Piloted at the University of Maine, Spring 2026

Make AI use a social discussion.

A shared, safe space to use AI with faculty, TAs, and classmates. Work out together where it helps, where it doesn't, and when not to use it at all.

Course channels with an @ai anyone can call on in front of everyone — so AI use gets discussed, critiqued, and learned together.

Try the demo coursesign in with Google

No setup. Sign in with your institutional (.edu) address and you land in a shared demo institution with a sample course. Request your own institution →

Already have an account? Sign in here
A shared conversation with a measured footprint Three different people and an AI take turns in one thread, shown as overlapping speech bubbles that alternate sides. Beside the thread, a stack of dashes grows taller after each AI reply, next to a tree for scale. AI AI
One shared conversation, with its footprint measured in the open.

1The platform

It's a human collaboration platform first

Channels, threads, notifications, and search, like any team chat. @ a person to collaborate in real-time. @ai to use AI in a shared space, which enables critique and discussion by default.

Ask in the open A class channel. A student asks @ai for an outline. The AI reply appears in the channel for everyone, with its energy estimate. A second student replies to the first, offering to help proofread, directly beneath it, and the class total at the top right gains one more dash. University of Maine ENG 201 · Persuasive Writing # general # peer-review # workshop-op-ed # maya-and-instructor PHI 232 · Ethics of Tech # workshop-op-ed Class total Maya R. 2:41 PM @ai Draft an outline for my op-ed on campus food waste. AI Claude 1. Hook with the dining-hall waste audit. 2. Argue trayless dining. 3. Rebut the cost objection… est. 0.4 Wh claude-opus-5.5 · 1,204 tokens · 1.8 s Devon L. 2:58 PM @Maya R Sam and I can give you feedback before you publish Message # workshop-op-ed
A real classroom move: the AI answers in public, and a classmate responds to the person who asked.

  • Visible by default

    AI answers land in the channel, next to the discussion — so a good prompt gets borrowed and a bad answer gets corrected out loud.

  • Explained, not assumed

    Every AI option says what it does in plain language, right where you switch it on.

  • Yours to leave

    Work in private channels, hide a sentence from the AI, or opt out of AI reading your posts entirely.

2Environmental impact

Discuss and critique environmental impacts

Every AI query carries energy and water estimates, and every member sees their own and the class's running totals.

  • Per-reply estimates — est. 0.4 Wh — sit next to the token count, so the footprint is part of the same conversation.
  • “Is this question worth its energy?” becomes a class discussion based on data.
Environmental impact your AI usage · this term
CO₂ Emissions
182 g
Energy Used
0.42 kWh
Water Used
1.6 L
Your AI usage is equivalent to…
  • 🔌 charging a smartphone 34 times
  • 🎬 watching a 1-hour Netflix movie, twice

Class pledges to offset72%

72%
estimates computed from usage logs · method and sources (details are shown in the app)

3For the course

Shape the AI to the course

AI use guidelines, course materials, purpose-built assistants — set by instructors, and with added guidelines from students in channels they create.

Guidelines that nudge, then get out of the way

Pause before you send A student has typed a message asking the AI to write a conclusion paragraph. A warning appears: this might go against the AI use guidelines, which say the AI may critique a draft but may not write it. The student chooses "Ask a human for help" instead of "Send anyway", and the message arrives with the instructor, Dr. Nelson, rather than with the AI. @ai write my conclusion paragraph for the op-ed Heads up! This might go against the AI use guidelines “AI may critique your draft; it may not write it.” Cancel Ask a human for help Ask a human for help Send anyway Dr. Nelson instructor

“Send anyway” is real — logged for discussion, not blocked. Trust first; revoke per student, never for the whole class.

  • Course knowledge, in the answers

    Upload the syllabus, readings, or a group's notes. Answers come from your materials, with the retrieved passages visible.

  • Custom bots with custom guardrails (similar to custom GPTs or Gemini Gems)

    Create a custom system prompt plus a knowledge base, to create what you think is helpful, like a writing critic or Socratic tutor that asks questions.

    @draft-critic · “Ask one clarifying question, then critique argument structure only. Never rewrite prose.”

  • A prompt library the class builds

    The prompt one group refined becomes where the next group starts, with the original conversation also visible.

4Class discussion

Empowering class discussion

A spectrum dialogue turns a hard question into a picture of where the class stands, and why. Every member places an opinion on a spectrum, writes a point in a sentence or two, and includes the points of others in their own view.

  • Opinions are continuous, not a vote: the spectrum shows a class that is split, leaning, or converging, and it moves as people read each other's points.
  • Points are short by design, and the best ones are the ones most classmates include. Quiet students get the same weight as loud ones.
  • Named, anonymous, or fully anonymous, chosen when the channel is created and never loosened afterward. The question of AI in the course itself makes a good first dialogue.

Design based on prior research on enhancing understanding and collaboration in discussion from the University of Washington and Travis Kriplean on Consider.it.

The class spectrum A spectrum from Oppose to Support. Nineteen opinions, shown as equal-size bubbles stacked on the axis, start spread across the axis and settle into two clusters, a larger one on the Oppose side and a smaller one on the Support side, as three points are included. A Lime marker and bar show your own position. ai-in-our-writing spectrum 23 members · 19 opinions Should AI be allowed to draft the first outline of an op-ed in this course? Oppose Support you Against The outline is the thinking. If AI does it, I skip the part I'm here to practice. included by 13 For Fine as a scaffold if you reorder it around your own strongest argument. included by 8 For My first language is not English, I use it to help translate my thoughts then I refine it into my own words. included by 9

5Privacy

Privacy by design with pseudoanonymization and private student accounts

Nobody's name goes to the AI company. Identifiers are swapped for role-labelled pseudonyms before a query leaves the platform, and swapped back before you read the answer.

Want less on the platform in the first place? Instructors can create anonymous student accounts: a random username and no email, so a student's identity is never stored here at all.

Name swap On the left, your class platform shows a student, Priya S., and the AI's answer addressed to her. In the middle, a gate swaps her name for a pseudonym before anything leaves. On the right, the AI provider sees only Student A, and its answer comes back to Student A. The gate swaps the real name back before Priya reads the answer. Your class platform AI provider Priya S. AI Priya S., Student A AI Student A, Priya S. Student A Student A Priya S.

6Getting started

From the demo to your own course in four steps

  1. 1. Try the demo course

    Sign in with your institutional (.edu) Google account and you land in a shared demo institution with a sample course. It is a worry-free sandbox where you can't make mistakes. You can always use it to try something new without affecting or access to any student data.

  2. 2. Request your institution

    A short form, from the demo or from this page: your name, your institution, what you teach. A person at the lab reads it, and sets up your institution.

    Request your institution →

  3. 3. Create your students' accounts

    From your course's Students tab, create accounts in bulk. Each gets a random username like User1384 and no email address, so students never hand their identity to us or to an AI provider. (You may also give students the option to sign in with institutional Google accounts, but don't require that.)

    See the Students tab in the demo →

  4. 4. CollaborAITE!

    A basic amount of usage is included on the platform's keys, with daily limits per person and per institution. Add your own API keys for your institution or for a single course to lift the limits. Keys are encrypted at rest.

    How to get an AI API key: Anthropic ↗ OpenAI ↗ Google AI Studio ↗

7The CollaborAITE Team

Teachers and students

CollaborAITE comes from the YESS Lab at the University of Maine, designed and built by teachers and students, including undergraduate students.

  • Team member Undergraduate developer

  • Team member Undergraduate developer

  • Team member Undergraduate developer

  • Team member Undergraduate developer

  • Team member Undergraduate developer

  • Jon Ippolito Environment and design consultant

  • Dr. Greg L. Nelson Faculty lead · YESS Lab

Instead of hunting down AI use, become the leader of a pack that learns to use and not use AI responsibly.

Power with responsibility

Students are given power with responsibility: they can override the AI guidelines checker when they think it is wrong, but it is logged.

Students need to be able to make mistakes and get feedback to learn responsible AI use. Otherwise, they will just use AI and hide it.

Students that abuse AI can have their privileges limited in response.

Designed with educational values first

Co-designed with, and mostly written by, undergraduate students — a tool meant to give students agency should be built with students.

If an instructor can do something to help learning, a student should be able to do it too — add guidelines, invite a TA, start a channel, add AI, give feedback on someone's AI use.

8Questions

FAQ

Can my instructor see everything I do?
No. Private channels are private. One deliberate exception: overriding an AI-use guideline is logged for your instructor.
Does my name reach the AI provider?
Identifiers are pseudonymized before a query leaves the platform, and you can read the exact text that was sent. Details you write about yourself in prose aren't caught, so the protection has limits.
What if I don't want to use AI at all?
Then don't. CollaborAITE works as a collaboration platform without it, and you can exclude your own posts from ever being read by the AI.
Is this more work for instructors?
Some, up front. Invisible AI use has its own cost: work you can't give feedback on, and misconceptions you find at grading time.
Won't students just override the guidelines?
Some will, but that enables a conversation about judgement. A hard block will just push AI use where you can't see it. Override rights can be turned off per student.
Is this more private than (tool X)?
CollaborAITE uses industry standard encryption for stored data at rest and in transit. But CollaborAITE might be better because we just store less information.
CollaborAITE supports anonymous student accounts created by instructors for their course, without any email or identifying information associated with them.
Pseudonymization (sending Person_X instead of “Maya”) provides an extra layer of privacy, on top of AI providers’ agreements to not train on any data sent. AI queries are all sent inside one institutional bucket instead of associated with individual users or emails.
Data on CollaborAITE is scoped per institution using row-level security. To create a special separate environment for your institution, contact us for hosting options (such as hosting on an institutional Microsoft Azure account).
Decide together how to use and not use AI responsibly in your course.
Try the demo coursesign in with Google

Ready for your own course? Request us to add your institution.

9Pilots

From the pilot writing courses

Piloted in Spring and Fall 2026 with 100+ students at the University of Maine. Four moments the shared space made possible.

  • “Am I on the right track?”

    A student @'d their instructor mid-draft. The earlier attempts were right there in the channel, so the feedback could be about that student's actual argument.

  • Seven minutes, one survey

    An instructor built a check-in survey in the open with students adding questions, then taught from the needs they'd shared anonymously.

  • A better prompt, taken up

    A teammate saw a research report in the channel, improved the prompt, and ran a better one. Peer learning about AI, in public.

  • A discussion with no AI in it

    For their final essay workshop, students had every AI feature available and used none. We count that as the platform working.