NoSQL, document-shaped
A MongoDB database of your own. Find, count and full aggregation pipelines, with the collections and their counts described to the agent before it asks anything.
Managed NoSQL and SQL, created in a click. Collections with their counts and tables with their columns are put in front of the agent at the start of every answer, so it reads the shape of your data before it writes a single row.
A document database and a relational one, described to the agent the same way and reached through the same calls - so the choice between them is about your data, not about what the tooling supports.
A MongoDB database of your own. Find, count and full aggregation pipelines, with the collections and their counts described to the agent before it asks anything.
Tables, columns and indexes, queried with ordinary SQL. Every query runs in its own process and is stopped at its timeout, so one heavy statement cannot hold up the rest.
A delete with no filter, a statement with no WHERE, several statements in one call: each is refused at the tool, before you are ever asked to approve it. A DROP exists only in Full access.
MongoDB and Cloudflare D1. The credentials are encrypted at rest and never leave the server; the agent sees collections, tables and columns, not your keys.
A SQL database is a file of your own, and each query runs in its own process with a timeout - one heavy statement cannot hold up the rest of your account. ATTACH, DETACH, VACUUM INTO and load_extension are refused outright, because each one would reach a file outside your database or load native code. What is left is the SQL you actually wanted.
One cannot eat another's room: the included space is counted per database, for NoSQL as for SQL. Beyond it, gigabytes are bought at 100 credits per GB per month - and those bought gigabytes are a pool the two services share.
NoSQL is a MongoDB database of your own; SQL is your own file, queried with ordinary SQL. Both are created in a click, and both are reachable from the agent, the workflows and the SDK with no connection string to carry around.
No. Collections with their counts and tables with their columns are put in front of it at the start of every answer. It reads the shape of your data before it writes anything.
Only with your approval, and only after the guards. A write is asked even in Auto mode, an empty filter or a missing WHERE is refused at the tool before the prompt is even shown, and a DROP exists only in Full access.
Buy gigabytes at 100 credits per GB per month. They form a pool shared by NoSQL and SQL, so an overrun on either draws from the same purchase.