Skip to content
Nappai Documentation

Batch processing

Many jobs are the same task repeated over a list: read each invoice, summarise each email, classify each row. In NappAI you do not need a loop: you turn on Batch Mode in the component and it runs once for each item of the list it receives.

In this video, a flow processes the twelve supplier invoices in a folder: it reads each PDF, extracts its data with an agent, checks it with Python, and leaves a spreadsheet and a summary for accounting.

How it works

  • Batch Mode is in the tab of the same name of every component that supports it. When it is on, the component shows the Batch Input input: that is where you connect the list.
  • Each field can be Fixed (the same value for every item) or Mapped: you write a key of the item between braces, for example {path}, and each run takes that item’s value. You can also mix text and keys: {Invoice {number} from {supplier}}.
  • After a run, the Previous execution panel shows a real item with its keys: drag one onto the field and it becomes mapped.
  • Parallel Executions sets how many items are processed at a time (3 by default).
  • The output of a batch component is another list, with one result per item. The next component can carry on in batch or take the whole list, for example to create a spreadsheet or write a summary.

The example in the video

  1. Storage Files, List Files operation: the PDFs in the files/october-invoices folder of your storage. It returns a list.
  2. Storage Files with Batch Mode, Get File operation: File Path mapped to {path} and Content set to Text. It reads each PDF as text.
  3. Agent with Batch Mode and structured output (Output Format tab): from each text it always returns the same fields (supplier, tax ID, number, date, category, net amount, VAT and total).
  4. Python Code with Batch Mode: it checks that the net amount plus VAT gives the total, that VAT is 21 % and that the invoice belongs to the month, and adds a status field.
  5. Two branches with the whole list: XLS Writer creates the spreadsheet and Storage Files (Upload Files) saves it in the same folder; Data to JSON and a Prompt hand every invoice to a second Agent, which writes the summary.

Show node outputs, in a component’s settings, shows its output as a table: one row per item.

Tips

  • Try it first with a few items and look at the output table of each step.
  • Calculations and checks are better in Python Code; leave to the agent what needs understanding the text.