idle.py logo

Tap. Compute. Learn Python.

An incremental idle game that teaches Python and modern AI engineering, one FLOP at a time. Tap your way from x = 1 through AGI Lab and into the Singularity. Try a slice right now — no install needed. ↓

β–Ά live demo
idle.py
learn python & ai engineering, one FLOP at a time
0
FLOPs
+0
per sec
0
insight
Γ—1
multiplier
+1 FLOPs / tap
buy
Just a slice of the real game β€” the full version adds 11 more Compute Units (through ASI), 8 always-available Python basics, 3 unlockable side curricula, 50-question practice banks, 70 AI Landmarks, multi-step engineering scenarios, surprise mid-run events, retrain with spaced-repetition review, and an optional Mentor Pack. Β·

Built to teach. Designed to be played.

Every Compute Unit unlocks an interactive lesson, a quick-check quiz, and a 50-question practice bank. Mid-game side curricula, surprise events, and applied engineering scenarios keep the loop alive long after the early-game hook fades.

⚑

Idle progression

Tap to generate FLOPs, or buy Compute Units that produce them while you sleep, work, or stare into the middle distance. Offline progress is applied on relaunch.

πŸ“š

40+ lessons

13 Compute Unit lessons (Variable β†’ AGI Lab), 8 always-available Python Basics, plus 12 lessons across 3 unlockable side curricula β€” Debugging, Performance, Frontier AI.

🧠

Mastery Β· Knowledge Β· Insight

Three reward axes from quizzes, practice banks, and retrain. Each layered. Each compounding. None of them grindable by accident.

πŸ†

70 AI Landmarks

From McCulloch–Pitts Neuron through GPT-4, Claude 3, and beyond. Each lifetime-FLOPs marker grants a permanent tap multiplier you keep across every retrain.

πŸŽ–οΈ

23 achievements

Cosmetic badges for tapping milestones, retrain counts, mastery, knowledge, and reaching landmarks. No grind walls β€” just recognition for the work you did anyway.

🧩

Engineering scenarios

Multi-step puzzles that drop you into a real situation β€” a hallucinating support bot, a slow endpoint, a silent bug. Pick the right diagnostic call, earn a FLOPs reward.

⚠️

Surprise events

OOM at step 47K. Memory leak in prod. Eval regression in CI. Random in-run interruptions test your applied judgement, not your trivia recall.

πŸ”

Retrain with review

Retraining quizzes you on your three lowest-mastery lessons before the reset. Drift, and mastery drops β€” just like real understanding does.

♾️

Transcendence end-game

Stack ten GPU Clusters to earn a Transcendence Unit. Spend TU to unlock Singularity (5 TU) and ASI (25 TU) β€” tiers that sit above the regular FLOPs ladder.

πŸ“–

Built-in docs

An in-app reference covering 44+ topics across Python Basics, Foundations, Type System, Standard Library, Numerical & ML, Training, AI Patterns, and Tools & Workflow.

🍎

iPhone & iPad ready

Built with React Native + Expo. Universal iOS binary, dark UI, accessible tap targets. Saves are stored locally on-device β€” no account required.

πŸ›‘οΈ

Privacy-respecting

No third-party ads. No personal-data collection beyond purchase records. Anonymous analytics only, and you can request deletion any time.

The curriculum

A guided path from your first variable binding through the Transcendence tiers. 13 Compute Unit lessons gate the main ladder; 2 more sit above it on the Transcendence loop. Tap a tier to peek at the code.

01
Variable
Name binding, the atom of every program.
x = 42
name = "ada"
02
Function
Pure compute, the first abstraction.
def square(x):
    return x * x
03
Class
State + behaviour, encapsulation.
class Node:
    def __init__(self, v):
        self.v = v
04
Module
Namespaces and imports.
from math import sqrt
import torch as T
05
Tensor
N-d arrays, vectorised ops.
x = torch.rand(8, 32)
y = x @ W + b
06
Neuron
activation(WΒ·x + b).
y = relu(x @ W + b)
07
Layer
Composed transforms + residuals.
x = norm(x + mlp(x))
08
Attention Head
Self-attention. The 2017 idea.
a = softmax(Q @ K.T / √d)
out = a @ V
09
Transformer Block
Attention + MLP + residuals + norms.
x = x + attn(norm(x))
x = x + mlp(norm(x))
10
Foundation Model
Pretrain β†’ adapt.
model = pretrain(corpus)
adapt(model, task)
11
Agent
LLM in a tool-calling loop.
while not done:
    a = llm.act(state)
    state = env.step(a)
12
GPU Cluster
DDP, TP, PP at scale.
model = DDP(model)
loss.backward()  # all-reduce
13
AGI Lab
The recursive endgame.
# left as an exercise
# for the reader 🐍
14
Singularity
Transcendence tier Β· costs 5 TU.
# beyond the regular ladder
while True:
    self = improve(self)
15
ASI
Superintelligence Β· costs 25 TU.
# the ceiling
# unlocked only via Transcendence Units

Python Basics always unlocked

Lists & Tuples
Dicts & Sets
Strings & f-strings
Control Flow
Comprehensions
Iterators & Generators
Decorators
Exception Handling

Side curricula unlocks mid-game

Three small-scope tracks that unlock at lifetime-FLOPs thresholds across the run, distributing the “new lesson” surprise instead of clustering it all in the early Compute Units.

Debugging β€” Print debugging Β· Reading tracebacks Β· breakpoint() & pdb Β· Logging vs Print
Performance β€” Profile first Β· Vectorization Β· Lazy generators Β· @lru_cache
Frontier AI β€” Distributed training Β· Attention internals Β· Scaling laws Β· RLHF Β· Alignment

Mentor Pack optional add-on

A bonus curriculum unlocked by the one-time Mentor Pack purchase. Same lesson + quiz format as Basics β€” purely additive content for players who want to keep going past the core syllabus.

LoRA
RLHF
Mixture of Experts
RAG
Function Calling
MCP
Tool Calling
Evals

Lessons test recall. Scenarios test judgement.

Two layered systems make the mid-game feel like real engineering work, not flashcards.

🧩

Engineering Scenarios

Multi-step puzzles drawn from real situations: The Hallucinating Support Bot, The Slow Endpoint, The Silent Bug. Each scenario walks you through a diagnostic with 3–5 judgement calls. Wrong answers don’t kill the run — you retry until you reach the correct call, then collect a FLOPs reward proportional to difficulty.

πŸ“Ÿ

Random Events

Once you’re past the early game, a long cooldown gives way to surprise interrupts: an OOM at step 47K, a memory leak in production, a 4% eval regression in CI, a prompt-injection report from security. Single-question, single-decision, paired with a real-world explainer.

Pop-up achievements & landmarks the whole way

Buy your first Compute Unit. Cross an AI Landmark. Ace a quiz cold. Every meaningful step lands with a satisfying ping — 23 cosmetic achievements and 70 lifetime-FLOPs landmarks across the run, each landmark granting a permanent tap multiplier you keep through every retrain.

πŸ‘‹
Hello, World
Cross the first lifetime-FLOPs landmark.
πŸ›’
Bound & Determined
Buy your first Variable.
πŸ€–
Automated
10+ FLOPs/sec from idle.
🧠
Backprop
Master a Compute Unit lesson to Γ—3.
πŸ†
GPT-4 Baby
Cross the GPT-4 AI Landmark.
♻️
First Retrain
Retrain and earn your first insight.
🌱
Recursive Improvement
Retrain 25 times.
♾️
Transcend
Earn your first Transcendence Unit (TU).
πŸ§™
ASI
Unlock the deepest Transcendence tier.

Free to play. Optional support.

The full game is free. There are no ads, no energy timers, no paywalls between you and the curriculum. The in-app store carries optional, fairly-priced items for players who want to support development or skip ahead after a long break.

πŸ’Ž

Founder Pack

One-time, non-consumable. Permanent 2× FLOPs/sec on every Compute Unit and on every tap, forever. Restorable across devices on the same Apple ID.

πŸŽ“

Mentor Pack

One-time, non-consumable. Unlocks a bonus curriculum: LoRA, RLHF, MoE, RAG, Function Calling, MCP, Tool Calling, Evals. Restorable.

⏩

Skip 4h / 24h / 7d

Consumable catch-up packs that apply offline progress at your current rate. The 24h and 7d skips unlock after your first / second retrain so they’re catch-up tools, not loop-skippers.

🌳

+5 / +15 Insight

Consumable packs that grant Insight directly — the same currency you earn by retraining. Each Insight you own multiplies your Compute Unit output, and they compound.

Purchases are processed through Apple In-App Purchase. Restore at any time from Store → Restore Purchases. Refunds are handled by Apple via reportaproblem.apple.com. Full terms in Terms & Conditions.

Ready to graduate from the demo?

Launching on iOS first — iPhone & iPad, free download, optional Founder & Mentor Packs. Android follows shortly after.

Need help? Support & FAQ Β· Press inquiries: press@idlepy.com