Extract Invoice Fields from Dropbox PDFs into Supabase

This workflow turns a Dropbox folder into a hands-off invoice intake pipeline. Whenever a new PDF lands in the watched folder, it downloads the file, sends it to an OpenAI model with structured-output instructions, and parses out the fields you care about — vendor name, invoice number, issue and due dates, line items, subtotal, tax, and total. The validated result is inserted as a typed row in a Supabase table, ready for reporting or reconciliation. Files that fail extraction are flagged rather than silently dropped.

Reach for this when finance or ops receives invoices as PDFs and someone is currently retyping them into a spreadsheet or ERP. It is ideal for accounts-payable teams, bookkeeping firms handling many clients, or any back office that needs every incoming document captured as queryable data within seconds of arrival. It also works well as the first stage of a larger approval or payment flow, since the Supabase row becomes the single source of truth other automations can build on.

The workflow is triggered by Dropbox's file-added event on a specific folder. The new file is downloaded as binary, then passed to OpenAI with a JSON schema describing the invoice shape, so the model returns strict structured output instead of free text. A logic step validates required fields (invoice number and total must be present and numeric) and normalizes dates and currency. Valid records are inserted into a Supabase `invoices` table via the service-role client; records that fail validation are written to a `invoices_review` table with the raw extraction so a human can correct them. Because it runs on Hive's Sim engine, each colony processes its own documents in isolation with no shared state.