Document models: what to read and how to read it, on your own terms
Every type of document — invoice, delivery note, customer order — is read by Data Alchemy according to a document model. A model is both a JSON document, with the structure of the fields to extract, and a natural-language text with the reading instructions: reading is accurate and you can edit your models yourself. To get started, pick a model from the marketplace and customise it.
Model marketplace available now. Basic licence: 1 model; Business: models sized to your needs; Enterprise: tailored models. The Demo uses 1 model chosen from the marketplace's predefined ones.
What a document model is
A document model has two parts. The first is a JSON document describing the structure of the data to extract: header fields, nested fields such as the supplier's details, and line items. The second is a text written in natural language, in Italian or another language, with the reading instructions: how to interpret a field, which format to return it in, what to do in special cases.
The instructions are written the way you would explain them to a colleague, not in a programming language: so whoever knows the documents can fix or extend a model on their own, without waiting for a developer. The model is also where you choose the AI engine that reads it and where you configure external fields and remote validation, on the licences that include them.
The model marketplace is available now: rather than starting from scratch, you pick a ready-made model and customise it for your documents. The predefined models you choose from in the Demo are the marketplace's models.
From the marketplace to a document read
You pick a model, adapt it to your documents and from then on every document of that type is read according to its structure and instructions.
- 01Pick from the marketplaceStart from a ready-made model for the type of document you want to read.
- 02CustomiseAdapt the JSON field structure and the natural-language instructions to your documents.
- 03Choose the AI enginePair the model with a single AI engine: Claude, GPT, a European provider or Data Alchemy AI.
- 04Document readingEach document for the model is read according to its structure and instructions and returned as structured data.
1. Pick a model from the marketplace
The marketplace collects ready-made models for recurring documents: you choose the one closest to your document instead of starting from a blank page.
2. Adapt the field structure
In the JSON document you add, remove or rename the fields to extract, for example to match the names used in your ERP.
3. Write the reading instructions
In the natural-language text you explain how to handle special cases: date formats, VAT number prefixes, lines spread over several pages. In Italian or another language.
4. Adjust it yourself when needed
If a supplier changes layout or you need an extra field, you edit the model: structure and instructions are yours, with no development requests.
Data read according to the model can then be verified against your systems and sent to the ERP: see the AI agent and ERP integration.
What it looks like in the platform
The editor of a model for delivery notes (DDT), as the customer sees it: the same information can be edited from the tabbed view or directly in the JSON.
A model at a glance: JSON structure and instructions
A simplified, easy-to-read model for supplier invoices, complementing the screenshots above: on the left the structure of the fields to extract, on the right the reading instructions written in natural language. Field names and instructions are an example: every model is adapted to the company's documents and ERP.
{
"number": "",
"date": "",
"supplier": {
"company_name": "",
"vat_number": ""
},
"taxable_amount": 0,
"vat": 0,
"total": 0,
"lines": [
{
"item_code": "",
"description": "",
"quantity": 0,
"unit_price": 0
}
]
}- Read the supplier invoice and fill in the fields of the structure.
- Write the date in DD/MM/YYYY format.
- If the VAT number has a country prefix (for example IT), remove it.
- Quantities and amounts are always numbers, with no currency symbol.
- Return one line per item, even when the invoice runs over several pages.
Illustrative example: it does not correspond to a specific marketplace model.
What you can do with document models
Get started straight away with a marketplace model
Pick a ready-made model for your document type and use it as it is or customise it: no configuration project from scratch.
Align the fields with your ERP
The JSON structure defines exactly which fields to extract and what to call them, so the data arrives already in the shape your ERP needs.
Handle special cases with written instructions
A supplier who writes dates differently or a document with lines over several pages is handled with a sentence in the instructions, not with a new template.
One model per document type
Invoices, delivery notes, customer orders, price lists: each document type has its own model, with its own AI engine and, on the licences that include them, its own checks against external systems.
How many models on each licence
The model marketplace is available now. The number of models depends on the licence:
Tailored models
Tailored models and storage, with everything in the Business licence, the native Salesforce integration, Order Dispatcher and the AI agent.
Models sized to your needs
Models and storage sized to your needs, with external fields and remote validation configurable in the models.
1 model
One document model and 500 MB of storage, with pay-as-you-go credits bought online: 1 credit = 1 document.
1 predefined model
10 days and 10 documents with 1 model chosen from the marketplace's predefined ones, in a dedicated environment where your data is never shared.
Compare the licences on the pricing and licences page or request a quote.
Related features and pages
- AI enginesThe engine that reads each model: Claude, GPT, the European providers or Data Alchemy AI.
- External fieldsModel fields checked against the external data the document refers to, such as the purchase order.
- Remote validationItem codes checked and customer or supplier code retrieved from master data.
- Data extraction accuracyHow reading accuracy is measured and the 99.8% measured with Claude.
Document models: frequently asked questions
What is a document model in Data Alchemy?
It is the definition of how to read a type of document. It has two parts: a JSON document with the structure of the fields to extract and a natural-language text, in Italian or another language, with the reading instructions. Every uploaded document is read according to its model.
Can I edit the models myself?
Yes. Structure and instructions are designed to be edited by the customer: fields are changed in the JSON and special cases are explained with a sentence in the instructions, with no code to write.
What is the model marketplace?
It is Data Alchemy's collection of ready-made models, available now: customers pick a model from the marketplace and customise it for their own documents. The predefined models you choose from in the Demo are the marketplace's models.
Which language are the reading instructions written in?
In natural language, in Italian or another language: the instructions are written the way you would explain them to a colleague who has to read the document.
How many models can I use?
The Basic licence includes 1 model; on Business models are sized to your needs and on Enterprise they are tailored. The Demo lets you try 1 model chosen from the marketplace's predefined ones.
Start from a model and read your documents
Activate the free Demo to try reading your documents with a predefined marketplace model, or plan a call to build the models around your flows.