Menu

Use case

Which documents should go into my RAG system?

Score each candidate document against your own evals before it reaches the knowledge base, and export the keepers as a manifest.

The problem

A RAG index is easy to fill and expensive to correct. Help-center pages, policies, and internal docs pile in without evidence, and each wrong document costs retrieval quality on every query it matches.

The workflow

  1. Baseline

    Run your agent over your RAG eval set and record the score.

  2. Value

    Each candidate document is injected into the slices it could affect and re-measured against fresh controls, producing a delta with a confidence interval.

  3. Select

    Kno builds a portfolio under your budget and rejects what does not pay for itself, no-effect, redundant, and harmful documents get a rejection reason instead of a slot.

  4. Export

    Render the keepers as a knowledge-base manifest.

What it looks like

kno
new_refund_policy.md   +18%   keep → knowledge base
example_42.json         +7%   keep → context
example_91.json         +1%   reject
old_refund_policy.md    -9%   reject → harmful

The recipe

Feed the pool from the documents you already have, or from a source you want to test, like a help center, with the Zendesk cookbook recipe. The pool adapter reads JSONL, CSV, or Markdown, so exporting from any system is enough to start.

What you get

A ranked decision per document, keep, reject, or move, with the measured delta behind it. Selected documents export as a knowledge-base manifest your ingestion pipeline can consume directly.

Stages involved

Try it →