Trace teaches students how AI works.

A student reads the idea, tries it in a lab, then builds an agent that uses it. A teacher assigns the material and reads what each student did. Trace is made in New York by one person, with the teachers and students who use it.

Try a lesson Talk to David

Students at desks in a lecture room, one with a hand raised, while a teacher listens.

How learning works on Trace

Every chapter moves a student through the same three steps, in the same order.

Read

A short explanation of what is actually happening inside the system. You cannot direct a model you do not understand.

Try

Change one thing in a lab and see what changes with it. The idea becomes something a student has done, not something they were told.

Build

Put the idea to work in an agent that runs, then read what it did step by step. The thing a student hands in works.

What is in the library

Written for high school. A teacher can assign a whole chapter or any single piece of it.

Readings

Short articles that explain one idea at a time, with a check question after each. A student always knows what they just learned.

Labs

Change one setting and watch the result move. An idea sticks once a student has seen it change under their own hands.

Quizzes and unit tests

Every answer is kept, so a teacher can see which questions a class missed instead of a bare percentage.

Agent projects

A build that runs. A student submits the exact version and the runs behind it, so the work shows how it was made.

Agent Builder

Blocks on a canvas, connected into an agent that runs on a real model. No syntax stands between a student and the idea.

AI tutor

Help inside a lesson, step by step. It will not hand over a finished answer, so it stays a tutor and not a shortcut.

Dev tools

A real model API for students who want to write code. Credits come with the account, so nobody needs a card or a provider key.

Resources

Guides and documentation for teachers and developers, kept in one public place.

What students learn

The course starts with examples students already recognize and ends inside the loop that runs an agent.

AI around us

Spot learned systems in everyday products, see what goes in and out, and judge whether one should be trusted.

Data, patterns, and labels

Where a dataset comes from, who is in it, and what it cannot support.

Classification and regression

How a model decides, and how to measure its mistakes honestly.

Neural networks

Compute one neuron by hand, then trace and train a whole network.

Reinforcement learning

Learning from reward instead of from answers, from bandits to games.

Language models

Tokens, the transformer, inference, and why an answer costs what it does.

How agents work

A model in a loop with tools, the transcript it carries, and when to stop.

Responsible AI

Where harm enters a system, fairness criteria, privacy, and oversight.

How the lessons are made

Every chapter is written in a reviewed source repository, separate from the app, and checked before Trace imports it. Publishing creates a fixed version. When a teacher assigns a lesson, students get that version, so a later edit never rewrites work already underway.

A chapter is made of units. Each unit pairs short readings with a lab and a quiz on a single idea, and ends with a test. A teacher adds an Agent Builder project where the material calls for one.

The lessons and the Agent Builder use the same words for the same things: the request a model receives, the transcript, a tool call, a stop reason. What a student reads is what they see on the canvas when their own agent runs. Lessons describe behavior a student can test. They do not say the model thinks or feels.

What Trace will not do

These are decisions, not gaps. They shape what gets built next.

Invent a mastery score

Trace records what a student opened, finished, answered, and built. The teacher sees that evidence and makes the judgment. A number would only hide it.

Do homework for other classes

Every message to a model and every reply is screened. A request to solve a school problem outright is refused, and the teacher can see the flag.

Give agents the open web

Agents use only the tools Trace ships. A student can learn how tool use works without anything leaving the classroom.

Run ads or sell data

No advertising and no ad cookies. Student work is never sold and never used to train a model. Schools can read the exact wording in the privacy policy.

Make student work public

Nothing is public by default. A student chooses what to share and can turn a share off.

Replace the teacher

Trace does not decide what a class studies. The teacher previews, assigns, and reads the work.

Who pays, and for what

Students in a class never pay. The person or school running the class does. Credits cover the model usage behind runs and refill every month.

A student on their own
Free, with every reading, lab, and quiz, the Agent Builder, and monthly credits for runs. A paid plan adds the tutor, study notes, chapter projects, and more credits.
A teacher
Free to start with a couple of classes. Paid plans add classes, students, and a larger shared credit balance for the class.
A school
One plan per teacher, one bill, seats managed by the school. Districts and networks get a custom agreement with single sign-on and invoicing.

See current prices

Who makes it

I'm David Wu. I study electrical engineering at The Cooper Union in New York, and I build Trace. Before this I spent four years tutoring computer science, from a first loop to students asking why a language model confidently made something up.

The same gap kept showing up. Students could use AI tools fluently and had no model in their head of what the thing was doing. Teachers wanted to cover it and had nothing that was both honest about the technology and usable in a class period.

Trace is the version of that tutoring that does not need me in the room. If you teach, learn, or run a school, I want your opinion on what this should be. I read every message.

Now
The Cooper Union
Electrical engineering, New York City.
Before
DAIR Lab, Hunter College
Research on how multiple language models reason and work together.
Before
Teachshare and Fathom TX
Built AI tools for teachers, and developer tools for a software team.
Throughout
Tutoring
Four years teaching computer science, in groups and one-on-one.

Talk to the person building it.

Book a call, or write to [email protected]. Bug reports, lesson corrections, and blunt feedback all land in the same inbox.

Book a meeting Send a message