Retrieval & RAG

Overview

WisebotAI uses retrieval-augmented generation (RAG): for each customer turn, the system may search your knowledge chunks (vector similarity + filters) and inject relevant context into the model before generating a reply.

Search tool vs automatic RAG

• Automatic RAG — Context is pulled based on agent and org configuration (chunking, folders, query rewrite).
• Search tool — The builtin:search tool lets the model explicitly query knowledge when phrasing requires it (see Built-in AI tools).

Configuration concepts

The retrieval pipeline applies chunking, embeddings, query rewrite, and folder filters so answers respect the right subset of documents.

Quality tips

• Keep titles and headings clear on crawled pages—chunk boundaries often follow structure.
• Remove stale documents to reduce contradictions.
• Use the agent chat playground to verify grounding on tricky questions.

Privacy

PII redaction and encryption may apply in your deployment—coordinate with your admin for regulated data.