What cayDoku can do
Not a chatbot that guesses. A search that truly knows the abas help – and cites its sources.
2,700 pages of help. One structured index.
cayDoku parses the complete abas online help – more than 2,700 pages – and turns it into a structured knowledge base: sections, field help, documents. The dashboard shows live how the knowledge graph is built, deduplicated and enriched.
Four methods. One result. 55 to 160 milliseconds.
cayDoku combines classic full-text search (BM25), trigram search for German compound words – it finds "Bankkontoauszug" even when you search for "auszug" –, semantic vector search with 768-dimensional multilingual embeddings and a knowledge graph boost. Results arrive in 55 to 160 milliseconds.
Over 31,000 technical terms. Almost 100,000 relationships.
The automatically built knowledge graph links more than 31,000 ERP terms with almost 100,000 relationships between them – such as "tax key → used in → ZUGFeRD". That is how cayDoku also finds matches in which the search term itself never even appears.
Ask like you would ask a colleague. Answers with sources.
The RAG pipeline answers questions in everyday language with the matching help sections – structured with title, description, step-by-step instructions and a source link to the original. An answer takes 2 to 12 seconds.
Runs in-house. Always cites the source.
cayDoku runs completely on-premise – no cloud, no third parties. And the AI answers always cite their source: every paragraph can be verified in the original abas help with a single click.
Seconds instead of page-flipping.
A consultant – or, in the future, S.E.L.M.A. – finds the right passage in the abas documentation in seconds instead of working through manual pages. That makes cayDoku the foundation for AI-assisted support around abas ERP.