Value your Zendesk knowledge

This tutorial follows the repository recipe Value your Zendesk knowledge , the pattern transfers to any support stack that can export transcripts and knowledge content.

What you will learn

• Building evals from production transcripts with kno mine
• Valuing real help-center content, not toy examples
• Selecting under budget and exporting for ingestion

1. Export from Zendesk

Export your help-center articles (each article becomes one asset in a pool) and historical support tickets with resolutions (each becomes a case).

2. Mine the transcripts into cases

kno mine --logs transcripts.jsonl --format jsonl-chat --mode immediate

Weak labels are marked derived so provenance survives ingestion. Add --review for a durable manifest of what needs human checking.

3. Baseline

kno baseline --evals mined.jsonl --agent openai:gpt-4.1 \
  --max-cost-usd 2.00 --yes

4. Value the help center

kno value --evals mined.jsonl --pool help-center.jsonl \
  --baseline-run-id <run id> --agent openai:gpt-4.1 \
  --max-cost-usd 5.00 --yes

Each article gets a delta with a confidence interval: the policy that resolves refund questions, the article that changes nothing, the outdated page that actively hurts.

5. Select and export

kno select --value-run-id <value run id> --pool help-center.jsonl
kno export --select-run-id <select run id> --pool help-center.jsonl --destination knowledge_base

The export is a manifest your ingestion pipeline can consume, Kno decides what belongs in the knowledge base; your stack does the moving.

Full recipe with export-format details: the cookbook entry.