Automating work with workflows

A workflow is an ordered list of steps you save once and run whenever you need it, against one database or every database in a project. Open Workflows from the sidebar, pick a project, and choose New workflow — the wizard walks you through the steps.

What a step can do

• Run SQL — reads or writes, held to exactly the same safety rules as a statement you type by hand.
• Check a number — run a scalar query and compare the result against a threshold, so a run can fail loudly when something is off.
• Export — run a query and write the rows to CSV, TSV, JSON or NDJSON.
• Import — load a delimited file into a table.
• Back up — dump a whole database to a file.
• Apply a migration — run a plain-SQL file.
• Compare — run one statement across several databases and diff the results.
• Schema drift — introspect two databases and report where their schemas have diverged.
• Personal-data scan — classify every column and report the categories and counts it finds.
• Notify — emit a notification and a line in the run report, touching no database.
• Ask AI — have a model write SQL grounded on your live schema, review an earlier step’s result, or plan mock rows for a table.

Steps can depend on each other

A later step can wait on an earlier one, so a run stops instead of charging ahead when a check fails. You can also point a review step at an earlier step to have its output summarised. AddisDB validates the order before you save and tells you plainly what is wrong — a step waiting on something that has not run yet, two steps sharing a name, or a workflow where everything is disabled.

Scheduling

Give a workflow a schedule and it runs on its own — while AddisDB is open. That is a real limitation and the Workflows view says so on screen. If a scheduled slot passed while the app was closed, AddisDB runs it once the next time you open it, rather than firing every missed run at once.

Reading the run log

Every run is recorded under Workflow activity — what ran, against which databases, what it produced and what went wrong. The log keeps counts, verdicts, file paths and errors; it never stores your result rows. Older runs are trimmed automatically.