RAG (retrieval-augmented generation)
ChatbotsDefinition
RAG (retrieval-augmented generation) is an AI pattern where a model retrieves relevant passages from a trusted source, like your knowledge base, and generates its answer from them. It grounds AI output in your actual facts, cutting hallucinations and keeping answers current without retraining the model.
The pipeline: index your documents, retrieve the passages relevant to the question, and hand them to the model as context for the answer. The model supplies language skill; your library supplies truth.
RAG quality is retrieval quality. Clean, current, well-chunked documents produce grounded answers; a stale or contradictory library produces confidently wrong ones. Maintaining the source is the real work.
How Savra puts it to work
Savra’s chatbots are RAG systems: they answer from your uploaded knowledge base, in your brand voice, with strict tenant isolation, so the bot cites your policies rather than inventing them.
People also ask
Why use RAG instead of fine-tuning?
RAG updates by editing documents and can cite its sources; fine-tuning bakes knowledge in, goes stale, and cannot point to where an answer came from. For factual Q&A, RAG wins.
Does RAG eliminate hallucinations?
It reduces them sharply by anchoring answers to retrieved text. Retrieval misses can still produce errors, which is why source curation matters.
Related terms
Where this lives in Savra
See how Savra puts rag (retrieval-augmented generation) to work.
Does your brand sound like itself?
Run the free 90-second Brand Genome audit and see where your marketing drifts off-voice.