AI engines: choose who reads each type of document
In Data Alchemy each document model is paired with a single AI engine, called via API. You can choose Claude (Anthropic), the engine we recommend by default; GPT (OpenAI); a compatible LLM through the European inference providers Scaleway (Generative APIs) and Seeweb Regolo.ai, on a subscription you take out directly; or Data Alchemy AI, our proprietary model on Data Alchemy's own servers and GPUs in Italy.
99.8% accuracy measured with Claude, the recommended engine. We do not publish an accuracy figure for the other engines.
One AI engine per document model
The AI engine is the language model that reads the document following the structure and instructions of the document model. Data Alchemy uses a dedicated-LLM architecture: the engine is chosen per model, for example the invoice model with Claude, the delivery note model with GPT and the order model with Data Alchemy AI, or several models with the same engine.
Each document is processed by a single engine, the one paired with its model. Claude is the engine we recommend by default: it is the one with which we measured 99.8% accuracy. We do not publish an accuracy figure for the other engines. The architecture is described on the Technology page.
How the AI engine is chosen and how it works
The choice is made in the document model; from then on every document for that model is sent via API to the paired engine, which returns structured data.
- 01Document modelJSON field structure and reading instructions for one type of document.
- 02Paired engineClaude, GPT, a European provider or Data Alchemy AI: one per model.
- 03Reading via APIThe document is processed by the chosen engine according to the model's structure and instructions.
- 04Structured dataThe extracted fields come back into Data Alchemy, ready for checks and for sending to the ERP.
1. Choose the engine in the model
For each document model you set the AI engine: Claude, GPT, a compatible model through Scaleway or Seeweb Regolo.ai, or Data Alchemy AI.
2. For European providers, your own subscription
With Scaleway (Generative APIs) or Seeweb Regolo.ai the customer takes out the subscription directly with the provider and chooses a compatible LLM.
3. One engine per document
Each document is processed only by the engine paired with its model: there is no parallel reading by several engines on the same document.
4. Change engine whenever you like
The engine is a model setting: different models can use different engines, or the same engine for all of them.
Where processing takes place depends on the engine: details on hosting, DPAs and sub-processors are on the Security and compliance page.
Four options, one engine per model
The engines you can choose from for each document model. Each document is processed only by the engine paired with its model: no parallel processing by several engines on the same document.
Want to try Data Alchemy AI on your documents?
Data Alchemy AI is Data Alchemy's proprietary model, on Data Alchemy's own servers and GPUs in Italy: with Data Alchemy AI, processing stays in Europe. Sign up to test it with our team.
Data in Europe, processing depending on the engine
Hosting and storage
Uploaded documents, extracted data, backups and logs stay in Europe, whichever engine you choose.
Data Alchemy AI and European providers
With Data Alchemy AI and with the European inference providers Scaleway and Seeweb Regolo.ai, AI processing also stays in Europe.
Claude and GPT
With Claude and GPT, documents are processed by the respective providers, Anthropic and OpenAI, under a DPA.
Which engine for which need
Highest measured accuracy: Claude
Claude (Anthropic) is the engine we recommend by default and the one with which we measured 99.8% data extraction accuracy.
A third-party alternative: GPT
GPT (OpenAI) can be selected for an individual document model, for example for a specific document type.
Processing in Europe with your own provider
With Scaleway (Generative APIs) or Seeweb Regolo.ai you use a compatible LLM on your own subscription with the provider, and processing stays in Europe.
The proprietary model in Italy: Data Alchemy AI
Data Alchemy AI runs on Data Alchemy's own servers and GPUs in Italy: processing stays in Europe. You can sign up to test it with our team.
AI engines and licences
The engine is chosen per document model; the number of models depends on the licence. For details on your licence, plan a call.
Engine per model
Tailored models, each with its own AI engine. The engine is chosen model by model.
Engine per model
Models sized to your needs, each with its own AI engine. The engine is chosen model by model.
1 model
One document model, paired with an AI engine. European providers require the customer's own subscription with the provider.
1 predefined model
10 days and 10 documents with 1 model chosen from the predefined ones, to try AI reading on your documents.
Compare the licences on the pricing and licences page or request a quote.
Related features and pages
- Data Alchemy AIData Alchemy's proprietary model, on servers and GPUs in Italy: sign up to test it.
- The dedicated-LLM technologyThe architecture that pairs an AI engine with each document model.
- Data extraction accuracyHow we measured 99.8% with Claude.
- Security and complianceHosting in Europe, DPAs with AI providers and processing by engine.
AI engines: frequently asked questions
Which AI engines can I use with Data Alchemy?
For each document model you choose an AI engine: Claude (Anthropic), the engine we recommend by default; GPT (OpenAI); a compatible LLM through the European inference providers Scaleway (Generative APIs) and Seeweb Regolo.ai, on a subscription taken out directly by the customer; or Data Alchemy AI, Data Alchemy's proprietary model, on its own servers and GPUs in Italy.
Can I use several engines on the same document?
No. Each document is processed by a single engine, the one paired with its model. Different models can use different engines, though: for example the invoice model with Claude, the delivery note model with GPT and the order model with Data Alchemy AI.
How accurate are the AI engines?
With Claude, the engine we recommend by default, we measured 99.8% accuracy. We do not publish an accuracy figure for the other engines.
How do the European inference providers work?
With Scaleway (Generative APIs) or Seeweb Regolo.ai the customer takes out the subscription directly with the provider and chooses a compatible LLM, which Data Alchemy calls via API for the document models it is paired with. Processing stays in Europe.
Where are my documents processed?
Hosting, storage and backups stay in Europe. With Data Alchemy AI and the European providers, processing also stays in Europe; with Claude and GPT, documents are processed by the respective providers under a DPA.
What is Data Alchemy AI?
It is Data Alchemy's proprietary AI model, on Data Alchemy's own servers and GPUs in Italy, available as an engine for document models. On its dedicated page you can sign up to test it with our team.
Try AI reading on your documents
Activate the free Demo to see AI reading at work, or plan a call to choose the right engine for each document model.