⚫ Try it now → https://naklitechie.github.io/KoLocal/

Play Go (Baduk/Weiqi) against an AI opponent — entirely in your browser. No install, no account, no ads, nothing uploaded anywhere.

• Full Go rules on 9×9, 13×13, or 19×19 boards
• Three difficulty levels (Easy / Medium / Hard) backed by Monte Carlo Tree Search
• Liberties, capture, ko, suicide, and Japanese territory scoring — all implemented correctly
• Stone hover preview shows where you're about to place
• Last-move marker so you always know what the AI just did
• Correspondence mode — every move updates the URL. Copy the link, send it to a friend, play async. No server, no login — the URL is the save file.
• SGF export for game records
• Tap-to-move on mobile (no drag required)

1. You are Black and move first. Tap an intersection to place a stone.
2. Surround opponent stones to capture them — a group is captured when all its adjacent empty points (liberties) are filled.
3. You can't play a move that leaves your own stone with zero liberties, unless it captures.
4. Ko rule: you can't immediately recapture a single stone that just captured yours.
5. Both players pass consecutively to end the game. Territory is then counted.
6. Score = your territory + your captures. White gets 7.5 points komi (compensation for going second).

MCTS needs thousands of random game simulations (playouts) per move. Running those through the full rules engine — with DFS group searches, string-keyed Sets, and board hashing — was far too slow.

The fix: a separate FastBoard playout engine that uses Int8Array for the board, pre-allocated scratch buffers (no GC pressure), inline capture/suicide checks, and simple ko tracking via a single forbidden-point index. It shuffles empty intersections and tries random moves directly instead of pre-computing all legal moves.

Result: ~500ms for 2000 playouts on 9×9 — the full rules engine is only used for the real game and tree expansion, never during simulation.

Coloured with japan-02 · 生成り KINARI — Edo merchant ledger, washi paper. Warm cream body, kaya-wood goban, classic black-and-white stones, Edo-navy action accent. The colour of a 19th-century Japanese paper-and-wood game room.

Palette pulled from Rangrez, the global colour-palette library that backs all NakliTechie projects.

Browser-native tools — no server, no API keys, no data leaving the device.

Built by Chirag Patnaik