A Chrome extension that records AI writing interactions alongside Google Docs as a TWFF-compatible process log: what was written, when AI assistance was used, and what the author decided to do with each suggestion (accept, modify, or reject). The author exports and submits the log voluntarily — nothing is sent anywhere automatically.
Full format specification: TWFF spec Live demo of the desktop editor: firl.nl/twff
- Records a writing session in Google Docs: edits, pastes, and AI-assisted text, each tagged with how it entered the document.
- Optionally runs local AI writing assistance (paraphrase, grammar, brainstorm) entirely on your own machine via a companion program — no cloud API, no data leaving your computer. This is off by default and fully optional.
- Lightweight writing-quality heuristics (long sentences, filler words, repetition) surface as suggestions in the side panel — a secondary feature, not a replacement for a dedicated grammar checker.
- Exports the recorded session as a
.twfffile (a structured, human- and machine-readable process log) that can be opened in this extension's own viewer or at colophon.firl.nl/viewer.
What it keeps: character counts and edit positions (not your typed keystrokes), paste event metadata (character count and source), AI interaction type/model/acceptance decision, and session timestamps — never your full document text, never individual keystrokes, and nothing is sent to any server unless you explicitly export and share the file yourself.
Not yet published to the Chrome Web Store. Until then, build from source (see below) and load it as an unpacked extension.
git clone https://github.com/Functional-Intelligence-Research-Lab/colophon
cd colophon
npm install
npm run build # produces dist/Load unpacked extension in Chrome:
- Go to
chrome://extensions - Enable Developer mode
- Click Load unpacked
- Select the
dist/folder
During development, use npm run watch instead of npm run build — it rebuilds on every save. Reload the extension in chrome://extensions after each rebuild.
To validate a .twff export:
python tools/validate.py path/to/your-session.twffRead CONTRIBUTING.md.
Apache-2.0. See LICENSE.
Contributions are credited in the repository and in any publications using Colophon as a research instrument.