Contents

The documentation is written in English and Dutch. You are reading the English version.

Retrieval settings

How the passages the model gets to see are chosen, and the one setting that steers it.

Before the model writes, the search picks the passages it may use. One setting under Settings, in the block Retrieval, steers that choice. Its default is measured, not guessed, and most agents never need to change it.

How the passages are chosen

  1. The search ranks your passages by how close they are in meaning to the question, and by the question's own words.
  2. Passages below the Minimum similarity are dropped.
  3. Of what is left, the best 10 go to the model.

If nothing is left after step 2, the model is told that nothing was found, and it says it does not know (or answers from a fact in your instructions).

How many passages

Every agent is given the best 10 passages, and the page the visitor is on adds its own. The number is the same for every agent and is not a setting. Each passage is about 500 tokens of reading. With fewer, an answer spread over two pages, or a price that sits on an overview page rather than on the product's own, was too often left out.

Minimum similarity

How close a passage must be to the question to count at all, from 0 to 1. The default is 0.25.

That looks low, and it is on purpose. Similarity is not a percentage of correctness. In our measurements a passage that is entirely about the question scores around 0.67, and a relevant paragraph inside a broader page scores around 0.29. A threshold of 0.5 would throw away most good answers.

  • Raise it a little (0.30 to 0.35) when the agent answers questions it should have declined, using passages that are only vaguely related.
  • Lower it (0.15 to 0.20) when the agent says "I don't know" about things your sources plainly cover. This happens most with very short questions and with questions in another language than your sources.

Change it in small steps and test with the same questions each time.

What this setting does not change

Confidence is measured the same way on every agent, always over the same number of candidates, whatever the minimum similarity is. Moving the slider never changes that score by itself, and the score is one of the measures the Overview uses to choose the questions in its tips. That is deliberate: a number you can change by turning a knob tells you nothing.

Before you touch it

Most answers that miss are not a settings problem. If the information is not in your sources, no setting finds it. Check Improving your answers first. Settings are the last step there, not the first.