Search Settings (RAG)

Knowledge fragments per question

How many pieces of your knowledge base the agent reads for each question. The field goes up to 20; a plan may allow fewer.

1-2: Precise, focused answers. Good for simple questions.

3: Balanced — the default.

4 and more: More context, but may include irrelevant info.

Relevance strictness

Minimum similarity score (0.0 to 1.0). Only pieces above this score are used.

0.1-0.2: Loose matching. More results, may include weakly related.

0.3: Balanced — the default.

0.5+: Strict. Only highly relevant, may miss some answers.

Example:

Customer asks: "What are your prices?"

With 3 fragments per question and strictness 0.3, the agent reads the 3 most relevant pieces about pricing.

If threshold is too high (0.7), it might not find anything and agent says "I don't know".