Can a prompt replace Stackby?

Build a records database whose distinguishing feature is a column type backed by an HTTP request, designed to work at thousands of rows.

Base schema: tables with fields typed as text, number, date, boolean, single select, multi select, attachment, link to another table, formula, and API column.

The API column, which is the point of this build. Its definition holds:

- A named connection (base URL, authentication method, credentials encrypted at rest, and a declared rate limit and quota).
- A request template with placeholders referencing other fields in the same row.
- A response path selecting the value to store (a JSON pointer), plus a type to coerce it to.
- A cache lifetime, after which a value is considered stale.
- A refresh policy: manual, on row change, or scheduled.

Working at scale, which is what separates this from the spreadsheet version of the idea:

- A shared token-bucket rate limiter per connection, respected across every column and every row using that connection. Refreshing three API columns on ten thousand rows must not exceed the provider's limit or exhaust a daily quota by lunchtime.
- A refresh queue with visible progress, pause and cancel, processing rows in batches and persisting after each batch so an interruption loses nothing.
- Per-cell state: fresh, stale, refreshing, or failed with the provider's own error message stored. A failed cell keeps its last successful value and shows both, because a table that blanks a column on a transient 429 is worse than one that shows an old number honestly.
- Quota accounting per connection: requests used today and this month against the declared limit, shown before any bulk refresh is started, with the projected cost of that refresh in requests.
- Conditional refresh: only rows matching a filter, so the expensive column can be refreshed for the fifty rows that matter rather than all ten thousand.

Honesty about staleness: every API column shows the age of its values in the column header, and a view can be filtered to rows whose values are stale. Data fetched three weeks ago presented as current is the failure mode of every tool like this.

Views: grid, kanban, gallery, with saved filters and sorts. Formula fields can reference API column values.

Also build: CSV import and export, a REST API over the tables themselves, and scheduled backups.

Out of scope: a catalogue of pre-built connectors, real-time collaborative editing, an automation builder, and any hosted multi-tenant deployment. Note in the README that this catalogue's spreadsheet-with-API-formulas entry covers the small-scale version of the same idea and may be the better fit under a few hundred rows.