at 32-bit floating point (f32) precision. Higher-precision values are accepted but lose precision when replicated to the index.
+ Fine-grained access control (FGAC) is not supported for the `SearchVectors` API.
+ `SearchVectors` responses are limited to 16 MB. Pagination is not supported. Because each result includes the projected attributes, using `ProjectionType: ALL` with large items and a high `TopK` value can approach this limit. If your items are large, use a narrower projection or reduce `TopK`.
+ Vector indexes do not support `Query` or `Scan` operations. Use the `SearchVectors` API to read from vector indexes.
+ Vector indexes are not accessible through PartiQL. Use the `SearchVectors` API to run similarity searches.

For numeric limits including maximum dimensions, TopK range, inline filters per index, indexes per table, and base table size thresholds, see [Vector indexes](ServiceQuotas.md#limits-vector-indexes) in [Quotas in Amazon DynamoDB](ServiceQuotas.md).