tomitankChess 2.0 was the first full BitBoard JavaScript chess engine on the World!
Why pure? Because it is No Emscripten (unlike stockfish-js), so the code is readable.
Pure JavaScript ~5X slower than "C" language, and ~2.5 slower than Emscripten JavaScript.
The Engine is running in your browser and with UCI GUI (eg: Arena) as well.

Usage:

tomitankChess (with node.js) run in Arena, in WinBoard and in Cutechess as well.
tomitankChess.exe was built with Node v24.14.1
- For other platforms or Node versions, see the instructions below.
Example for Arena GUI with node.js
- Command line: direct acces to node.exe (C:\Program Files\nodejs\node.exe)
- Command line parameters: direct acces to tomitankChess.js (C:\Program Files\nodejs\tomitankChessUCI.js)
Web browser: tomitankChess runing with and without WebWorker as well. (Recommended: WebWorker)
- Input (standard UCI commands) to the engine is posted as a message to the worker. Example:
- var tomitankChess = new Worker('tomitankChess.js');
- tomitankChess.postMessage('ucinewgame');
- tomitankChess.postMessage('isready');
- at this point you need to wait (eg with Promise) for readyok response
- tomitankChess.postMessage('go depth 12');
JSUCI: https://sourceforge.net/projects/jsuci/

Estimated level (CCRL 40/40):

7.0: ~3050 elo /2026.07.05/
6.0: ~3000 elo /2026.03.31/
5.3: same as 5.1 only the uci code changed
5.1: ~2920 elo (Strongest JavaScript Chess Engine /2021.08.08/)
5.0: ~2900 elo (Strongest JavaScript Chess Engine /2021.01.18/)
4.2: ~2830 elo (Strongest JavaScript Chess Engine /2020.09.24/)
4.0: ~2820 elo (Strongest JavaScript Chess Engine /2020.01.24/)
3.0: ~2780 elo (Strongest JavaScript Chess Engine /2019.01.14/)
2.1: ~2660 elo (Strongest JavaScript Chess Engine /2018.11.26/)
2.0: ~2660 elo (Strongest JavaScript Chess Engine /2018.07.11/)
1.5: ~2590 elo (Strongest JavaScript Chess Engine /2017.12.03/)
1.4: ~2550 elo (Strongest JavaScript Chess Engine /2017.09.30/)
Click here to see all CCRL 40/40 results of tomitankChess
Click here to see all CCRL 40/4 results of tomitankChess

Changes log:

v7.0
- Focused on reducing memory lookups (arrays/functions), which are costly in JavaScript
- Replaced Rank/File/Mirror arrays with bit manipulation
- Switched to direct indexing instead of function calls on BitBoard arrays
- Bit faster NN index (~2% NPS gain). In this workload, optimized JS performs similarly to WASM SIMD, making SIMD unnecessary complexity
- TT optimizations: improved replacement scheme, faster access, fewer lookups
- Added a custom symmetry-safe evaluation cache (~5% NPS gain even with the main TT enabled)
- Optimized pawn eval, capture eval, and NN updates
- Added plain magic bitboards (~9% NPS gain)
- Minor time-management improvements
- Overall NPS gain: ~20% (compared to 6.0)
- Same eval (HCE + small NN), retrained with a better trainer
- Estimated strength: around 3050 elo
v6.0
- use Math.clz32 when available
- own native popcount64 wasm code when wasm is supported
- improve move sorting (use see only inside the move picker)
- use Typed Array rest of the code
- improve see a little bit
- more agressive nmp and razoring
- fix memory overflow after "ucinewgame" command (don't realloc tt table)
- fix "forfeits on time" in short time control (substract 50ms for lag)
- revisited and refactored code (would be better rewrite from scratch..)
- fix en-passant hash key update -> only when sq attacked by opponent
- main eval is not changed (boring for me), still use HCE + small NN
- 60% faster NPS, so tomitankChess play full of strength with same eval
- bit better draw detection
- Known issues:
- Fifty-move repetition may incorrectly ignore mate positions
- Stop and quit commands are ignored in the Node.js environment
- Future improvements aka TODOs: better time management
- Estimated strength: Senpai 1.0 (around 3000 elo)
v5.3
- add root moves result instead of multiPv for analysis
- maybe a bit stronger than 5.1
v5.2
- small refact on UCI code + fix three fold repetition code
v5.1
- network is trained with 4M example instead of 2.7M as in the previous version. (same size 768x16x1)
- better hash table usage (prune more in Qsearch as well, and improve hash store)
- around 20 elo better than previous version
v5.0
- Added eval (NN)! Network size is only 768x16x1. Smaller than I've seen before.
- it does not replace evaluation, it only compensates that.
- trained only with 2.7M example.
- The network is not compatible with other engines and it's integrated into the code. (hardcoded)
- I wrote the network from scratch. I don't use machine learning platform.
- I have around 10 elo better net (768x32x1) in fix depth test, but JavaScript is too slow, and don’t have AVX or similar methode in JavaScript (so i use vanilla approach). Essentially the smaller net has nearly the same strength in both short and long TC tests.
v4.2
- revisited code
- rewrite pawn eval
- fixed a bug that's caused problem in IE
v4.0
- see pruning
- better lmr
- rewrited pawn eval
- new pawn shield code
- new see (add discovered checks)
- lots of little things (revisited code)
v3.0
- new pawn shield code (idea from Senpai 2, but a bit better)
- use 3x3 squares king ring (idea from Senpai 2)
- new king safety code (simple, but bit better)
- all parameters scoring by logistic regression
- better time management (take more time when unsure)
- 4-way transposition table
- lots of little things (revisited code)
v2.1
- Revisited code (threats, move gen, some bug-fix, etc..)
- Store eval score in hash
- Smaller king-ring
- Added logo.png
v2.0
- first full BitBoard JavaScript chess engine on the World
- added check evasions code
- added threats for all pieces
- added new passed pawn eval
- new King Safety (bigger and pawn-safe king ring)
- new non-linear and pawn-safe mobility (Same in mg and eg)
- better aspiration window
- a bit better delta pruning
- Don't give back the "upper bounds" moves
- pre-calculate and transmit the "gives check"
- added pawn hash (there is no elo gain yet, but it will be useful for a more complex pawn evaluation.)
- Typed Arrays Hash table for better memory usage
- added "Hash option" (default 32 min 1 max 256)
- added "eval" command (show the static evaluation)
- lots of little things
v1.5
- New UI interface code (now working with JSUCI as well)
- Tuned Search algorithm
- Use TT at depth == 0
v1.4
- first public version

Usage:

Estimated level (CCRL 40/40):

Links:

Changes log: