Krex AI · Analyzer suite
Document integrity, measured honestly
Six tools that analyze writing — AI detection, grammar, plagiarism, citations, authorship, and rewriting — built on the one principle this category avoids: publish what the detector can do and what it can't.
Try it now — the real engine, no signup
2 free scans a day · 40–1,200 words
0.87
AUC catching unedited AI drafts — the honest use case, measured
1%
false-positive budget on human writing. Not a goal — a design constraint
0
false flags on real non-native-English writing (n=5, growing) — the group detectors hurt most
every number traces to our published evaluation · re-run monthly against current models
What we don't claim
Our detector catches text that reads like an unedited AI draft, and explains every flag. It does not reliably catch AI text that was prompted casually or run through a humanizer — and our testing says nothing affordable does. So a clean scan is reported as exactly that: no AI-draft signals found — not proof of human authorship. If a detection tool tells you otherwise, ask to see the curve.
One suite, six instruments
AI Detection
A per-passage heatmap with the measured signals behind every flag — and the human counter-signals pointing the other way. Signals, never accusations.
Grammar
Discrete, explained corrections — grammar, spelling, punctuation, style — each one quoting the exact text it's about. Apply fixes individually or all at once; your voice stays untouched.
Plagiarism
Your most distinctive sentences searched across the public web, matches shown side-by-side with linked sources. Honest scope: a clean result is not a guarantee of originality.
Citations
Every reference verified against real registries — DOIs resolved, bibliographies cross-checked, fabricated citations exposed before a reviewer finds them.
Authorship
Style consistency across a document, passage by passage — plus a comparison against your own writing sample when you need evidence that a text is yours.
Rewrite
Rework your own writing in five modes — natural, paraphrase, professional, academic, or matched to a sample of your voice — with every change shown as a reviewable diff.
Documents flow between tools — check a draft for AI signals, verify its citations, then polish it in Rewrite without pasting anything twice.
Straight answers
How accurate is the AI detector?
We publish curves, not confidence: 0.87 AUC catching unedited AI drafts, a 1% false-positive budget on human writing set by design, and a full public writeup of what evades detection — including casually-styled and humanized AI text, which our testing shows defeats every method in this category. Every number comes from our evaluation runs, re-run monthly.
Will it falsely accuse a human writer?
It is built so that the cost of caution falls on the catch rate, never on the writer. At our threshold, at most 1% of human documents flag by design, accusatory verdicts require high combined evidence, templated or formal writing gets structural protection, and in our tests real non-native-English writing produced zero false flags.
Does a clean result prove a document is human-written?
No, and we say so on every scan. A clean result means no AI-draft signals were found — current-generation AI told to write casually can evade detection. Any tool claiming otherwise should show you the measurement.
What happens to documents I analyze?
Scans live in your private history so you can reopen them, and you can delete them any time — trash, restore, or purge permanently. Your documents are never used as marketing examples or shared.
The only detector worth shipping is one that knows its limits.
Free to start — 50 credits a month, no card. The methodology is public, the numbers are re-measured monthly, and every verdict shows its work.