Brand voice enforcement
Marketing copy that matches your editorial principles on the first try, every time.
Answer qualityEarly access
Every answer, graded against your standard.
Write a rubric in plain English. A separate grader scores each answer in its own context, and the agent iterates until it passes — without anyone having to review every attempt.
No card · one-click signup · unsubscribe anytime
The problem
Enterprise AI fails in the long tail: the answer looks fine, but it's off-brand, missing a disclaimer, or leaks a name it shouldn't. Manual QA doesn't scale. Auto-evaluation pushes that judgment into the workflow — the agent self-grades against your rubric and keeps going until the output is good enough to ship.
How it works
Four steps from zero to running in production.
- Step 1
Define
Write a rubric in plain markdown: what good looks like, per criterion. Reuse across sessions or attach per-task.
- Step 2
Grade
A separate grader evaluates each draft in an isolated context window, so it can't be influenced by the writer's reasoning.
- Step 3
Iterate
Failed criteria come back as concrete gaps. The agent revises and retries — up to 20 cycles per outcome.
- Step 4
Deliver
Only answers that clear the rubric are returned to users. Every attempt is logged for audit and analysis.
What you get
Built for the way teams actually ship work.
- Rubrics are plain markdown — no DSL, no code
- Per-criterion pass/fail with concrete gap explanations
- Isolated grader context resists prompt-injection gaming
- Configurable iteration budget per task (default 3, max 20)
- Works for chat replies, drafted documents, slide decks, spreadsheets
- Full evaluation log feeds your admin dashboard
Real scenarios
Where it earns its keep.
Policy compliance
Support replies that always cite the correct policy, never contradict it, and flag escalations.
PII and security
Drafts that must not contain personal data, internal URLs, or unreleased product names — checked automatically.
Structured deliverables
Postmortems, status reports, customer-facing docs that must hit every section the template requires.
Early signal
What the numbers say.
Benchmarks from the underlying platform research and early-customer pilots. Your mileage will vary with scope and setup.
up to +10 pts
Task-success lift
+8.4%
Higher docx quality
+10.1%
Higher pptx quality
Frequently asked
Things people ask us.
Who writes the rubric?
Whoever owns the quality bar. We ship starter rubrics for common tasks (support replies, release notes, status updates) that your admin can fork and customize.
What if the rubric is too strict and the agent loops?
Every outcome has an iteration cap. If the agent can't pass after the configured attempts, the latest draft is returned with the gaps surfaced so a human can finish it.
Does the grader see my raw data?
The grader runs inside your Knoq tenant, against the same compliance boundary as every other agent call. No data leaves your environment to be evaluated.
Can I combine it with Company Brain?
Yes — and you should. Brain gives the agent your context; the rubric gives the agent your standard. Together they close the loop between 'what to say' and 'how to say it'.
Get early access
Be first on Answers that grade themselves.
We’re onboarding a small group before general release. Tell us a bit about your team and we’ll reach out when the next slot opens.
Keep exploring
One ask, a team of specialists
A coordinator breaks a hard question into pieces and hands each one to a focused specialist. Results fan in, synthesised, in seconds.