Supabase rows read, enriched and written back

Supabase exposes every project's Postgres tables as a REST API, and the Supabase integration speaks that API directly: read rows with a PostgREST filter, insert one or many, update or delete the rows a filter matches. That makes a table a natural source and sink for workflows: watch it for new rows, enrich them with an AI agent, write the result back, and hand a row the agent is unsure about to a person in Slack. Runs keep the rows each step read or changed, so a bad update traces back to its filter and input. Describe the workflow in plain English and MadMax writes the filters, picks your project connection and runs the read step against your real table during the build so downstream steps are shaped by real rows.

What teams build

• A watch on the profiles table keyed on id fires per new signup, an AI agent with the browser tool reads the company's website for industry and size, and update_rows writes those columns back before Slack #signups gets a one-liner.
• A webhook from your app fires on a new post, an AI classify step sorts it into fine, borderline or remove, and a switch publishes it, sends it to an approval in Slack, or runs delete_rows and emails the author the reason.
• A Sunday schedule runs list_rows on sessions with expires_at=lt.<now> and deletes them, then loops over orphaned uploads deleting each row and its storage object with a web request, and emails the counts.

How it connects

Paste your Project URL (https://your-project.supabase.co) and the project's service role secret from Settings, API in the Supabase dashboard. MadMax sends it as both the apikey header and the bearer token, the way PostgREST expects. The service-role key bypasses Row Level Security, so it is stored encrypted, used only server-side at run time and never rendered back into the UI. Filters use PostgREST syntax (status=eq.active, email=ilike.*@acme.com), and update and delete require one, so a workflow can never touch the whole table by accident. See managing connections.

Related

• Watch trigger
• AI classify
• Approval step
• Managing connections
• Postgres