Keep AI speed. Train on the code you actually ship.

Practice reading, debugging, and testing code yourself — with browser exercises and tasks from your own repositories.

Start free 5-minute review

3 exercises + 1 repo task free

See an example report

Today · Example2 / 5 VERIFIED

Stop stale search results

Selected from your repository because async reasoning is under-trained.

Proof: Regression test
Time: 18 minutes
Skill profile: Async reasoning

search-results.js

async function search(query) {
  const request = ++latestRequest;
  const response = await api.search(query);
  setResults(response.items);
}

Newest request wins

Expected “second”, received “first”

FAIL

See the missing guard

Ignore an older response before updating the results.

if (request !== latestRequest) return;
setResults(response.items);

Illustrative example, not a live test run.

Run the full example yourself

3 Gym + 1 Project

Free before you choose Pro

Your repositories

Useful tasks reserved by your agent

Browser practice

Gym exercises when projects do not fit

The verification gap

66%say AI output is almost right, but not quite.

Stack Overflow 2025 survey ↗

AI-GENERATED PATCHLOOKS SAFE

01 function resolveLimit(limit) {
02 return limit || 10;
03 }

One silent assumption changes behavior

Evidence

default limitPASS

custom limitPASS

explicit zero0 ≠ 10

01Read the assumption

02Prove the failure

03Make the smallest fix

How it works

One loop. Each rep gets more relevant.

No course catalog. No task hunting. ManualMode picks the work and learns from what you prove.

CODE READING74

DEBUGGING61

TEST DESIGN38

Measure

Three Gym reps calibrate your starting point; one Project rep proves the workflow.

Project · 18 min

BEST FIT

Guard request ordering

Gym fallback · 12 min

Debug stale async state

Train

Reserve useful project work first. Use Gym when no right-sized task fits.

Tests pass

Evidence verified

Queue changes

Difficulty +1

Adapt

Verified outcomes sharpen the next recommendation.

Project-first training

Your repository becomes the training ground.

ManualMode gives your coding agent a bounded skill target. It proposes useful work from the project already in front of you; you reserve one task and complete it manually in your IDE.

PROJECTAGENT RESERVED

src/search.ts

Useful work from your repository

Reserved by your coding agent · completed in your IDE

One profile

VERIFIED EVIDENCE

GYM FALLBACKBROWSER

async-search.js

Targeted reps when needed

Use when your project has no right-sized task

Not another puzzle site

Your real work becomes the workout.

LeetCodeA fixed puzzleCorrect answer

AI coding toolsYour promptGenerated code

ManualModeYour codebase + skill evidenceUseful work you complete yourself

Before your first exercise

Start small. Know what to expect.

Can I try it before creating an account?

Yes. The free code review exercise lets you review two synthetic patches and compare your decisions with an explanation. It takes about five minutes, runs no code, and does not count toward your free training allowance.

What do Gym and Project mean?

Gym is coding practice in your browser: read code, fix a bug, and run tests. Project is a useful task from your own repository, reserved through a connected coding agent and completed manually in your IDE. Both contribute evidence to the same skill profile.

Which programming languages can I use?

Browser Gym currently executes JavaScript. The free review exercise offers JavaScript/TypeScript, Python, Go, Java, and C++ as review-only scenarios. Project Mode uses your repository language and local tests when the connected workflow can verify the task.

Do I have to upload my repository?

No. Start with browser exercises without connecting a repository. In Project Mode, raw source stays local by default; the workflow sends bounded task metadata and verification summaries. Your coding agent has its own provider settings. See what ManualMode collects.

What is free, and when would I pay?

Create an account for 3 Gym exercises and 1 verified Project task, with no card required. After those free allowances, Pro provides continued practice at $20/month or $200/year. You choose whether to subscribe.

4 reps free · no card

Prove the loop before you pay.

3 Gym reps find your starting point. Then one free Project rep proves the workflow on code you actually ship.

Start free

GYM

PROJECT

Monthly

$20per month

Pay month to month while you build the habit.

Project tasks from your repositories
Tailored Gym fallback
Coach reads your code + test output

Billed monthly

Annual

$200per year

Train for a full year at the lowest monthly rate.

Project tasks from your repositories
Tailored Gym fallback
Coach reads your code + test output

$16.67/month · save $40

Unlimited adaptive repsGym drills and Project tasks

ManualMode CoachReads code + latest failure only when asked

Full evidence historyProgress across six skill axes

Try the method

A real example before you commit.

All practice guides

Review an AI-generated patch

Find an async bug, inspect the failure, and check the correction.

Practice in your own repository

Take a bounded stale-search bug from failing test to manual fix.

Write tests that catch AI mistakes

Turn the contract and edge cases into evidence before merging.

Copy an AI code review prompt

Install it in Claude Code, Cursor, or Copilot, then verify one finding yourself.

Check an AI-built app before you ship

Secrets, ownership checks, database rules, packages, and cost limits.

Practice debugging without AI

Five JavaScript bugs with failing tests and answers.