Kno v0.1: the measurement loop is complete

Kno v0.1.1 is out. The measurement loop is complete: you can baseline an agent, value a pool of candidate assets, select a portfolio under budget, export it to a destination, and read the whole story in a report, with the caveat that nothing is validated on holdout yet. validate is the next stage.

What shipped:

• baseline, run the agent over your dev cases, score against your goal, persist every result.
• value, route each asset to the slices it could affect, inject, re-measure against fresh controls, record the delta with a confidence interval.
• select, build a portfolio under budget with a rejection log; every decision at a Bonferroni-corrected interval.
• export, render selected assets into context packs, knowledge-base manifests, or tuning-set JSONL.
• report, one page: verdicts, portfolio, gaps, and the caveat.
• mine, turn production transcripts into an eval set, with weak labels marked as derived and PII warnings built in.
• Adapters, Langfuse, LangSmith, and Braintrust evals; Hugging Face evals and pools; CSV and Markdown pools; the exec:, bedrock:, and vertex: agent adapters; a Zendesk recipe and more in the cookbook.
• Budget guard, estimate, confirm, checkpoint; caps are enforced before the call, and interrupted runs resume without paying twice.

The default agent is fake:, a deterministic local agent that costs nothing, so the whole loop can be seen working before you point it at anything that bills you. The deltas read zero because the fake answers every case with what the case expects. That is the point: it proves the loop, not the asset.

Install:

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

Then follow the first baseline walkthrough.

Kno is early. The honest status, the roadmap, and every accepted piece of debt, with repayment triggers, are public.