Skip to main content
A Data Source is a named, reusable read: the operation to run, its inputs, the columns it returns, and any supported cache settings. A Connection supplies authentication to the provider. A Data Source uses that connection; it does not contain or share the author’s credentials. Use sources for NetSuite queries, supported connector reads, and Erstan data such as Sheets, Databases, Tasks, or tabular Files. Write operations belong in an App’s Actions, not in a Data Source.

Create a source

1

Open Data sources

Choose Data sources from your profile menu, then create a source. You can also start from Edit App → Data to keep the setup inside the App.
2

Choose the data

Search or browse the available connected systems and Erstan resources. Select a supported read and configure its query, resource, or arguments. NetSuite options depend on the available MCP, Add-on, or query connection and your access.
3

Define inputs and preview

Give the source a clear name. Define any inputs, types, defaults, and required values. Choose Run preview with representative inputs and inspect the returned rows. Where available, Use preview schema declares the returned columns for later compatibility checks.
4

Save and publish

Use Save draft, then Review and publish. Review the compatibility notes and affected Apps before confirming. A saved draft is not yet available to consumers.
Previews use your current access and read live data. They return a limited sample, so preview rows and totals must not be treated as the complete dataset. Configure and test paging where the read supports it.

Reuse and share

Published sources can be used by Apps and Agents that have access. Grant Can use for reads or Can edit for authoring. An App may fix some inputs and allow its users to supply the remaining declared inputs. Each consumer still needs access to the underlying provider or Erstan resource. A source share does not grant those permissions.

Update a published source

Changes stay in the draft until published. Apps and Agents use the current published definition; publication updates the read definition rather than freezing its results. Publish review checks input and output compatibility and shows known App dependencies. A breaking change is blocked. Preserve the existing contract or create a new source and migrate consumers deliberately.
Used by lists known App consumers. It does not inventory every dynamic Agent read, and inaccessible Apps may be represented by a count. An empty list does not prove that a source is unused.

Caching and usage

Caching is available only where the provider’s access can be checked safely. Results are scoped to the reader, inputs, page, and published definition; they are not shared across users. Preview bypasses the cache. A Data Source read itself has no tool-call charge. Calls to paid third-party connector tools can incur usage when dispatched, including previews. Cache hits, Erstan-hosted data reads, and the customer’s NetSuite or own MCP tools do not incur that per-call charge. AI work that creates or interprets a source can still consume model usage.

Retire a source

  • Deprecate source removes it from default selection while existing reads keep working.
  • Revoke and archive stops future reads, including cached reads. Review affected Apps before revoking.
  • Removing an App’s binding leaves the shared source intact.
If an App loses data after a source change, check its inputs, declared output columns, source status, and the current user’s access.