Whetstone

Why Whetstone

Skills need three kinds of care, and Whetstone gives all three:

• Learn what you don't know. Pick a system you want to understand, or a topic your next role will need. Your agent writes it as a series of steps, each one a little harder than the last, starting from what you already know.
• Relearn what you know. The things you learned years ago, or now hand to your AI, fade quietly. Whetstone brings them back as fresh problems, so you find out what is still there.
• Keep your skills sharp. Every idea you solve returns on a new surface after 3 days, 3 weeks, 3 months, 6 months and a year. Each skill shows how sharp it is today, and your agent suggests what to practice next.

Built around you. Your agent already knows you from the work you've done together: what you build, your stack and experience, where you're headed, and what it now writes for you. It drafts your profile from that, and from your GitHub activity if you agree, so there's nothing to fill in. You only check the draft. Every question is then written for you: a scenario from your world, at the step you're ready for, explained the way you learn.

Real problems from real interviews. The question bank is grounded in 2,771 interview reports from 397 companies, organised into 271 modules and 78 systems to build. The problems are practical: ledgers that must balance, retries that must not charge twice, rate limiters, caches and schedulers. There are no throwaway puzzles. That makes Whetstone good interview preparation too: filter by the companies you're aiming for, practice system design on a whiteboard, and ask your agent for a mock interview.

Your AI as reviewer and interviewer. The same agent that writes your questions also reviews your solutions like a staff engineer, nudges you one step at a time when you're stuck without giving the answer away, pairs with you in the editor's margin, and runs mock interviews.

Private by design. Whetstone runs on your machine. Your profile, your questions and your progress never leave it. There's no account and no telemetry, and it runs on the AI subscription you already have.

Same idea, written for you

Three engineers ask their agents for the next question in the rate limiter module. Each one practices the same idea, a token bucket, in the world they work in:

The learner	The question their agent writes
A payments engineer	Throttle the webhook sender. Card networks accept 50 calls a second per merchant. Queue the rest instead of dropping them, and never let one busy merchant starve the others.
A trading engineer	Stay under the exchange's order limit. The venue rejects a burst over 200 orders a second and bans you for a minute. Smooth a burst at the open without losing the order.
An LLM platform engineer	Share a token budget across model providers. Each provider allows a number of tokens a minute. Spend the budget across many users fairly, and fall back when one provider is full.

You pick the world, too. When you press Prepare, choose your own work, a company you're preparing for, or a domain such as payments or trading. Before an interview at Uber, the same rate limiter question can come back set in Uber's world, built on the questions Uber candidates report.

[!NOTE] Whetstone has no AI of its own. Your agent (Claude Code, Codex, Cursor or another) writes the questions on the subscription you already pay for, and Whetstone checks and grades them on your machine.

What you practice

What your AI does for you	What you practice in Whetstone
Writes the code	Build questions: the problem comes first; the lesson opens after your first try
Explains code you didn't write	Predict questions: read working code and say what it does before you run it
Finds the bug	Fix questions: a bug report with symptoms, never the cause
Reviews its own pull request	Review questions: find the defects planted in plausible AI-written code
Writes the tests	Tests questions: your tests must catch hidden bugs you never see
Sketches the architecture	Design questions: draw it on a whiteboard, and your agent grades it against a rubric
Works through a whole task with you	AI-allowed rounds: a small codebase, solved in stages with the AI tool you use at work, like the interview rounds that allow one. Then, with the AI closed, you say what your code returns
Writes the one-liner	Terminal questions: text pipelines, jq and git rescues in a real shell, with no network and nothing else to break. Your script is replayed on copies of the scenario you never saw

Quick start

You need Docker and an AI coding agent that can run commands in a folder.

git clone https://github.com/boffti/whetstone && cd whetstone
cp .env.example .env
docker compose up -d --build        # then open http://localhost:8000
claude                              # or codex, or open the folder in Cursor

Open http://localhost:8000 in a browser pane next to your agent, such as the built-in browser in Claude's desktop app, or in any browser tab. Then tell your agent: "Set up my Whetstone profile." If something doesn't work, run bin/whetstone doctor: it checks your setup and says how to fix each problem.

The exercises are Python 3.12 today. Building a question takes your agent a few minutes. To update Whetstone, run git pull && docker compose up -d --build.

How it works

flowchart LR
    P["Your profile"]:::clay --> W["Your agent<br/>writes a question"]:::steel
    W --> C["Whetstone<br/>checks it"]:::steel
    C --> S["You solve it,<br/>tests grade it"]:::green
    S --> R["It comes back<br/>days and weeks later"]:::clay
    R --> W
    classDef clay fill:#B8572F,stroke:#9C4526,color:#FFFFFF
    classDef steel fill:#2F6497,stroke:#244C77,color:#FFFFFF
    classDef green fill:#2E7A4D,stroke:#245F3C,color:#FFFFFF

여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗