Product

Meet Pip.

Ask Pip to write it, plan it, debug it, or argue with your thesis. And under the school side sits a floor of hand-written lessons no model improvises.

An assistant you can hold to its word

  • Write, plan, code & research
  • Cites its sources
  • Admits when it isn't sure
  • No ads, ever
  • Authored courses for 12 exams

Capabilities

What Pip does today.

One section per capability — nothing aspirational, nothing "coming soon". Every claim is checkable in the product, and every "Learn more" lands on the doc that says exactly how it works, limits included.

Chat

One cockpit, 8 models.

Auto reads each message and routes it to the lightest model that can handle it — the scoring happens locally, before any model is called, so choosing Auto costs nothing and adds no latency. Or pick a model yourself: the cockpit says what each one is for, and third-party models are always your explicit choice, never Auto's.

Model
AutoAdaptive
GPT-4oTasks
GPT-4o miniFast
Claude Sonnet 5Balanced
Claude Opus 5Most capable
GPT-5.4Frontier
GPT-5.6Reasoning
GPT-5.6 ProDeepest
Grok 4.6Frontier

Research

Sources attached, or it says so.

Deep research reads real pages and pins citations to the claims they support, so you can open what it read. A run that finds no real sources is labelled that way — it is never dressed up as a finished report.

Research
4 sources read
1
1 claim labelled: no source found. Labelled, not dressed up.

Images

From sentence to image.

Describe what you want, pick a style, refine the concept you like. Styles are honest prompt expansions — the preset shows you exactly what art direction it added to your words, rather than doing something unexplained to them.

Images
a lighthouse at duskWatercolor

"Watercolor" becomes the full art direction — soft edges, flowing colors, paper texture — and you can read what was added.

Code

Code that runs, not just reads.

The code workspace writes code, runs it, and reads the failure — inside the conversation. Source files are read natively across dozens of file types, so debugging starts from your actual code rather than a paraphrase of it.

Code
app.jsxstyles.css
Runs in the conversation — and reads its own failures.

Study

And then there's the part nobody else has.

Most assistants improvise school. CHUD ships 12 written courses — the SAT, the ACT and 10 AP exams — covering 471 topics with units, lessons, quizzes and flashcards that a person sat down and wrote. It teaches the concept, then makes you prove you have it. Here is the whole ledger, counted from the shipped curriculum:

SAT28 topics
ACT28 topics
AP Biology30 topics
AP US History41 topics
AP Calculus AB29 topics
AP World History95 topics
AP Physics C: Mechanics34 topics
AP Chemistry92 topics
AP Human Geography28 topics
AP Physics 129 topics
AP Physics 222 topics
AP Physics C: E&M15 topics

Counted from the shipped curriculum, not typed into the page. If a course ever ships unfinished, this table says so on its own.

Memory

It remembers you — on the record.

Pip carries what matters between sessions: your classes, your goals, how you like things explained. What it stores is a list you can read, and every entry can be deleted on its own — memory you can audit, not a vibe you have to trust.

Memory
Prefers steps over summaries
Taking AP Physics this year
Working on a personal essay draft

Every entry visible. Every entry deletable. Nothing implied.

Install

The app is the download.

CHUD ships a web app manifest and a service worker, so a browser that supports installation can add it to a home screen or dock and open it without browser chrome. There is no app store listing to wait on and nothing to update by hand.

Install

Install CHUD

Adds to your home screen or dock. No app store required.

Opens without browser chrome

What you get

Four things it does properly.

Explain hydrogen bonding simply
Rewrite this paragraph, keep my voice
Quiz me on Unit 3 — 10 questions

One assistant, every lane

Write, plan, debug, argue with your thesis — one thread, one model picker, no app-switching.

syllabus.pdf
unit-3-notes.docx
lab-data.xlsx
essay-draft.md
review-set.pdf
syllabus.pdf
unit-3-notes.docx
lab-data.xlsx
essay-draft.md
review-set.pdf

Bring your own files

Drop in a syllabus, a lab sheet, a chapter. It reads them and cites what it used.

MONUnit 3 recall
TUEPractice set
WEDWeak spots

A week that plans itself

Give it the exam date. It works backwards into days you can actually do.

Docs
Sheets
PDF
Markdown
Slides

It ends up where you work

Google Docs, Sheets, PDF, Markdown — the answer leaves the chat as a real file.

Models

The roster, on the record.

This list is read from the same catalog the in-app picker renders, engine names included — so what is advertised here is what you get. Auto routes within your plan's roster by task difficulty — the engine names below are exactly what runs, and the reply always says which one answered. On pricing these same engines are grouped into tiers as Pip Lite, Pip, and Pip Max — two names, one roster.

No per-message prices are printed here on purpose: cost depends on how long the answer is, so a flat number would be a fiction. Monthly allotments live with the plans.

Auto

Automatically routes each message to the right tier by difficulty.

from Free

GPT-4o

Runs the app's structured jobs — lesson answers, key terms, quizzes, plans.

from Pro128K context

GPT-4o mini

The fastest, lightest tier. Streams answers instantly for quick questions.

from Free200K context

Claude Sonnet 5

Fast enough for chat, smart enough for real work — with live thinking and web search. The default.

from Student1M contextlive thinking

Claude Opus 5

Best-in-class reasoning, long-form output, live thinking, and web search.

from Pro1M contextlive thinking

GPT-5.4

OpenAI's most capable frontier model — strongest coding, long-context and multi-step work.

from Pro400K context

GPT-5.6

OpenAI's current reasoning model — plans before it answers, and thinks harder as the question gets harder.

from Student400K context

GPT-5.6 Pro

The highest-intelligence option here: OpenAI's -sol model run in pro reasoning mode. It does substantially more work per answer, so it is slower and costs more — pick it for genuinely hard problems, not for chat.

from Pro400K context

Grok 4.6

xAI's frontier model, with the largest context window here — 500K tokens, enough for a whole textbook chapter in one go.

from Pro500K context

CHUD is in beta. Things ship most weeks, some of them break, and when they do we say so rather than quietly patching it. Read what changed.

Pricing

Free is a real tier.

Free

$0forever

120 tokens a month

Real chat, and deep research that shows its sources. No card, ever.

What counts as a token?

Student

$8/mo

640 tokens a month

Every AP unit opens, and Pip steps up to the model that handles real work.

What counts as a token?

Pro

$18/mo

1,440 tokens a month

Every model, the code workspace, and the design canvas.

What counts as a token?