Cortex64 is a desktop AI chess application built entirely from scratch in Python. No Stockfish. No python-chess. No external chess engines. Everything — the move generator, the AI search, the GUI — is hand-written.

Feature v1 v2 Main Menu ❌ ✅ Full animated main menu Board Skin Grey flat ✅ Obsidian Classic (warm oak + walnut) Animations ❌ ✅ Piece slide, shake, check pulse Evaluation Bar ❌ ✅ Live centipawn bar Chess Clock ❌ ✅ With increment support Difficulty CLI flag only ✅ In-app slider (5 levels) Move Quality ❌ ✅ ✓ △ ✗ badges per move Results Screen Basic popup ✅ Full stats + accuracy % Analysis Mode Basic ✅ Eval graph + engine arrows Settings ❌ ✅ Skin, sound, animation toggles Drag & Drop ❌ ✅ Sound ❌ ✅ Move, capture, check, illegal

Full legal move generation (castling, en passant, promotion)

Check, checkmate, stalemate detection

Reversible board state (push / pop)

Negamax + alpha-beta pruning with move ordering

Depth ceiling and time-bounded search

CNN-based evaluator (PyTorch) — falls back to material/positional if no model

Multi-screen architecture — MainMenu → GameSetup → Game → Results → Analysis

Obsidian Classic board skin — warm oak light squares, dark walnut dark squares

Smooth piece animations — 120ms ease-out slide on every move

Illegal move shake — board shakes instead of silent failure

Check pulse — king square flashes red when in check

Evaluation bar — smooth vertical bar with live centipawn label

Chess clock — countdown with increment, danger flash under 30s

Move quality badges — ✓ good / △ inaccuracy / ✗ mistake per half-move

Accuracy stat — computed at end of game

Eval sparkline — evaluation graph in analysis mode

Drag & drop — click-to-move and drag-and-drop both supported

Sound effects — move, capture, check, illegal, game end

4 board skins — Obsidian Classic, Classic Walnut, Ice Blue, Neon Dark

Settings persistence — all preferences saved to gui/data/settings.json

Argument Description Default --depth Maximum search depth Uses in-app slider --move-time AI move time budget (seconds) Uses in-app slider --human Your color: white or black Uses in-app selector --model Path to CNN model weights Falls back to material eval

Input Action Left-click piece Select piece (shows legal moves) Left-click destination Move selected piece Click-hold + drag Drag and drop piece Q / R / B / N Promote pawn to chosen piece ESC Clear selection / close popup R Quick rematch (on results screen)

Button Action ↩ Undo Undo last move (disabled while AI is thinking) 💡 Hint Show best move arrow + explanation ⚑ Resign Resign current game (with confirm)

Input Action ← / → arrow keys Navigate moves Click on eval graph Jump to that position Click move in list Jump to that move

If no model file exists, Cortex64 falls back to material + positional scoring.

Level Depth Move Time Style Beginner 2 0.5s Makes frequent errors Casual 4 1.0s Plays reasonable moves Intermediate 6 1.5s Solid club player level Advanced 8 2.0s Challenging for most players Master 12 3.0s Strong positional play

Python 3.11+

Windows / macOS / Linux (display required for Pygame)

Dependencies: numpy, pygame-ce, torch

No external engines — every chess rule implemented from scratch

No python-chess — move generation is fully custom

Clean separation — engine/ai/gui are fully independent layers

Educational focus — hints, move quality feedback, analysis mode

Time-bounded AI — strict wall-time control prevents UI blocking

CNN model missing The application runs using material/positional fallback evaluation automatically.

Complete multi-screen architecture

Obsidian Classic board skin

Piece slide animations, board shake, check pulse

Evaluation bar, chess clock, eval graph

Move quality badges and accuracy scoring

Drag-and-drop support

Sound effects

Settings screen with skin selection

Custom chess engine (negamax + alpha-beta)

Pygame GUI

CNN evaluator with fallback

Undo, Hint, Resign, PGN export

Profile persistence

Built with ♟ by vivekyarra · No Stockfish · No python-chess · Just Python