Manage Data Store

Create and evolve the tables, columns, indexes, and views of a structured knowledge base

What it does

manage_data_store lets an agent shape the schema of a data store (a structured knowledge base): create tables, add or alter columns, add indexes, create views, and inspect the current schema. Changes apply to the store’s own isolated Postgres database.

• create_store creates a brand-new data store (name + optional description) — no knowledge_base_id needed. Use it when work deserves its own store rather than adding unrelated tables to an existing one. Creation only: deleting stores stays with manage_knowledge_bases.

For an overview of structured data stores — what they are and when to use one — see Data Stores.

Parameters

Parameter	Type	Required	Description
knowledge_base_id	integer	Yes	The structured knowledge base to modify
action	string	Yes	One of create_store, create_table, add_column, alter_column, rename_table, rename_column, create_index, create_view, drop_column, drop_table, drop_view, describe
table	string	For most ops	Table (or view) name — lowercase snake_case
columns	array	create_table, add_column	Column definitions (name, type, nullable, default, primary_key)
column	string	alter_column, drop_column, rename_column	Target column
new_type	string	alter_column	New column type
new_name	string	rename_table, rename_column	New name. Renaming preserves the primary key, unique constraints and indexes — unlike dropping and recreating
index_columns	array	create_index	Columns to index
unique	boolean	No	Make a create_index unique
view_sql	string	create_view	A full SELECT computed server-side (joins, aggregates, window functions, CTEs)
confirm	boolean	For drops	Required true for drop_column, drop_table, drop_view

Key behaviors

• A primary key is added for you. create_table adds an id primary key automatically unless you define your own.
• Views give apps aggregation power. create_view stores an agent-authored SELECT that apps read like a table — the way to let an app show a rollup or join over a large table without pulling every row. Views run with the reading role’s privileges, so they stay safely scoped to the store.
• Destructive changes are gated. Dropping a column, table, or view requires confirm: true, because a published app may depend on it.
• Every change is audited, and the store’s cached schema — which the agent sees in its context — refreshes automatically after each change.

Common use cases

Create a table

operation: create_table
knowledge_base_id: 954
table: project_stages
columns:
  - { name: stage_name, type: text, nullable: false }
  - { name: status, type: text, nullable: false }
  - { name: sort_order, type: integer, nullable: false }

Create a view for an app to chart

operation: create_view
knowledge_base_id: 954
table: stage_summary
view_sql: SELECT status, count(*) AS n FROM project_stages GROUP BY status

Related

• Query Data Store — read data with SQL
• Write Data Store — insert, update, delete, upsert
• Data Stores — the feature overview