Understand how your students
write with AI.
Colophon logs every edit, paste, and AI interaction as a verifiable timeline, right inside Google Docs. No cloud, no surveillance, no subscription.
Running research case studies this fall: free for one course section, tool + onboarding + a published report included. Apply →
Read the preprint →It's one file: a verifiable record you control.
An inscription placed usually at the end of a book, giving facts about its publication.
A colophon is a section of a book that describes the technical and production details of how the book was created, including printing methods, materials, typography, and contributors.
See it in action
A real look at the extension and the TWFF viewer.
At a glance, in the popup
Click the toolbar icon any time to see the current document's own-writing / AI / pasted breakdown and a running activity feed; no need to open a dashboard.
Lives beside your document
The panel floats right next to the page you're actually writing in; nothing to switch to, no separate app to keep open.
Open any session in the TWFF viewer
Drop a .twff file into the viewer at colophon.firl.nl/viewer, entirely in your browser with nothing uploaded, and get the full annotated, exportable report.
Don't have one? Try our sample essay or download the raw file.
Is Colophon an AI detector?
No. Colophon does not try to guess whether text was written by AI. Instead it gives students and teachers a verifiable, timestamped record of the entire writing process: every keystroke block, paste, and AI interaction. A percentage score can be disputed; a timestamped edit history can be checked against the document itself.
How we work
-
01
Never optimize for surveillance. We record process, not verdicts; no keystroke-content logging, no AI-detection score, no feature that flags a student.
-
02
The author owns their process. A student's writing record is theirs to export, share, or delete, not a log kept from them.
-
03
Evidence beats prediction. Charts show what happened (counts, timelines, composition), never a predicted "quality" or "originality" score.
-
04
Publish what we learn. Real case studies end in a findings report anyone can read, not a private result.
-
05
If a feature needs a "trust us" explanation, don't build it. What's collected, and why, should be visible and checkable, not taken on faith.
In conversation with
Institutions and standards groups we've been talking with about Colophon and the TWFF format.
More as these conversations become formal partnerships.
Standards alignment: our TWFF process-log schema maps onto the W3C Data Privacy
Vocabulary's dpv:AutomatedDecisionMaking concept, a contribution the DPV Community Group
has already accepted.
Try it on one assignment, no district approval needed
Grading a stack of essays and need a quick read on which ones leaned hard on AI, without handing down a score you can't defend? Set up one class for one assignment yourself, free, no IT ticket, no procurement process. Students write with the Colophon extension on their own device; you get a scannable, color-coded view of what was typed, pasted, or AI-suggested in each submission, evidence, not a verdict. This data isn't used for research unless you opt in.
Request access, freeWe set up individual teachers quickly, no institutional sign-off required.
Get early access
Colophon is live on the Chrome Web Store, and we're working toward our first real classroom case study. Join the waitlist to bring your class in, or to get researcher access to the data once a study is actually running.
Join the Colophon waitlistWhat's next
- Getting a first real pilot section through the whole pipeline end to end
- Google Classroom and Canvas integration, so students don't upload files by hand
- Aligning the AI-content icons shown in the viewer with the EU's AI-content-labelling recommendations
- Refining which in-editor writing suggestions are worth surfacing, based on real usage