Import invoices with AI
This is the agent-first heart of Aikount. Instead of typing every purchase invoice by hand, you drop the PDF and the AI agent does the mechanical work: it reads it, extracts the data and books it into your accounting, already categorized and ready to review.
How it works
- You drop the PDF of the purchase invoice from the Purchases section.
- The agent reads the document by OCR, even a scan or a photo.
- It extracts the key fields: supplier, taxable base, VAT and total.
- It books it into the Spanish chart of accounts (a group-6 account), already categorized.
- The expense stays editable: you can correct any field before signing off.
The result isn't a raw draft but an expense already posted to your books, with its accounting account and input VAT computed. You only review.
What the agent extracts
| Field | Where it comes from | What it feeds |
|---|---|---|
| Supplier | Invoice issuer | Linked to a supplier-type contact |
| Taxable base | Amount before tax | The expense account (group 6) |
| Input VAT | VAT amount | Account 472, deductible on Modelo 303 |
| Total | Amount due | The basis for bank reconciliation |
Always editable
The agent makes the first pass; the final judgment is yours. You can:
- Change the accounting account if the expense didn't land where you expected — see Categorization (PGC).
- Correct the base, VAT or rate if the OCR missed.
- Adjust the supplier or create the contact — see Suppliers.
Correcting the agent isn't an exception, it's part of the flow. Aikount does the repetitive part; you bring the judgment.
No duplicates
Each invoice enters only once. If you drop the same PDF twice, or the same expense arrives through two channels, Aikount detects it by the invoice's identity, not by the file. This is a hard product rule: a duplicate purchase would corrupt your expenses, input VAT and bank reconciliation. See Purchase deduplication to understand how it works.
For agents
The same flow is available over the API: an agent or script can send the PDF to the purchase-ingest endpoint and receive the expense already booked and deduplicated. It's the recommended path for bulk loads, because it runs through the same invoice-identity deduplication.