The new Amagno.AI services (Chat, Autopilot and Automate) enable automated, comprehensive information retrieval, analysis and processing that go far beyond the capabilities of a traditional search.
Amagno.AI services are paid premium features that can be used in the Amagno Business Cloud in Germany. To use them, your Amagno environment must have organisations set up.
Components
Amagnio follows a universal approach in its products. The Amagno.AI Premium Service therefore combines various features, which are divided into three key elements.
Chat
The chat feature allows users to interact with one or more documents, as well as with entire folders or an entire organisation. The user asks a question and Amagno.AI provides the answers shortly afterwards, allowing for follow-up queries.
Autopilot
The autopilot currently provides various data within the document context.
Standard document data
From every document the first 500 characters from the full text are analysed during import. The AI attempts to recognise the following data and automatically organises it into five categories, so that it can be used, for example, in business processes or for navigating the category structure (rather than folders). This data comprises the following:
- Document type (e.g. ‘Contract’)
- Document sub-type (e.g. ‘Lease agreement’)
- Dokumentendatum (das Datum eines Anschreibens, nicht ein Dateidatum)
- Sender (e.g. name of the sending company or person)
- Recipient (e.g. name of the receiving company or person)
Invoice details
If the AI determines from the standard document data that the file is an invoice or credit note, and it is not in an e-invoice format compliant with EN 16931, the AI will attempt to recognise the invoice data that Amagno currently outputs as tags for an EN 16931 document. This includes:
- Comprehensive sender and recipient information
- Standard invoice line items
- Payment information
Automate
With Automate, you can significantly increase the level of automation in the recognition, validation and calculation of tags and their associated values.
To this end, Automate allows you to store commands (prompts) for document types, multi-tag sets and tags.
This allows numerous complex scenarios to be automated using simple formulations, for example:
- A command for a document type can be used to assign document types very flexibly and dynamically; for example, ‘Document is an invoice and the sender is Clean Power AG’ recognises this as the document type for the outgoing invoice of the company in question.
- A command for a multi-attribute set can be used to respond very flexibly to recurring items in a document’s list.
- A command for a specific tag can be used to read a value from a document very flexibly, e.g. contract number, reference number, address and much more. This can also be used to perform validations or calculations.
Automate is a highly effective tool that can perform complex tasks on documents quickly and in natural language.
Technology
Amagno.AI is currently based on language models from the GPT-4.1 series. The data is processed in data centres within the EU, primarily in Germany, via Microsoft Azure in dedicated, secure instances. Consequently, no company data is transmitted to public APIs.
All documents are checked against the user’s permissions prior to AI evaluation. This influences the result individually for each user.
The Amagno.AI services process text, graphics or videos. The processing volumes are technically limited by the language model.
Tokens (Billing)
Amagno.AI is billed monthly using tokens as the unit of measurement. The technology used relies on numerous other cloud technologies, the cost of which is allocated via tokens.
Tokens are a smaller unit than words and are composed of many elements relevant to AI. The OpenAI Tokenizer provides an overview of how tokens are calculated in AI.
Every activity involving the AI consumes tokens, including, amongst others:
- Asking a (follow-up) question
- Receiving a (follow-up) answer
- Processing the necessary text elements from files to compile an answer
- Processing new document content and its data
- Processing existing document content and its data
The Amagno.AI services and the available tokens are billed monthly. Unused tokens are not carried over to the following month. If more tokens are used than have been purchased, the service is paused until the start of the next month. Alternatively, a token quota can be purchased at any time, which should also be used up by the end of the month.
Functionality
Artificial intelligence in modern software systems is generally a straightforward technology with a few distinctive features. The distinctive features of the technology are:
- The AI possesses an enormous body of knowledge, trained on a global scale, across a vast number of languages.
- The AI processes inputs and outputs via a language model. A language model (or Large Language Model (LLM)) is a system that has been trained to understand and generate human language. It is based on algorithms and statistical methods that analyse text and recognise patterns in order to process natural language.
Durch dieses Verständnis erkennt die KI Through this understanding, the AI uses the data it has been trained on to recognise, for almost all documents, what type of file it is and which data within it is important. It can also establish common relationships between documents, e.g. between invoices and delivery notes.die antrainierten Daten zu fast allen Dokumenten, um welche Art von Datei es sich handelt und welche Daten darin wichtig ist. Es kann ebenfalls häufige Beziehungen zwischen Dokumenten erstellen, z.B. zwischen Rechnungen und Lieferscheinen.
Special features
Completeness
In the context of document management, queries are often run to process large volumes of data at once, e.g. ‘Sum all invoices from Company XY for the year 2025’ or ‘Are there delivery notes for all (200) invoice items?’.
For this process, Amagno.AI will find the most important, but not all, document fragments. Due to the quality of the fragments and the token limit per query, not all documents can be taken into account.
Consequently, completeness cannot be guaranteed for queries that involve a large number of documents.
Mathematical accuracy
For technical reasons, an AI cannot perform calculations reliably. Therefore, particular caution is required when using AI for mathematical tasks.
Hallucinations
Hallucinating means that the model generates false or fabricated information which, although linguistically correct and convincing, has no basis in reality. The more up-to-date the language models are and the more they check themselves before providing an answer, the more reliable the results are.