Skip to content
Trace

Everything you need to teach AI, ready to assign.

More than 200 interactive labs, readings, quizzes, question banks, and AI projects across data science, neural networks, large language models, and building agents. Written for the classroom.
See how educators use it

A written library across the whole stack.

Readings, quizzes, interactive labs, and agent projects come authored and versioned. Pick one piece or a full path and it shows up for your class.

Readings

Short explanations in order, written for high school.

Quizzes

Questions on the reading, with each student's answers kept.

Labs

Change one setting and compare what comes back.

Agent projects

A build that runs, submitted with the runs behind it.

Students always get the published version you assigned, so editing your material later never rewrites work already underway.

What the courses cover.

The library is one path from what a dataset is to a working agent, split into courses you can assign a piece at a time.

Data and prediction

Datasets and labeling, classification, regression, and clustering — with the confusion matrix, ROC curve, and k-means running as labs, not slides.

How models learn

One neuron to backpropagation, then reinforcement learning from a reward signal to Q-learning and reward hacking. Students turn the crank and watch the loss move.

Inside a language model

Tokens, attention, sampling, the KV-cache, and why a model makes things up — the actual mechanisms behind the tools students already use.

Prompting and agents

Prompts and chain-of-thought, then the agent loop: a tool call, the transcript, permissions, and the blast radius when something goes wrong.

Fairness, privacy, governance

Where bias enters, re-identification risk, documentation, and how to audit a system — the questions a school has to answer about AI.

Written to be taught

A reading leads into a lab, the lab into a quiz, the quiz into a project. You assign a path and it holds together instead of reading like scattered files.

Every lab hands over one control.

A lab exposes the single thing its lesson is about and nothing else. Students predict what will happen, change the value, and compare the result to their guess. Move the classification threshold and watch precision and recall trade off. Take one gradient-descent step and see the loss drop. Raise the sampling temperature and watch the answer drift.

The lab below is the same kind students get inside a lesson. Change the temperature or the prompt and read what comes back.

See how an AI response changes

Edit the instruction, the question, or the response variety

Response variety0.70
More consistentMore varied

Example response

Change the instruction, question, or response variety, then try it to compare what the model does.

Turn the same questions into a printable worksheet.

Pull questions from the bank, drop in headings and directions, set how much answer space each one gets, then export a clean PDF — with an answer key when you want one.

Question bank
  • What does a model turn your sentence into first?
  • Why does a higher temperature change the answer?
  • Name one reason a model states something false.

Drag a prompt onto the sheet, or add a heading, directions, or a page break.

Worksheet
Answer keyExport PDF

How a language model picks the next word

Use what you noticed in the sampling lab. Answer in your own words.

  1. 1.

    A model turns your sentence into ___ before it reads it.

  2. 2.

    Raising the temperature makes the next word more ___.

  3. 3.

    Give one reason a model can state something false with confidence.

The questions carry over from what you already assign, so a printed review sheet and the on-screen quiz stay in step.

One catalog, the right depth for the grade.

The same chapter is authored for a grade band rather than watered down. Younger students point and choose from a few large blocks; older students get the numeric dials and the full pipeline.

Early grades

A helper is one big block — ask, remember, allow, stop. Students wire behavior by choosing, not by reading a metric.

Middle grades

Pipelines stay black-boxed — retrieve, cite, refuse when there is no support — with rubric scoring instead of numbers.

High school

Everything opens up: sampling and temperature dials, precision and recall, and pipelines students take apart stage by stage.

The course on how language models work already ships in two grade bands, each written on its own — the same mechanism, a different amount of machinery on screen.

One submission

The work and the runs behind it, kept together

Submitted

Responses

Quiz answers, lab observations, and written thinking.

The agent

The exact block graph turned in for the assignment.

Run history

Each model turn, tool call, output, and error.

See exactly what each student did.

Open a submission and you get the real thing: their answers, the agent they built, and every run behind it down to the model turns and tool calls. You talk about a specific choice a student made, because Trace never turns clicks or time into a mastery score it invented.

Shortcuts are visible on the record instead of hidden.

Real AI, included. Not your credit card.

Students run on the same production models a professional would, through the visual builder instead of raw code. The credits behind those runs come with your plan as a shared classroom pool. You hand credits to students from it, so no student signs up for a model account and nothing is billed to you per run.

See what each plan includes

Class balance

AP Computer Science · 26 students

Refills monthly

Credits left this month

24,180of 30,000

Recent runs

  • Build a research agent

    24 students, 61 runs

    −512

  • Temperature and sampling lab

    26 students

    −143

  • Tokenization warm-up

    26 students

    −64

Students pay nothing

You are not billed per run

Unused credits do not roll over

Example figures on an Educator Pro plan, which includes 30,000 credits a month.

Privacy enforced on every model call.

These run on the request itself, not on a student's browser or a consumer app that lives off their data.

What a student types

Inside Trace

Hi, I'm Maya Chen. Can you check the notes I emailed from [email protected]? My student ID is 4471029.

What reaches the model

After Trace rewrites it

Hi, I'm [REDACTED_STUDENT_IDENTIFIER]. Can you check the notes I emailed from [REDACTED_EMAIL]? My [REDACTED_STUDENT_IDENTIFIER].

Sent on the same request

"provider": { "data_collection": "deny", "zdr": true }

The provider is told not to keep the request or use it for training.

Not used to train models
Not sold
Only tools Trace provides
An example message. Trace also redacts phone numbers, addresses, and dates of birth.

A student's own identifiers are removed alongside any personal details in what they type, a separate safety check reviews both the input and the answer, and agents can only call the sandboxed tools Trace ships — no student keys, no arbitrary URLs.

Common questions

What teachers ask before they assign the first path.

Readings, quizzes, interactive labs, and agent projects across roughly a dozen courses — more than 200 hands-on labs in all — from data science and neural networks to how language models and agents work.

Assign your first path this week.

Create an account