What are Boxes?
Boxes are Harold's way of organising extracted documents by type. Every document Harold processes ends up in a Box. A Box is essentially a smart inbox for a single document category — Purchase Invoices, Sales Invoices, Receipts, Credit Notes, or any custom type you create.
Think of Boxes as the place where extraction results live. DocuTrain teaches Harold what to extract. Automations handles the incoming flow. Boxes is where you see, review, and manage everything that has been extracted.
The four default Boxes
Harold comes with four pre-configured Boxes that cover the most common business document types.
Purchase Invoices — invoices your suppliers send to you. Typically contain supplier details, line items, VAT, payment terms, and due dates.
Sales Invoices — invoices you send to your customers. Harold can extract these too, which is useful if you are reconciling what was issued against what was received.
Receipts — till receipts, expense receipts, and similar. Typically shorter documents with fewer fields but high volume.
Credit Notes — supplier or customer credit documents. Structurally similar to invoices but require separate handling for accounting purposes.
What you see inside a Box
Each Box shows a list of every document of that type Harold has processed. The list shows the supplier or sender name, the document date, the extraction status (Processing, In Review, Approved, Failed), and the confidence level of the extraction.
You can filter the list by sender email, by status, and search by supplier name or document reference.
Selecting and batch processing
You can select multiple documents using the checkboxes on the left of each row. Once you have a selection, you get options to batch-approve, batch-export, or send to Review and Export. This is the workflow for processing a week's worth of invoices in a single session — select all approved ones, export, done.
The Review and Export option opens the full batch review table where you can see all selected documents' extracted fields side by side, make corrections, and export as a structured file.
The document detail view
Clicking a document opens a split-panel view. The left panel shows the original document — rendered from PDF — with navigation controls for multi-page documents. The right panel shows every extracted field alongside its value.
From here you can approve the document, send it to review, mark it as failed, or view the underlying extraction run for debugging purposes. You can also see the extraction method and the overall confidence score.
Ease of use
Boxes is the most straightforward part of Harold. There is nothing to configure — documents arrive in the right Box automatically based on their type. The interface is intentionally list-like because the primary job is triage: see what has come in, check anything low-confidence, approve the rest.
The main workflow friction is multi-page documents where extraction may miss a field on a later page. The PDF viewer makes it easy to cross-reference quickly.
The aim
Boxes exist because extracted data without organisation is just noise. The goal is that every Box reflects the current state of your document processing — you can see at a glance what is pending, what is approved, and what needs attention. Combined with Automations handling the inbound flow and Zapier handling the outbound export, Boxes is the human checkpoint in an otherwise automated pipeline.