RAG (Retrieval-Augmented Generation)

RAG (Retrieval-Augmented Generation) is the AI pattern behind the Papyrus Copilot. It combines two steps:

1. Retrieval: Given a question, semantic search finds the most relevant document chunks from the tenant's corpus
2. Generation: A language model synthesises an answer using only those retrieved chunks as context, with citations back to the sources

RAG addresses two limitations of pure language models:

• Knowledge cutoff: language models don't know about your specific documents unless retrieval brings them in
• Hallucination: by constraining the model to answer from retrieved context, made-up answers become detectable (the cited source doesn't actually contain the claim)

RAG quality depends heavily on retrieval. If the right chunks aren't retrieved, the model will either decline to answer or guess. Papyrus's hybrid retrieval (keyword + semantic) is tuned to maximise the chance of retrieving the right context.

RAG also strictly respects RBAC — only documents the user can access are eligible for retrieval, so the Copilot never reveals what the user couldn't otherwise see.

See also

• Semantic Search
• Asking the Copilot Questions About Your Documents
• RAG vs Keyword: Picking the Right Search for Your Org