Bot Trainer
Train an AI to play your board game
You can train an AI to play your own board game in a browser tab with DiceyTable's bot trainer. It learns by playing against itself, AlphaZero style, and the finished network becomes a bot players can seat in your game's lobby. Nothing is uploaded while it trains.
Write your rules as a deterministic rules kernel, then open the trainer from the editor under Manifest, Bots, Train a model. Pick a tiny, small or medium network and leave it running. After each generation, a gauge shows how the new checkpoint does against random, greedy, minimax and search opponents.
When you are happy with it, export the checkpoint into your project as an ONNX model and attach it to a bot. No game yet? Open the trainer without a project to try a Reversi demo.
What is DiceyTable?
DiceyTable is a free 3D virtual tabletop with a game editor built in. For game makers it includes a bot trainer and a bot system, so your game can offer computer opponents you trained yourself, in your own browser.
Training needs pure, deterministic rules, so games decided by dice or physics can only have hand-written bots. A short run gives a modest bot: our 6-minute test beat random play and roughly matched greedy.
Train a bot for your game
Open the trainer from your project, start a run and watch the strength gauge. It is free and stays in your browser.