Route
Assignment
No student work is used here, so drafting a rubric can use the larger cloud model when you're online. Offline, it falls back to the on-device model.
Rubric
Scored 1 (beginning) to 4 (exceeds). Edit anything.
What the class needs next
Built only from anonymized scores and next steps. No essays and no names are used.
Student writing never leaves this computer with a name on it.
Red Pen uses Firebase AI Logic's hybrid inference, and it picks the route on purpose instead of leaving it to automatic fallback.
Student papers
Chrome can run Gemini Nano?Feedback is drafted on this device with
ONLY_ON_DEVICE. Zero bytes are sent. Works offline.Anonymize first, on this deviceStudent and classmate names, teacher, school, emails, phones, IDs, addresses and dates become placeholders like
[STUDENT].Only the safe copy goes to the cloud
ONLY_IN_CLOUD, never automatic fallback, so a real name can't slip through.Names are put back on this deviceThe teacher sees a normal note to Maya. The cloud only ever saw
[STUDENT].Planning work
Rubrics and class insightsNo student data, so these use
PREFER_IN_CLOUD for the larger model when online.Offline? Still works.The same request falls back to the on-device model.
Every result shows a receipt
"Stayed on this device"or "Anonymized, then sent," with a button to see the exact text that left.
Why it matters: student work is protected by FERPA, and many districts bar teachers from pasting it into AI tools. Many school Chromebooks can't run a local model, which is why the anonymized route exists.