Knowledge search (RAG)
How your agent looks up answers in your knowledge base — and when to keep it on.
Knowledge search is how your agent finds specific answers in the content you've indexed. When a visitor asks something the agent can't answer from its system prompt alone, it searches your knowledge base for the most relevant items and uses what it finds to reply. This is the retrieval step behind grounded, on-topic answers — often called RAG (retrieval-augmented generation).
You'll find it as the first item in the Tools card on the agent page, labelled Knowledge Search.
What it does
When knowledge search is on, the agent decides for itself when a question needs a lookup:
- Greetings, small talk, and questions it can already answer from the system prompt don't trigger a search.
- A specific question — about pricing, a particular service, opening hours, "do you do X?" — prompts the agent to search your indexed content and answer from the best matches.
The agent can search more than once for a single question if it has several parts, so a multi-topic question still gets a complete answer.
Why keep it on
Knowledge search is enabled by default, and for most agents it should stay that way. It's what lets the agent answer from your actual content instead of guessing. Without it, the agent is limited to whatever is captured in the system prompt — fine for a fixed FAQ, but unable to reach the depth and specifics held in your wider knowledge base.
What happens when it's off
You can turn the toggle off for an agent. When it's off:
- The agent no longer looks anything up in your knowledge base.
- It answers only from its system prompt and the current conversation.
- It uses fewer credits per reply, since it skips the lookup step.
Turning knowledge search off is rarely what you want for a customer-facing agent — it's the main thing keeping answers grounded in your content. Leave it on unless you have a specific reason to limit the agent to its system prompt.