Menu

Explanation

Roadmap

What ships today, what is planned, and where the project deliberately goes next, including the debt ledger, which is public on purpose.

Current status

Generated, not written: every table here is read from docs/status.json in the Kno repository, as of v0.1.5.

Stages

A Stage is a step of the measurement pipeline. status is a tri-state: a Stage that ships only part of its job says partial and says what it excludes.

StageWhat it doesStatusMilestone
baselineRun the agent over your dev cases, score against your goal, persist every result. A holdout is sealed here; nothing reads it until validate.shippedv0.1
valueRoute each asset to the slices it could affect, inject it, re-measure against fresh controls, record the delta with a confidence interval.shippedv0.1
selectBuild a portfolio under budget with a rejection log; every keep/reject decision at a Bonferroni-corrected interval.shippedv0.1
validateMeasure the portfolio as a set against the sealed holdout.shippedv0.2
exportRender selected assets into the destination grammar: context pack, knowledge-base manifest, or tuning-set JSONL.shippedv0.1

Commands

Stages and commands are different lists on purpose. Every command below is registered in the CLI at this ref; the ones that are not Stages are marked.

  • kno baseline
  • kno demonot a Stage
  • kno doctornot a Stage
  • kno eval inspectnot a Stage
  • kno export
  • kno initnot a Stage
  • kno judge calibratenot a Stage
  • kno minenot a Stage
  • kno purgenot a Stage
  • kno reportnot a Stage
  • kno select
  • kno validate
  • kno value

Goals

A Goal is what a Case is scored against.

  • exact-match

Agent adapters

Which agent runtimes Kno can measure against, and what each one can tell it about tokens and spend.

SchemeAvailableToken countsSpendsNotes
anthropic:yesyesbills yourequires --max-output-tokens
bedrock:yesyesbills youConverse; requires --max-output-tokens; AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION
exec:yes—freeruns a local command per Case; free unless --cost-per-call-usd is set
fake:yesyesfreelocal, deterministic, makes no network call
openai:yesyesbills youOpenAI and any compatible endpoint via --base-url
tuned:not yet—freefine-tuned outputs land with the tuner
vertex:yesyesbills yourequires --max-output-tokens; GOOGLE_APPLICATION_CREDENTIALS, or GOOGLE_CLOUD_PROJECT + GOOGLE_CLOUD_REGION

Debt ledger

65 of 116 ledger entries are open — accepted debt with a repayment trigger and no disposition yet. Publishing the count is the point; several open entries make good first issues. Provider price table: 2026-08-28.

Everything above this line is generated. Kno’s repository produces docs/status.json from its own command tree and enum, gates it in CI, and this page fetches it at a pinned release ref. When validate ships, this page changes because the code changed, not because someone remembered to edit it.

Everything below is written by hand, because it is judgement rather than fact: what a milestone is for, and how to get involved.

What is next

Milestone Scope
v0.2 validate (holdout measurement); knowledge-injection writable KBs; redundancy detection; judge calibrate; Tuner and the proxy-FT bridge
v0.3 serve, plugin system (Ring 2 exec plugins), SDKs, OTel export
v0.4 Transfer prior

Alongside the agent adapters listed above, v0.1 also ships eval adapters for Langfuse, LangSmith, and Braintrust; Hugging Face evals and Pools; CSV and Markdown Pools; and the budget guard with resumable runs. Those are not in the generated tables yet — the status artifact describes agent adapters only.

The full architecture, and what is deliberately out of scope: DESIGN.md.

Debt is public on purpose

The open-entry count above is read straight from the debt ledger, where every accepted piece of debt carries a repayment trigger and an owner. When something here looks stale, check the ledger; when a debt is repaid, the ledger is where it shows.

Help define what comes next

← All docs