Managing Sandcastle Data

Every sandcastle with persistent state has one or more collections — document stores that hold the app's data. This guide covers how to view, edit, and manage that data outside the app itself.

Before you begin

  • You need at least execute access to the sandcastle to view data.
  • You need edit or admin access to modify data or reset collections.

Viewing collections and documents

Go to Sandcastles, click the sandcastle, then click Data. This opens the data browser.

  • The left sidebar lists every collection the sandcastle has. Click a collection to see its documents.
  • The main panel shows the documents in the selected collection as a scrollable list. Each document displays its fields and values.

Use this page to:

  • Verify that the app is writing data correctly.
  • Inspect a specific document's fields (useful for debugging).
  • Understand the data shape before asking the AI to build a new view.

Creating and editing documents

From the data browser, you can create new documents and edit existing ones directly. This is useful for:

  • Seeding test data while developing the app.
  • Fixing a bad record without going through the app's UI.
  • Adding data that the app's forms don't cover yet.

Documents are JSON. Each document has a unique _id field assigned automatically.

Resetting a collection

If the data is in a bad state (broken seed data, test pollution, schema drift), you can reset a collection from the data page. Resetting deletes all documents in the collection. This is a destructive action — you will be asked to confirm by typing the collection name.

Reset is useful during development. Be careful in production — there is no undo.

How the linked subagent interacts with data

The linked subagent has its own set of state tools for reading and writing to the sandcastle's collections:

  • Find documents — query with filters, sort, and pagination.
  • Insert documents — add one or more new documents.
  • Update a document — modify fields on an existing document.
  • Delete a document — remove a document by ID.
  • Describe schema — inspect the collection's field names and types.

These are the same operations the app's frontend uses, just invoked by the subagent instead of by a user clicking a button. This is how you automate data maintenance: schedule the subagent to clean up stale records, sync data from an external system, or generate summary documents.

Limits

  • 100 documents per query. If a collection has more than 100 documents, use filters to narrow the result set, or paginate with skip and limit.
  • Collections are workspace-isolated. A sandcastle cannot read another sandcastle's collections, and a user in another workspace cannot see yours.
  • Structured data only. Collections store JSON documents, not files. For file storage, use the linked subagent's filesystem.

Next steps