Rabbithole is integrated into a school day that also includes direct instruction from our teachers, hands-on projects, physical activity, and collaborative work. Students engage in a three-way dialogue with teachers and an AI thought partner, where continuous, individualized assessment happens naturally, complementing other types of learning.
The on-screen and off-screen parts of the day reinforce each other. Inquiry on Rabbithole fuels what students build in the maker space, investigate in the lab, or debate in seminar—and those experiences come back into Rabbithole as students reflect on what they did and what they learned.
Tradewinds is a nonprofit school, not a software company. Rabbithole is built to serve teachers and families, not to maximize screen time or sell data. When a student is done exploring, the system is done—there's no engagement loop keeping them scrolling.
A project of the Tradewinds Center for Advanced Learning, built by Andy Szybalski—designer and co-creator of Google Street View and Uber Eats—and Carl Sabatino, a nationally recognized gifted educator with 30+ years building unconventional schools.
For Students
A Thought Partner, Not a Tutor
When students sit down with Rabbithole, they get a Socratic dialogue partner. The AI doesn't give answers—it asks questions, offers different angles, and adapts based on how the student responds. It adjusts its language to each student's level so vocabulary never becomes a barrier to sophisticated thinking.
Each activity comes with a teacher-designed process: structured checkpoints where students check in with their teacher for sign-off and feedback. The AI guides students through these phases, gently steering them away from dead ends and prodding them to go deeper when they're ready.
Debriefs:Rabbithole isn't just for screen time. After a lab experiment, a field trip, or a maker space session, students sit down with Rabbithole to reflect on what they did and what they learned. These debriefs turn experience into understanding. John Dewey argued that we don't learn from experience—we learn from reflecting on experience. Structured reflection builds metacognition: the ability to think about your own thinking, which research consistently identifies as one of the strongest predictors of academic growth.
Socratic dialogue — structured checkpoints on the right
For Teachers
Teachers Set the Vision. AI Lets Them Be Everywhere at Once.
Every child learns differently, but a single teacher can only differentiate so far. Rabbithole lets teachers effortlessly personalize curriculum to keep each student in the sweet spot of challenge—engaged by their interests, pushed just beyond what they already know. Teachers design the activities, set the checkpoints, and decide where learning goes next. The AI makes it possible to do that for every child simultaneously.
Structured Activities with Human Checkpoints
Teachers design activities—a unit of work with a set of scholars and a timeframe. Each activity has a structured process: checkpoints where students check in with their teacher for feedback the AI might have missed. Teachers see every scholar's progress at a glance: who's in the “View” phase, who's moved on to “Origins,” who needs a nudge.
Activity overview — all scholars' progress at a glance
AI Suggests Where to Go Next. Teachers Decide.
As students work, Rabbithole observes what genuinely engages them and generates “seeds”—suggestions for where their learning could go next. A student fascinated by coordinate geometry might get a seed connecting it to game design. A student exploring narrative structure might be nudged toward data visualization.
Teachers review and approve seeds during prep periods, after class, or in real time. They can also plant their own. The result is a growing queue of interesting topics tailored to each scholar—ready to spark the next rabbithole.
Rabbithole passively assesses understanding from every conversation—including post-activity debriefs where students reflect on hands-on work. It maps each concept against Bloom's Taxonomy, from basic recall through creation.
Crucially, mastery tracking isn't limited to a fixed rubric. Most ed-tech starts with a predetermined knowledge graph and moves students through it. Rabbithole starts with the student and builds the map as they explore. We cross-reference against state and federal standards and research-based frameworks like UCLA Historical Thinking—but those represent a fraction of what's worth learning. A student who explores topology through sailor's knots or materials science through rice cookers is building real knowledge, even if no curriculum standard covers it. The mastery profile isn't a checklist—it's a living portrait of how your child thinks.
Mastery profile — concepts tracked by Bloom's level
Bloom's Taxonomy — Extended for Gifted Learners
We Focus on Higher Order Thinking
Rabbithole assesses understanding using Bloom's Taxonomy—a framework developed by educational psychologist Benjamin Bloom in 1956 after observing that most classroom questions only test recall and basic comprehension. Bloom identified six levels of cognitive complexity, from simple remembering through creation. We extend his framework for gifted learners, where the goal is to spend as much time as possible in the upper tiers.
CreatingProduce, Design + Divergent creativity, open inquiry, innovation
RememberingRecall — Gifted learners often enter beyond this tier
A Portrait of How Your Child Thinks
Beyond subject mastery, Rabbithole tracks the habits that matter most: self-direction, task commitment, creative approach, emotional engagement, productive struggle, metacognition, and intellectual intensity. These are assessed passively from conversation data and updated over time, giving teachers a rich portrait of each scholar's growth.
Rabbithole also detects cross-domain connections—moments when a student spontaneously links ideas across disciplines. A student connecting Chinese cultural symbolism across literature, philosophy, and psychology is doing exactly the kind of transdisciplinary thinking we want to cultivate. We're still exploring the best ways to actively spark more of these connections, but we believe recognizing them is the first step.
Learner Profile — habits of mind + cross-domain connections
For Families
The Bridge Between School and Home
Most parents want to ask “what did you learn today?” and get more than “nothing.” Rabbithole makes that conversation real. At the end of each day, it generates a personalized Topics to Go sheet for each scholar: a short list of what they explored, questions they asked, and conversation starters for the dinner table.
We see education as a partnership between school and family, not a handoff. Rabbithole is the bridge that keeps parents connected to their child's learning without adding to teacher workload.
Topics to Go
Daily conversation starters based on what your child actually explored. Specific enough to spark a real conversation at dinner.
Shared Mastery Profiles
Parents see the same mastery profile that teachers see—emerging strengths, growth areas, and the full landscape of what your child knows.
Parent-Suggested Seeds
“She's really into volcanoes right now.” Parents can suggest exploration seeds that teachers review and approve into their child's queue.
From Our Early Testing
“How Do Rice Cookers Work?”
One of our earliest Rabbithole testers was an 8-year-old who sat down one afternoon with a simple question: how does a rice cooker know when the rice is done?
It's the kind of question most adults can't answer well. And it turns out the answer is genuinely fascinating.
Bimetallic Strips
Rabbithole guided him to discover that different metals expand at different rates when heated. A strip made from two metals bonded together will bend as it warms up—and this bending can flip a switch. That's how simple rice cookers measure temperature without any electronics at all. He spent twenty minutes just on this, asking follow-up questions about which metals and why.
Phase Transitions
Then came the real breakthrough: the temperature of boiling water stays locked at 100°C until all the water is gone. As long as there's liquid water in the pot, the temperature can't rise past the boiling point. The moment the last of the water turns to steam, the temperature spikes—and the rice cooker detects that spike and clicks off. He was genuinely amazed. “So the rice cooker doesn't know the rice is done,” he said. “It knows the water is gone.”
Feedback Loops
Rabbithole nudged the conversation toward a broader idea: this is a feedback loop. Sense something, respond, sense again. He explored how the same principle appears in thermostats, cruise control, and even the human body regulating its own temperature. The concept clicked immediately—he could see the pattern everywhere once he had the name for it.
The Marshmallow Roaster
Then he had an idea of his own: what if you built a marshmallow roasting device that worked like a rice cooker—but instead of sensing temperature, it used a camera to sense the shade of brown on the marshmallow? A visual feedback loop instead of a thermal one. He iterated with Rabbithole on the concept, working through how you'd mount the sensor, what “golden brown” means in terms a camera could understand, and how the device should pull the marshmallow away from the flame as it approaches the perfect color. By the end he had a detailed concept for his perfect marshmallow roaster.
In about forty-five minutes, an 8-year-old explored materials science, thermodynamics, phase transitions, control systems engineering, and computer vision—then pulled it all together into an original design concept. None of this is in any curriculum standard for third graders. The AI didn't lecture him through a lesson plan. It followed his curiosity, gently steered him away from dead ends, and prodded him to go deeper when he was ready.
After the session, his teacher saw new mastery concepts logged—thermal expansion, phase transitions, feedback loops—each mapped to Bloom's levels. And Rabbithole generated three new exploration seeds based on where his curiosity was heading. That's the system working as designed.
Open Source Is the Goal. Here's Where We Are Today.
You wouldn't want a tutor who refused to explain their teaching methods. Why accept that from software? The educational philosophy behind Rabbithole is published in a public repository — the application code and every system prompt, so you can read exactly what the AI is told to do. When we write “use the Socratic method” or “track mastery using Bloom's taxonomy,” that's pedagogy you can read and question. Our incentives are aligned with yours because we're a nonprofit, not a software company.
We intend to make Rabbithole open source. We're not there yet, and we'd rather say so than blur it. Today the code is source-available: published for inspection under an all-rights-reserved license. You can read every line, but deploying, modifying, or redistributing it needs our written permission for now.
The reason for the gap is that a license is a promise about how something can be run, and we don't yet know enough about how schools need to run this. Hosted or self-hosted, one institution or many, a district or a nonprofit maintaining its own version—those are real differences, and getting them wrong in a license is harder to undo than getting them right slowly. So we're working through them with design partners first, and choosing a license once we understand what it has to support.
We're looking for design partners
What works at Tradewinds shouldn't stay locked inside one school. If you want to deploy or adapt Rabbithole, get in touch and we'll talk through what you need—those conversations are what will decide the license. Permission is granted case by case today, and the shapes we're exploring with partners are:
Joining the hosted rabbithole.school service as one institution
A self-hosted deployment for a single institution
A self-hosted, multi-institution deployment
A philosophy-, language-, district-, or nonprofit-specific version, serving one school or many
None of these are granted by the current license on their own. Each one starts with a conversation and our written permission. Which of them a future open-source license has to support is exactly what we're trying to learn.
