Datasette JSON Extractor

Point this Actor at any Datasette instance and pull its data out through Datasette's uniform JSON API — no per-site scraper, no HTML parsing. Datasette publishes SQLite databases as browsable, queryable websites, and every instance exposes the same JSON endpoints, so one Actor works across all of them: the project's own datasette.io, covid-19.datasettes.com, the UK Register of Members' Interests, timezones.datasette.io, and the thousands of sites published with datasette publish.

Three modes, chosen automatically from your input

1. Discovery — map an instance

Give it just a base URL (and optionally turn on List databases & tables only) and it returns one row per (database, table) with the column list, primary keys, row count, and the direct table/database URLs — a cheap map of everything the instance serves.

{ "baseUrl": "https://latest.datasette.io", "listDatabasesOnly": true }

2. Rows — page a whole table

Add a database and table and it pages that table into a clean dataset, following Datasette's cursor until the table is exhausted or Max records is hit. Each row's own columns are spread to top-level fields (so the dataset mirrors the source table) plus a lossless raw object.

{ "baseUrl": "https://latest.datasette.io", "database": "fixtures", "table": "facetable", "maxRecords": 5000 }

3. SQL — run a read-only query

Provide a SQL string and it runs it against Datasette's read-only SQL API and returns one record per result row. Datasette blocks writes by default; this Actor never attempts them.

{
  "baseUrl": "https://latest.datasette.io",
  "database": "fixtures",
  "sql": "select state, count(*) as n from facetable group by state order by n desc"
}

Input

Field	Description
baseUrl (required)	The Datasette instance root, e.g. https://latest.datasette.io. Don't include /-/databases.json — just the root. A path-mounted instance (https://host/data) works too, and if you paste a database or table URL the Actor recovers the root automatically.
database	The database to read. Leave empty in discovery mode to list all; in rows/SQL mode the Actor defaults to the sole database when an instance has just one.
table	The table (or view) to page. Setting this switches to rows mode.
sql	An optional read-only SQL query. Takes precedence over table.
listDatabasesOnly	Discovery-only switch — map databases + tables without reading rows.
maxRecords	Max total records to emit (default 1000). Each record is one billable event.
pageSize	Rows requested per page in rows mode (default 1000, server-capped).
bearerToken	Optional bearer token for a private/gated instance. Never logged.
extraHeaders	Optional extra request headers (e.g. an API-gateway key).

Output

Every record shares a stable set of normalized fields — _source, _database, _table, _mode, _rowIndex, primaryKey — plus mode-specific data:

• Discovery rows carry tableName, rowCount, columns, primaryKeys, isView, tableUrl, databaseUrl.
• Rows / SQL records spread the source columns to top-level fields, plus a lossless raw object of the original row.

Pricing

Pay-per-event: one record charge per emitted row (a table in discovery mode, a row in rows/SQL mode). You pay only for what you extract.

Notes

• A table with 0 rows returns 0 records and exits cleanly.
• A non-Datasette host fails fast with a clear message.
• Public instances need no credentials. The bearerToken / extraHeaders inputs cover gated deployments and are never logged.