Getting started

This tutorial walks the whole loop against the free local agent: no API keys, no cost, and nothing leaves your machine. You will finish with a working understanding of every stage.

1. Install

curl -sSfL https://raw.githubusercontent.com/uknoAI/kno/main/install.sh | sh

The script verifies a SHA-256 checksum (and a cosign signature if you have cosign) before placing the binary on your PATH. Alternatives: Homebrew (brew tap uknoAI/homebrew-tap && brew install kno) or go install github.com/knograph/kno/cmd/kno@latest. See Install Kno.

2. Write some cases

A case is one scoreable interaction, one JSON object per line, each with a stable id:

{"id":"refund-01","input":"How do I get a refund?","expected":"Refunds are processed within 5 business days."}
{"id":"ship-01","input":"When does my order ship?","expected":"Orders ship within 1 business day."}

Save this as cases.jsonl. Real projects usually mine these from production transcripts, see the mine how-to.

3. Measure the agent as it is today

kno baseline --evals cases.jsonl

The baseline is the reference every later number is compared against. It also seals a holdout: a slice of your cases that nothing, not valuation, not selection, is allowed to read until validate. That separation is why any number Kno reports later means anything.

4. Write a pool of candidate assets

The data you are considering adding, one JSON object per line:

{"id":"refund-policy-v3","content":"Refunds are processed within 5 business days.","kind":"knowledge"}
{"id":"refund-example-17","content":"Example: a 30-day-old refund request is declined.","kind":"knowledge"}
{"id":"brand-guide","content":"Use sentence case everywhere.","kind":"knowledge"}

Save as pool.jsonl. Pools also load from CSV (csv:pool.csv) and Markdown (md:dir/), a bare path always means JSONL.

5. Value them

kno value --evals cases.jsonl --pool pool.jsonl --baseline-run-id <run id from step 3>

Each asset is injected into the slices it could affect and re-measured against fresh controls. The output is one row per asset, a delta with its 95% confidence interval, plus a control reading.

The deltas read 0.0000 here because the free local agent answers every case with what the case expects. That is the point: this run proves the loop, routing, injection, controls, intervals, before real money is involved.

6. Select and export

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

select builds a portfolio under budget with a rejection log; export renders the selected assets into a destination grammar (context pack, knowledge-base manifest, or tuning-set JSONL).

Next

• Point Kno at a real provider when the free path has shown you the loop.
• What the numbers mean before you put a delta in a slide.
• The repository’s full walkthrough and cookbook index.