chore(databases): clarify default_connection_id usage and catalog attachment rules

.openapi-generator/FILES

@@ -89,6 +89,7 @@ docs/ListWorkspacesResponse.md
 docs/LoadManagedTableRequest.md
 docs/LoadManagedTableResponse.md
 docs/ManagedSchemaResponse.md
+docs/ManagedTableConstantPerKeyResponse.md
 docs/ManagedTableResponse.md
 docs/MintUploadPartsRequest.md
 docs/MintUploadPartsResponse.md
@@ -115,6 +116,7 @@ docs/TemporalProfileDetail.md
 docs/TextProfileDetail.md
 docs/UpdateEmbeddingProviderRequest.md
 docs/UpdateEmbeddingProviderResponse.md
+docs/UpdateManagedTableRequest.md
 docs/UpdateSavedQueryRequest.md
 docs/UploadSessionResponse.md
 docs/UploadsApi.md
@@ -224,6 +226,7 @@ src/models/list_workspaces_response.rs
 src/models/load_managed_table_request.rs
 src/models/load_managed_table_response.rs
 src/models/managed_schema_response.rs
+src/models/managed_table_constant_per_key_response.rs
 src/models/managed_table_response.rs
 src/models/mint_upload_parts_request.rs
 src/models/mint_upload_parts_response.rs
@@ -247,6 +250,7 @@ src/models/temporal_profile_detail.rs
 src/models/text_profile_detail.rs
 src/models/update_embedding_provider_request.rs
 src/models/update_embedding_provider_response.rs
+src/models/update_managed_table_request.rs
 src/models/update_saved_query_request.rs
 src/models/upload_session_response.rs
 src/models/upsert_database_context_request.rs

CHANGELOG.md

@@ -7,6 +7,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Changed
+
+- chore(databases): clarify default_connection_id usage and catalog attachment rules
 
 ## [0.18.1] - 2026-09-18
 

docs/AddManagedTableDecl.md

@@ -4,8 +4,8 @@
 
 Name | Type | Description | Notes
 ------------ | ------------- | ------------- | -------------
+**constant_per_key** | Option<**Vec<String>**> | Columns whose value is determined by this table's `key`: for every uploaded row, every stored row sharing its key holds the same value of these columns.  Declaring this lets a keyed mutation (`delete`, `update`, `upsert`) restrict its search for prior versions to the values the upload carries, which prunes far harder than the key alone when the key's own file statistics are weak. Omit (the default) for the unrestricted search.  **Correctness-affecting, not a hint.** If the assertion is false, a mutation supersedes one version of a key and appends beside another, silently duplicating it. Declare it only where the invariant is established. | [optional]
 **key** | Option<**Vec<String>**> | Columns that uniquely identify a row, enabling the key-based load modes (`delete`, `update`, `upsert`) on this table: those loads match rows by these columns' values. Omit (the default) to declare no key; the table can still be loaded with `replace` and `append`, but key-based modes are then rejected. | [optional]
-**key_determines** | Option<**Vec<String>**> | Columns whose value is determined by this table's `key`: for every uploaded row, every stored row sharing its key holds the same value of these columns.  Declaring this lets a keyed mutation (`delete`, `update`, `upsert`) restrict its search for prior versions to the values the upload carries, which prunes far harder than the key alone when the key's own file statistics are weak. Omit (the default) for the unrestricted search.  **Correctness-affecting, not a hint.** If the assertion is false, a mutation supersedes one version of a key and appends beside another, silently duplicating it. Declare it only where the invariant is established. | [optional]
 **name** | **String** |  | 
 **partition_by** | Option<[**Vec<models::TablePartitionKey>**](TablePartitionKey.md)> | Partition keys for this table, applied in order. Omit for no partitioning. Declared when the table is created and fixed thereafter. | [optional]
 **sorted_by** | Option<[**Vec<models::TableSortKey>**](TableSortKey.md)> | Sort keys for this table, applied in order. Omit for no sort order. Declared when the table is created and fixed thereafter. | [optional]

docs/AddManagedTableRequest.md

@@ -4,8 +4,8 @@
 
 Name | Type | Description | Notes
 ------------ | ------------- | ------------- | -------------
+**constant_per_key** | Option<**Vec<String>**> | Columns whose value is determined by this table's `key`: for every uploaded row, every stored row sharing its key holds the same value of these columns.  Declaring this lets a keyed mutation (`delete`, `update`, `upsert`) restrict its search for prior versions to the values the upload carries, which prunes far harder than the key alone when the key's own file statistics are weak. Omit (the default) for the unrestricted search.  **Correctness-affecting, not a hint.** If the assertion is false, a mutation supersedes one version of a key and appends beside another, silently duplicating it. Declare it only where the invariant is established. | [optional]
 **key** | Option<**Vec<String>**> | Columns that uniquely identify a row, enabling the key-based load modes (`delete`, `update`, `upsert`) on this table: those loads match rows by these columns' values. Omit (the default) to declare no key; the table can still be loaded with `replace` and `append`, but key-based modes are then rejected. | [optional]
-**key_determines** | Option<**Vec<String>**> | Columns whose value is determined by this table's `key`: for every uploaded row, every stored row sharing its key holds the same value of these columns.  Declaring this lets a keyed mutation (`delete`, `update`, `upsert`) restrict its search for prior versions to the values the upload carries, which prunes far harder than the key alone when the key's own file statistics are weak. Omit (the default) for the unrestricted search.  **Correctness-affecting, not a hint.** If the assertion is false, a mutation supersedes one version of a key and appends beside another, silently duplicating it. Declare it only where the invariant is established. | [optional]
 **name** | **String** |  | 
 **partition_by** | Option<[**Vec<models::TablePartitionKey>**](TablePartitionKey.md)> | Partition keys for this table, applied in order. Omit for no partitioning. Declared when the table is created and fixed thereafter. | [optional]
 **sorted_by** | Option<[**Vec<models::TableSortKey>**](TableSortKey.md)> | Sort keys for this table, applied in order. Omit for no sort order. Declared when the table is created and fixed thereafter. | [optional]

docs/ConnectionsApi.md

@@ -13,6 +13,7 @@ Method | HTTP request | Description
 [**get_table_profile**](ConnectionsApi.md#get_table_profile) | **GET** /v1/connections/{connection_id}/tables/{schema}/{table}/profile | Get table profile
 [**list_connections**](ConnectionsApi.md#list_connections) | **GET** /v1/connections | List connections
 [**load_managed_table**](ConnectionsApi.md#load_managed_table) | **POST** /v1/connections/{connection_id}/schemas/{schema}/tables/{table}/loads | Load managed table from inline data, upload, or query result
+[**set_managed_table_constant_per_key**](ConnectionsApi.md#set_managed_table_constant_per_key) | **PUT** /v1/connections/{connection_id}/schemas/{schema}/tables/{table}/constant-per-key | Declare which columns are constant per key
 
 
 
@@ -292,3 +293,36 @@ Name | Type | Description  | Required | Notes
 
 [[Back to top]](#) [[Back to API list]](../README.md#documentation-for-api-endpoints) [[Back to Model list]](../README.md#documentation-for-models) [[Back to README]](../README.md)
 
+
+## set_managed_table_constant_per_key
+
+> models::ManagedTableConstantPerKeyResponse set_managed_table_constant_per_key(connection_id, schema, table, update_managed_table_request)
+Declare which columns are constant per key
+
+Replace the columns a table declares constant for a given key: for every row, any other row sharing its key holds the same value of these columns. Declaring this lets a keyed mutation (`delete`, `update`, `upsert`) narrow its search for prior versions to the values the upload carries, which prunes far harder than the key alone when the key's own file statistics do not discriminate.  Unlike `partition_by` and `sorted_by`, this is NOT fixed when the table is created. It changes only which files a mutation opens, never how rows are written, so nothing stored becomes wrong when it changes and a populated table can adopt it with no rewrite. It takes effect on the next load.  Send an empty array to revoke it, restoring the unrestricted search — this is the way to undo a declaration that turns out to be false.  **This is correctness-affecting, not a hint.** If the assertion is false, a keyed mutation supersedes one version of a key and appends beside another, silently duplicating it, and the pruning conceals its own evidence because the file holding the missed row is never opened. Declare it only where the invariant is established.
+
+### Parameters
+
+
+Name | Type | Description  | Required | Notes
+------------- | ------------- | ------------- | ------------- | -------------
+**connection_id** | **String** | Connection ID | [required] |
+**schema** | **String** | Schema name | [required] |
+**table** | **String** | Table name | [required] |
+**update_managed_table_request** | [**UpdateManagedTableRequest**](UpdateManagedTableRequest.md) |  | [required] |
+
+### Return type
+
+[**models::ManagedTableConstantPerKeyResponse**](ManagedTableConstantPerKeyResponse.md)
+
+### Authorization
+
+[WorkspaceId](../README.md#WorkspaceId), [BearerAuth](../README.md#BearerAuth)
+
+### HTTP request headers
+
+- **Content-Type**: application/json
+- **Accept**: application/json
+
+[[Back to top]](#) [[Back to API list]](../README.md#documentation-for-api-endpoints) [[Back to Model list]](../README.md#documentation-for-models) [[Back to README]](../README.md)
+

docs/CreateDatabaseResponse.md

@@ -6,7 +6,7 @@ Name | Type | Description | Notes
 ------------ | ------------- | ------------- | -------------
 **created** | Option<**bool**> | Whether this call brought the database into existence.  Only `false` when `if_not_exists` found a database already carrying the requested name, in which case nothing was created and the existing one is returned. The response status says the same thing — `201` against `200` — but generated clients often surface only the body, so it is stated here as well.  Always sent. It is declared optional so that a client built against a newer version of this API still accepts a response from a deployment that predates the field. Absent therefore means \"this deployment cannot say\", which is not the same as `false` — test for the two values explicitly rather than for truthiness. | [optional]
 **default_catalog** | **String** | Name the database's default catalog answers to inside its query scope (`default` unless overridden at create time). | 
-**default_connection_id** | **String** | Internal id of the connection that backs this database's `default` catalog. Workspace-level connection endpoints (list, get, health, delete, cache purge) refuse to act on this id — it is exposed only for the managed-tables load endpoint (`POST /v1/connections/{id}/schemas/{s}/tables/{t}/loads`) so callers can load data into tables declared at database-create time. Addressing it directly in SQL is not the recommended path — use `default` inside an `X-Database-Id` scope instead. | 
+**default_connection_id** | **String** | Id of the connection that backs this database's `default` catalog. Two uses: pass it as `connection_id` to `POST /v1/databases/{other}/catalogs` to attach this database's catalog into another database, and as the connection in the managed-tables load endpoint (`POST /v1/connections/{id}/schemas/{s}/tables/{t}/loads`) to load tables declared at create time. Other connection endpoints (list, get, health, delete, cache purge) refuse to act on it. In SQL, address the catalog as `default` inside an `X-Database-Id` scope, not by this id. | 
 **default_schema** | **String** | Schema that unqualified table names resolve to inside this database's query scope. `main` unless the database declares a single schema or a `default_schema` was set at create time. | 
 **expires_at** | Option<**String**> | When this database expires. | [optional]
 **forked_from** | Option<[**models::ForkedFromInfo**](ForkedFromInfo.md)> |  | [optional]

docs/CreateIndexRequest.md

@@ -14,6 +14,7 @@ Name | Type | Description | Notes
 **index_type** | Option<**IndexType**> | Index type. `sorted` supports range queries, `bm25` full-text search, and `vector` similarity search. (enum: sorted, bm25, vector) | [optional][default to Sorted]
 **metric** | Option<**String**> | Distance metric for vector indexes: \"l2\", \"cosine\", or \"dot\". When omitted, defaults to \"l2\" for float array columns or the provider's preferred metric for text columns with auto-embedding. | [optional]
 **output_column** | Option<**String**> | Custom name for the generated embedding column. Defaults to `{column}_embedding`. | [optional]
+**vector_precision** | Option<**VectorPrecision**> | How precisely a vector index stores each number of a vector. Lower precision shrinks the index so a larger table can be indexed within the same memory, and lets searches run on a smaller instance. Omit this field to store vectors at the same precision as the column, which is the default.  The quality figures below come from one benchmark — 1536-dimension text embeddings, cosine distance, default search settings — and are a guide, not a guarantee. Other models, dimensions, distance metrics and data distributions behave differently, so measure on your own data before moving a production index to a lower precision.  `float32` — on a `float64` column this halves the index. Widely used embedding models emit 32-bit values, so for those nothing is lost; vectors that genuinely carry more than 32 bits of precision will lose some.  `float16` — half the memory of `float32`. In that benchmark its results matched `float32` to within 0.1 percentage points.  `float8` — a quarter of the memory of `float32`. In that benchmark it scored about 4 percentage points below `float32`, and raising the search effort did not close the gap, so treat the reduction as permanent for a given index.  `float64` — accepted only for a column that already holds double-precision values; it cannot add precision the stored data does not have.  Changing this means dropping the index and creating it again. It affects only the index: the table's own values are never altered, and text columns indexed with a generated embedding are not re-embedded. (enum: float64, float32, float16, float8) | [optional]
 
 [[Back to Model list]](../README.md#documentation-for-models) [[Back to API list]](../README.md#documentation-for-api-endpoints) [[Back to README]](../README.md)
 

docs/CreateUploadRequest.md

@@ -8,7 +8,7 @@ Name | Type | Description | Notes
 **checksum_value** | Option<**String**> | Integrity checksum value, paired with `checksum_algo`. Optional. | [optional]
 **content_encoding** | Option<**String**> | Content encoding to record for the uploaded file (for example `gzip`). Optional. | [optional]
 **content_type** | Option<**String**> | Content type to record for the uploaded file (for example the Parquet, CSV, or JSON MIME type). Optional. | [optional]
-**declared_size_bytes** | Option<**i64**> | The exact size, in bytes, of the file you will upload. Optional. When provided, it is validated at create time against the maximum allowed size, and again at finalize against the bytes actually uploaded — a mismatch fails the finalize. Omit it to create a streaming (unknown-size) upload: the session is always multi-part and returns no part URLs up front; instead you mint part URLs on demand from `POST /v1/uploads/{upload_id}/parts` as you upload, and finalize validates only that the file is non-empty. | [optional]
+**declared_size_bytes** | Option<**i64**> | The exact size, in bytes, of the file you will upload. Optional. When provided, it is checked at create time against the maximum upload size (16 GiB by default, the same for every file format), so an oversized file is refused before you transfer any of it; it is checked again at finalize against the bytes actually uploaded — a mismatch fails the finalize. Omit it to create a streaming (unknown-size) upload: the session is always multi-part and returns no part URLs up front; instead you mint part URLs on demand from `POST /v1/uploads/{upload_id}/parts` as you upload, and finalize checks only that the file is non-empty and within the maximum upload size. | [optional]
 **filename** | Option<**String**> | Original file name, recorded with the upload for your own bookkeeping. Optional and advisory — it does not affect how the file is uploaded or loaded. | [optional]
 **part_size** | Option<**i64**> | Preferred size, in bytes, of each part for a large (multi-part) upload. Optional hint — the service clamps it to the allowed part-size range and to the maximum number of parts, and ignores it for small files uploaded with a single `PUT`. Omit to let the service choose. | [optional]
 

docs/DatabaseDefaultTableDecl.md

@@ -4,8 +4,8 @@
 
 Name | Type | Description | Notes
 ------------ | ------------- | ------------- | -------------
+**constant_per_key** | Option<**Vec<String>**> | Columns whose value is determined by this table's `key`: for every uploaded row, every stored row sharing its key holds the same value of these columns.  Declaring this lets a keyed mutation (`delete`, `update`, `upsert`) restrict its search for prior versions to the values the upload carries, which prunes far harder than the key alone when the key's own file statistics are weak. Omit (the default) for the unrestricted search.  **Correctness-affecting, not a hint.** If the assertion is false, a mutation supersedes one version of a key and appends beside another, silently duplicating it. Declare it only where the invariant is established. | [optional]
 **key** | Option<**Vec<String>**> | Columns that uniquely identify a row, enabling the key-based load modes (`delete`, `update`, `upsert`) on this table: those loads match rows by these columns' values. Omit (the default) to declare no key; the table can still be loaded with `replace` and `append`, but key-based modes are then rejected. | [optional]
-**key_determines** | Option<**Vec<String>**> | Columns whose value is determined by this table's `key`: for every uploaded row, every stored row sharing its key holds the same value of these columns.  Declaring this lets a keyed mutation (`delete`, `update`, `upsert`) restrict its search for prior versions to the values the upload carries, which prunes far harder than the key alone when the key's own file statistics are weak. Omit (the default) for the unrestricted search.  **Correctness-affecting, not a hint.** If the assertion is false, a mutation supersedes one version of a key and appends beside another, silently duplicating it. Declare it only where the invariant is established. | [optional]
 **name** | **String** |  | 
 **partition_by** | Option<[**Vec<models::TablePartitionKey>**](TablePartitionKey.md)> | Partition keys for this table, applied in order. Omit for no partitioning. Declared when the table is created and fixed thereafter. | [optional]
 **sorted_by** | Option<[**Vec<models::TableSortKey>**](TableSortKey.md)> | Sort keys for this table, applied in order. Omit for no sort order. Declared when the table is created and fixed thereafter. | [optional]

docs/DatabaseDetailResponse.md

@@ -7,7 +7,7 @@ Name | Type | Description | Notes
 **attachments** | [**Vec<models::DatabaseAttachmentInfo>**](DatabaseAttachmentInfo.md) |  | 
 **created_at** | Option<**String**> | When the database was created. | [optional]
 **default_catalog** | **String** | Name the database's default catalog answers to inside its query scope (`default` unless overridden at create time). | 
-**default_connection_id** | **String** |  | 
+**default_connection_id** | **String** | Id of the connection backing this database's `default` catalog. Pass it as `connection_id` to `POST /v1/databases/{other}/catalogs` to attach this database's catalog into another database. In SQL, address the catalog as `default` inside an `X-Database-Id` scope, not by id. | 
 **default_schema** | **String** | Schema that unqualified table names resolve to inside this database's query scope. `main` unless the database declares a single schema or a `default_schema` was set at create time. | 
 **expires_at** | Option<**String**> | When this database expires. | [optional]
 **forked_from** | Option<[**models::ForkedFromInfo**](ForkedFromInfo.md)> |  | [optional]