lorebit
  • English

      • RAG workflow

      CORE CONCEPT · RETRIEVAL IS A FLOW, NOT A TABLE

      RAG is lorebit's primary product path: a question becomes an evidence-backed answer through a sequence that preserves source identity and makes each decision inspectable.

      The path from material to an answer

      01Ingest

      Collect approved material with source ownership and version information.

      02Normalize

      Turn it into addressable units without losing the link to the original.

      03Retrieve

      Use the capabilities needed for a question: semantic, lexical, or structured.

      04Ground

      Build context and citations that let a reader verify the answer.

      Reranking, filtering, and context assembly belong in this flow because they affect what evidence reaches the answer. A vector index can help with one step; it is not the system boundary.

      What a good result preserves

      PropertyWhy it matters
      Source identityA reader can find the original material behind a claim.
      Version awarenessA result can be reconsidered when a source changes.
      Retrieval rationaleOperators can understand why evidence was chosen.
      Citation-ready contextThe final answer can expose evidence rather than merely sounding confident.

      What this does not promise yet

      These docs describe the intended workflow, not a released SDK contract. Names, interfaces, and provider support remain preview material until an implementation and compatibility policy are published.

      Continue with the knowledge lifecycle.