Making one
You make a database where you need it, from a Database node's own Database field: open the picker, choose New Database…, name it, Create. No schema step, no file path, no format to pick. The first Save to Database builds the table to fit whatever you store, alongside the two columns Watchflows manages — id and created_at, named plainly so any SQLite tool reads them as you would expect.
Settings ▸ Databases is the management surface, not the starting point. On a Mac with none it says so and points you back at the node.
Living with them
A tile carries the name as an editable label, the size on disk and the creation date, Show in Finder, and Delete.
- Renaming never breaks a flow. A node finds its database by a hidden id, so rename as often as you like. Two live databases cannot share a name, ignoring case.
- Private keeps a database out of the picker other flows browse, and out of the list the AI Flow Builder is told about — so it is never offered and never resolved into a draft. Flows already pointing at it run exactly as before.
- Delete counts the damage first. It names the nodes and flows pointing at the database before you agree, and says plainly when it could not check rather than implying nothing uses it. Agreeing is a soft delete: the file stays, the entry moves under Deleted, and Restore puts it back.
A missing file is left missing. Move the file away and the entry wears an orange MISSING badge. Nothing is recreated and flows pointing at it fail until it is back — a flow that appears to work while writing into nothing is worse than one that stops.
Settings ▸ Storage names the folder the files live in and can move it. A node points at a database by id, never by path, so relocating breaks nothing.
Writing SQL yourself
The everyday shapes need no SQL at all. When you want a join, a GROUP BY or an index, Run SQL takes one statement you write, and the Workbench gives that statement a full-window editor with its results beside it.