Engine by engine: the objects with Ask Forge tools of their own, the engine-specific settings on shared tools, and what Ask Forge reports about each engine.
By the SchemaSmith Team · Last reviewed
They are platform-gated, and the gate is not cosmetic.
Most Ask Forge tools work the same on every supported engine. A few manage objects that only some engines have, and some shared tools accept settings that only some engines read. This page lists both, engine by engine.
The engine a tool works against is the platform of the product in context: the Platform in its Product.json. platform is required on schemasmith_create_product and cannot be changed afterward, because the engine reads every definition in the package as that platform's own model (a SQL Server column and a PostgreSQL column are different classes), so set it when the product is created.
Platform-specific tools are available whatever the current platform. They return a clear error message if called against an unsupported platform, and they write nothing.
Over MCP, the tools that create or change these objects are Tier 3 tools, so they appear in the tool list and run only with confirm: true. The tools that list, show and compare views are Tier 1 tools that the AI reaches through schemasmith_execute_tool. Platform-Specific Tools lists the engine-specific tools in one table.
On SQL Server, the object type with tools of its own is the indexed view:
| Object | Explore | Modify |
|---|---|---|
| Indexed views | schemasmith_list_indexed_views, schemasmith_show_indexed_view, schemasmith_diff_indexed_views |
schemasmith_create_indexed_view |
schemasmith_create_indexed_view creates a new indexed view JSON definition file holding the view's defining SELECT; SchemaQuench adds WITH SCHEMABINDING itself when it creates the view. The stored Name and Schema are written in [brackets], the form SQL Server's own extraction writes and its deployment expects. schemasmith_validate_json reports an indexed view with no index, or with no clustered index, and schemasmith_audit_version_floor scans indexed views' indexes for features above the product's floor.
With verbose, the list adds each view's index count and whether it has a clustered index:
ask-forge list-indexed-views --verbose
schemasmith_apply_change has change types for SQL Server's spatial indexes, JSON indexes, edge constraints and ledger view, and set_cdc for change data capture. On any other platform they are refused, and nothing is written. set_fulltext writes a SQL Server table's full-text index.schemasmith_manage_connection takes integrated_security, which uses Windows Authentication for the connection.server_to_quench on schemasmith_configure_template, and on product-level folders in schemasmith_add_folder, selects which replica the scripts run against: Primary, Secondary, or Both.schemasmith_show_platform_capabilities reports this for SQL Server:
| Report line | SQL Server |
|---|---|
| Features reported as supported | Schemas, temporal tables, full-text search, XML indexes, computed columns, partitioning, JSON columns, CDC, data masking, Always Encrypted, row-level security |
| Maximum identifier length | 128 |
| Identifier quoting | [example] |
| Default schema | dbo |
| Identity syntax | IDENTITY(1,1) |
| Naming style | PascalCase |
| PK, FK, index and unique prefixes | PK_, FK_, IX_, UQ_ |
On PostgreSQL, the object types with tools of their own are materialized views, enum types, domain types, sequences and row-level security policies:
| Object | Explore | Modify |
|---|---|---|
| Materialized views | schemasmith_list_materialized_views, schemasmith_show_materialized_view, schemasmith_diff_materialized_views |
schemasmith_create_materialized_view |
| Enum types | None | schemasmith_create_enum_type, schemasmith_rename_modeled_object |
| Domain types | None | schemasmith_create_domain_type, schemasmith_rename_modeled_object |
| Sequences | None | schemasmith_create_sequence, schemasmith_rename_modeled_object |
| Row-level security policies | None | schemasmith_configure_policies |
schemasmith_create_materialized_view creates a new materialized view JSON definition file with its defining query. with_data populates the view on creation (default true); set it false to create it unpopulated and refresh later. It also takes tablespace and access_method.
The enum type, domain type and sequence tools write a typed JSON definition into a script folder: Enum Types/, Domain Types/ and Sequences/. For an enum type, the order you give the labels is the declared order, and that order matters: PostgreSQL sorts an enum by declared position. A domain type's data_type is create-time only, because PostgreSQL has no ALTER DOMAIN ... TYPE. SchemaQuench creates a sequence, or alters an existing one in place, and never resets its current value.
schemasmith_configure_policies adds, updates or removes one row-level security policy on an existing table, and turns the table's row-level security on or off. The two settings work together: a policy is enforced only while the table's row-level security is on.
This writes Enum Types/public.OrderStatus.json, with the labels in the order given:
ask-forge create-enum-type --name OrderStatus --values "draft,active,archived"
schemasmith_configure_data_delivery takes merge_disable_rules and merge_update_descendents, which disable rules during the merge and cascade it to inherited tables. On any other platform they are refused.schemasmith_configure_drop_policy takes drop_exclude_constraints, which drops exclude constraints removed from the product.schemasmith_show_platform_capabilities reports this for PostgreSQL:
| Report line | PostgreSQL |
|---|---|
| Features reported as supported | Schemas, full-text search, computed columns, partitioning, enum types, domain types, array types, JSON columns, sequences, rules, row-level security, deferred constraints, exclude constraints, table persistence |
| Maximum identifier length | 63 |
| Identifier quoting | "example" |
| Default schema | public |
| Identity syntax | GENERATED ALWAYS AS IDENTITY |
| Naming style | snake_case, lowercase |
| PK, FK, index and unique prefixes | pk_, fk_, ix_, uq_ |
On MySQL, the object types with tools of their own are scheduled events and table partitioning:
| Object | Explore | Modify |
|---|---|---|
| Events | None | schemasmith_create_event, schemasmith_rename_modeled_object |
| Partitioning | schemasmith_show_table |
schemasmith_configure_partitioning |
schemasmith_create_event writes a typed JSON definition into the template's Events/ folder. MySQL has no schema concept, so the event takes no schema: its name is the whole identity. schedule_type is EVERY (recurring, the default) or AT (one-shot), and each takes its own fields: EVERY uses interval plus the optional starts and ends bounds, and AT uses execute_at.
schemasmith_configure_partitioning sets or removes the partitioning of an existing table in one call. The call declares the whole partitioning and replaces what the table declared, so a part you leave out is gone. SchemaQuench partitions a table when it creates it. On a database where the table already exists, it refuses a changed method, expression, count, subpartitioning or partition list, because repartitioning rewrites every row; the one change it applies is a RANGE or LIST partition added after the last one, when that one is not MAXVALUE. schemasmith_show_table lists the table's partitioning, so read the declaration first.
MySQL has no sequence at any version, so schemasmith_create_sequence refuses on a MySQL product.
This writes Events/purge_old_sessions.json, an event that runs every day:
ask-forge create-event --name purge_old_sessions --interval "1 DAY" `
--definition "DELETE FROM sessions WHERE expires_at < NOW()"
ask-forge create-event --name purge_old_sessions --interval "1 DAY" \
--definition "DELETE FROM sessions WHERE expires_at < NOW()"
time_zone, the session time zone the content file's TIMESTAMP values were extracted in, given as an offset such as +00:00.schemasmith_show_platform_capabilities reports this for MySQL:
| Report line | MySQL |
|---|---|
| Features reported as supported | Full-text search, computed columns, partitioning, JSON columns, storage engines, events |
| Maximum identifier length | 64 |
| Identifier quoting | `example` |
| Default schema | None |
| Identity syntax | AUTO_INCREMENT |
| Naming style | PascalCase |
| PK, FK, index and unique prefixes | PK_, FK_, IX_, UQ_ |
On MariaDB, the object types with tools of their own are sequences, application-time periods, scheduled events and table partitioning:
| Object | Explore | Modify |
|---|---|---|
| Sequences | None | schemasmith_create_sequence, schemasmith_rename_modeled_object |
| Application-time periods | None | schemasmith_configure_periods |
| Events | None | schemasmith_create_event, schemasmith_rename_modeled_object |
| Partitioning | schemasmith_show_table |
schemasmith_configure_partitioning |
A MariaDB product has its own platform, MariaDb, and its own connection, chosen separately from MySQL's and never folded into it.
On MariaDB a sequence's name is not schema-qualified, so schemasmith_create_sequence writes Sequences/order_no.json (or order_no.<variant>.json), and refuses schema, persistence_type and description. data_type takes tinyint, smallint, mediumint, int, integer or bigint, each optionally followed by unsigned. Sequences need MariaDB 10.3 or later; below it, a sequence is skipped and recorded as a downgrade, or refused when the unsupported-feature policy is fail. A type other than a plain bigint needs 11.5 or later: on 10.3 through 11.4 every sequence is a signed bigint, and one that declares another type is recorded as a downgrade, or refused under the same policy. SchemaQuench creates the sequence, or alters an existing one in place, and never resets its current value.
schemasmith_configure_periods adds, updates or removes one application-time period (PERIOD FOR) on an existing table, without rewriting the table. A period gives a table a business-time validity window between a start column and an end column. Adding a period needs both start_column and end_column, and both must be columns of the table. MariaDB allows one application-time period per table, so adding a second is refused, and nothing is written.
schemasmith_create_event writes a typed JSON definition into the template's Events/ folder, with a recurring EVERY schedule or a one-shot AT. schemasmith_configure_partitioning sets or removes a table's whole partitioning in one call, and the declaration it writes replaces what the table declared. On a database where the table already exists, SchemaQuench refuses a changed partitioning, because repartitioning rewrites every row; the one change it applies is a RANGE or LIST partition added after the last one, when that one is not MAXVALUE. schemasmith_show_table lists the table's partitioning.
This writes Sequences/order_no.json, a sequence that starts at 1000:
ask-forge create-sequence --name order_no --start 1000
drop_periods, valid only at the table tier, which drops application-time periods removed from the table's definition.schemasmith_show_platform_capabilities reports this for MariaDB:
| Report line | MariaDB |
|---|---|
| Features reported as supported | Full-text search, computed columns, partitioning, JSON columns, sequences, storage engines, events |
| Maximum identifier length | 64 |
| Identifier quoting | `example` |
| Default schema | None |
| Identity syntax | AUTO_INCREMENT |
| Naming style | PascalCase |
| PK, FK, index and unique prefixes | PK_, FK_, IX_, UQ_ |
schemasmith_show_platform_capabilities, a Tier 1 tool, reports on one engine at a time. It reports two separate things, which are easy to confuse:
MinimumVersion, what each one needs, and whether it is skipped outright, created with reduced fidelity, or refused, which stops the deployment. Where the engine records a downgrade, the exact manifest entry is quoted so you can search for it after a deployment.A feature can appear in both: SQL Server supports temporal tables, and still degrades them below 2016. For some engines, the report also lists target prerequisites: database or server setup that a feature needs on any version, such as change data capture enabled on a SQL Server database.
The floor is read from the product in context. Ask about a different platform than your product's and the report falls back to that platform's assumed minimum instead, rather than misapplying your product's version to another engine. To see where a template actually uses an above-floor feature, run the version floor audit.
The platform is optional and defaults to the context's platform:
ask-forge show-platform-capabilities --platform MariaDb
Free schema-as-code for SQL Server, PostgreSQL, MySQL, and MariaDB.
The source is on GitHub for anyone to read.
Get Started on GitHub