Skip to main content
Database is the resource type for tabular data — customers, orders, inventory, lookup tables, conversation logs, anything you’d keep in rows. Collections come in two flavors: Structured (with a schema and typed columns) and Unstructured (schema-less, JSON-like documents). Both are readable and writable from agents, workflows, and the built-in Datastore embedded services. Find it under Tools → Database. Database list — collections with Max, Current, Created by, Refresh, and New Database action

​
The Database List

Each row is one collection. Columns:
ColumnDescription
NameCollection name. For structured datasets, a small Table badge is shown.
MaxThe maximum row count allowed by the collection’s size limit.
CurrentThe actual number of rows currently stored.
Created byThe email of the user who created the collection.
ActionsShare, Rename, and a More menu with Delete and Export.
Use the search box to filter by name. Click Refresh to re-fetch counts.

​
Creating a Collection

Click + New Database to open the creation dialog. New Resource dialog — Database Type dropdown with "Unstructured Database" selected, Name field, Advanced options Pick a Database Type:
  • Unstructured Database — default. Schema-less; rows are JSON-like documents. Best for semi-structured data where fields vary per row.
  • Structured Database — define columns with types (string, number, boolean, date, enum) up front. Best for known schemas.
Give the collection a Name and, under Advanced, optionally set a row-count ceiling. Click Create.

​
Working Inside a Collection

Clicking a collection navigates into its detail view. The toolbar includes:
ButtonAction
SchemaOpen the schema editor (structured only) — add/remove columns, change types.
Import CSVBulk import rows from a CSV.
Download CSVExport the whole collection.
RefreshRe-fetch rows from the backend.
WrapToggle cell text wrapping.
A natural-language search bar sits above the rows: type something like get active users and Pinkfish converts the query into a SQL filter (structured) or a semantic search (unstructured). You can also type raw SQL when the search is three characters or longer. Row actions (right-click or the Actions column): Edit, Duplicate, Delete.

​
Structured vs unstructured

FeatureStructuredUnstructured
SchemaRequired, typed columnsNone — any JSON shape
SearchSQL + natural languageSemantic + text
Best forReference data, reportsLogs, events, flexible payloads
MCP serverdatastore-structureddatastore-unstructured

​
Sharing

Click Share on any collection row to open the share dialog. ACLs work the same as Connections: Read (use in workflows), Write (add and modify rows), Admin (share further). Shared collections show a small “Shared” indicator next to the name.

​
Using Databases in Agents and Workflows

Agents and workflows reach databases through the Datastore MCP servers. See the embedded reference:

Structured Datastore

Tools for querying and mutating rows in structured collections with SQL.

Unstructured Datastore

Tools for searching and writing to schema-less document collections.
Both servers pick up every collection your user can read. Reference a collection by its ID in the tool input.

​
Notes

  • Database collections are builder-only. Non-builder users can still reference collections shared with them in agent/workflow runs, but they can’t see or manage the list.
  • Row counts update asynchronously after bulk imports — the Current column may lag for a minute after a large CSV upload.
  • For file uploads (PDFs, images, spreadsheets) use File Store; for indexed retrieval use Knowledge Base.