Skip to main content

RAG (Retrieval-Augmented Generation)

RAGs provide a configurable "document memory" for the AI. In this ecosystem they are managed from the Portal Admin UI.

Configuration model (high level)​

  • RAG prompt: system instructions used when this RAG is active.
  • Embedding model: model used to vectorize text/documents.
  • Vector store: Qdrant is used as the vector database.

Data ingestion​

Common source types:

  • Files (.pdf, .txt, .md, ...)
  • Folders (server-side paths)
  • Single web pages
  • Sitemaps

After adding sources, an indexing action triggers ingestion.

  • Qdrant: docs/internal/services/data-layer/qdrant.md
  • Portal: docs/internal/services/platform/portal.md