Plug-in Game Detection
Adapters snap onto any 2048-style board, reading the grid in milliseconds and adapting to obfuscated DOMs without shipping new builds.
PLAY · OBSERVE · OPTIMISE
Where automation artists craft AI that plays for the thrill of it.
Playkit.ai is a living lab for mastering 2048-style games. Inject our runtime, watch the WebAssembly solver wrestle the board, then remix strategies from the `.kiro` specs to build your own competitive agents.
Move latency
82 ms
Avg per Expectimax decision
Win rate
99.4%
4096 tile attainment
Runtime targets
≤ 20 KB bundle · CSP safe · desktop-first runtime with mobile expansion on deck
What you get
A modular playground engineered for fast iteration and cinematic runs.
WebAssembly Solver Core
Expectimax, Monte Carlo, and heuristic hybrids compiled from C++ deliver lightning-fast moves that feel deeply human and ruthlessly efficient.
HUD Built for Observability
Shadow-DOM overlay with live telemetry, strategy switching, and step-through controls—engineered for spectators and tinkers alike.
Runtime journey
From a single paste to a solver you can trust, replay, and remix.
Drop-in runtime
Bookmarklet or MV3 extension loads a single runtime bundle. No external calls, no fingerprints, no risk of breaking CSPs.
Adaptive sensing
Detectors normalize any 4×4 grid into a canonical state, while the driver waits on confirmed DOM mutations before committing moves.
Autopilot with intent
Strategy engines favor stable corner builds, dynamically pivoting when entropy spikes. Expectimax owns the board; heuristics own the vibe.
Insights that teach
Every run emits scoring, tile peaks, and move density so you can benchmark tweaks or train the next generation of micro-solvers.
Under the hood
Solver architecture built on shared runtime primitives.
The runtime coalesces adapters, observers, and driver loops into one orchestrated timeline. Strategies compiled from C++ live in WebAssembly modules that hot-swap without reloading the page.
- Adapter priority stack resolving Play2048 → GetMIT → Generic 4×4 heuristics
- Mutation observers and board hashing ensure moves confirm before the next dispatch
- Expectimax heuristics tuned for corner builds, ready for ML-assisted rollouts
- Telemetry and HUD overlays sealed inside a Shadow DOM for zero CSS collisions
Stack
Next.js 15 · Turbopack · Tailwind · Vanilla runtime modules · C++20 → WebAssembly via Emscripten · Binaryen optimisations
Specs live in `.kiro`
Architecture, testing, and rollout playbooks live alongside the codebase so contributors can bootstrap new experiments in hours, not weeks.
What's next
A community-powered arcade of automation is loading.
Phase · Canvas intelligence
Vision-based adapters and hotspot trainers to conquer fully drawn boards and mobile-first layouts.
In design
Phase · Strategy marketplace
Sandboxed agent API so creators can ship solvers, browse matchups, and earn reputation in an open leaderboard.
In design
Phase · Playkit OS
Multi-game orchestration for Minesweeper, Sudoku, and incremental worlds with replay timelines and community labs.
In design