Document Extractor

When you share a file, the agent can:

• Read a PDF and return its text. Contracts, invoices, bank statements, reports, order confirmations: whatever text is in the PDF comes back so the agent can summarize it, find a figure, or answer questions about it.
• Read a Turkish e-fatura (XML, UBL-TR). The full document comes back plus, for a quick read, a summary of the key fields: invoice number, issue date, supplier, customer, currency, tax and totals.
• Turn an e-fatura into structured data. For a Turkish e-fatura (UBL-TR) the skill extracts the invoice precisely into a clean, structured form your agent can act on: every line item (description, quantity, unit price, VAT rate and amount, line total), the supplier and customer with their tax numbers (VKN / TCKN), the per-rate VAT breakdown and withholding (tevkifat), and the totals, with a built-in check that the amounts add up. Because this is read straight from the e-fatura rather than guessed, the figures are exact, so another skill can rely on them to record the invoice.
• Read plain text, JSON, or CSV files you attach and return their contents as-is.
• Recognize an attached image and hand it off. Photos and screenshots are not "extracted" here because your agent can already see images directly. The skill simply notes that the file is an image so the agent uses it from the conversation.
• Feed what it read into your other skills. Because it returns clean text in the same turn, the agent can chain it: read now, act next. This is what makes it useful for real workflows rather than just displaying a file.

1. Open the Skill Marketplace in your Botonom dashboard and find Document Extractor (Productivity category). It is free.
2. Install it, then assign it to the agent that should be able to read documents, for example your Bookkeeper, Operations Helper, or Executive Secretary.
3. That is it. There is nothing to configure. From now on, whenever you attach a file in that agent's chat, it can read it.

To use it, just share the file in the conversation and ask in plain language. The agent picks up the attachment automatically; you never paste links or file paths.

User: [attaches acme-invoice.pdf]
      Can you pull the total and the due date from this invoice?
Agent: This invoice from Acme Ltd is for 4,250.00 USD, issued 12 July,
       payment due 26 July. The invoice number is INV-2044.

User: [attaches efatura.xml]
      Who is the supplier on this e-fatura and how much is it for?
Agent: Supplier: Yildiz Tekstil A.S. Customer: Botonom A.S.
       Invoice GIB2026000001234, issued 14 July 2026, total payable
       11,800.00 TRY (tax 1,800.00 TRY).

User: [attaches efatura.xml]
      Extract the line items, VAT and total as structured data.
Agent: Invoice ABC2026000000999 from Acme Teknoloji A.S. (VKN 1234567890)
       to Botonom Yazilim Ltd. Two lines:
       1) Widget A  x10 @ 100.00  VAT 18% = 180.00  line 1,180.00
       2) Widget B  x5  @ 100.00  VAT 20% = 100.00  line   600.00
       Net 1,500.00 + VAT 280.00 = 1,780.00 TRY payable. Figures reconcile.

User: [attaches expenses.csv]
      Which three line items are the largest in this file?
Agent: The three largest rows are: Cloud hosting 1,920 TRY,
       Office rent 1,500 TRY, and Software licenses 940 TRY.

There is nothing to connect and no key to paste. Some skills need you to link an account (like Google or Business Central) or enter an API token. Document Extractor needs neither, because it only reads files that already live in your workspace conversations.

So the entire setup is: install it from the marketplace and assign it to an agent. Setup takes about a minute. Once assigned, it works the moment you share a file in that agent's chat.

Limits and honest boundaries:

• Read-only. It never creates orders, posts, invoices, or edits. To act on a document, pair it with an action skill (for example Business Central or an orders skill) and the agent does the reading and acting across skills.
• Only your workspace files. It reads files shared in your own conversation and nothing else. It cannot open arbitrary web links or fetch files from outside, which keeps your agent from being used as a fetch service.
• Scanned or photo-only PDFs come back empty. If a PDF is really just a picture of a page with no text layer, there is no text to extract. Automatic text recognition (OCR) for scanned pages is on the roadmap but not available yet.
• Big or very long files are capped. Files up to roughly 20 MB are accepted, and extremely long documents may be trimmed. For most invoices, statements, and reports this is not a concern.
• e-fatura (XML) extraction is exact; PDF is best-effort. For a Turkish e-fatura in XML (UBL-TR) the structured figures are read deterministically from the file, so they are precise and reconcile. For a PDF the skill returns the text and a lower-confidence summary, so glance over PDF figures before relying on them for accounting.

Security and privacy:

• Files are read in the moment and not stored by the skill.
• Access is limited to signed-in team roles (staff, manager, admin, owner). An anonymous visitor in a public chat cannot use it, so your documents are not exposed through a public widget.

On the roadmap: a direct read-to-record combo (your agent reads an e-fatura and, with your confirmation, records it in Business Central or another system), structured extraction from PDFs, and OCR for scanned pages. Structured extraction from e-fatura XML is already live.