No signup, no ads, no limits. The Nardex neural-network engine, trained on 20+ million positions, is waiting for your move.
Play free →Difficulty levels — from casual games to training
In the lobby you can pick the AI level: from a light bot useful for drilling openings and learning the rules, up to a strong engine that holds its own at club-player level. The same neural-network evaluator runs underneath — the difference is in search depth and move-pick temperature. Want a quick win and a taste of victory? Pick the easy level. Want to see what the AI is really capable of? Pick the strong one and brace for a serious game.
No signup, no ads
You click «Play» and the game starts. No email confirmations, no banners, no popups. Progress is saved in the browser, and if you want to enter the global rating or sync game history across devices — you can register in one click. But it is optional: nothing is required to play the AI.
A real AI, not a script
Most online narde sites use simple if-else logic: «if the roll is 6-6, go there». Nardex is a full neural network trained on TD-Gammon principles (Tesauro, 1995) and adapted to long narde with its specific game phases (blocking / race / x-escaped / o-escaped). Behind every move of the network there is:
- 20+ million positions in the training set (random + self-play + extracted from real games)
- 192,000 expert positions from real played games, labelled by GPU rollouts
- ONNX inference on CUDA — millisecond evaluations on a modern GPU
- Multi-phase networks — separate models for blocking and race endgames
All Russian Ministry of Sports rules respected
Nardex plays by the official long-narde rules approved by the Russian Ministry of Sports order №347 of 30.04.2025 (Article 23). One checker off the head per turn (except for the first-move 6-6 / 4-4 / 3-3, where two are allowed). Scoring: mars = 2, koks = 3 (see the separate article on mars and koks). The doubling cube with the Crawford rule is available in match play. No improvisation, no «grandma's house» variants.
Want to go deeper?
If you are serious about levelling up your game — see our guide to long narde strategy: opening templates, building a prime, the race endgame and the doubling cube. Still getting familiar with the rules? Start with the long narde rules. Want to play on your phone — Download the Android app (APK); the iOS version is in development.
Frequently asked questions
Is it free?
Yes, playing against the AI on nardex.ai is free. There is no cap on the number of games, no embedded ads, no premium walls. If we add paid options in the future (tournaments, advanced analytics), playing the AI will remain free.
Do I need to sign up?
No. You can hit «Play» and start a game against the bot anonymously. Your progress is stored in the browser. Signup is only required if you want to play matches against other people, appear in the rating, or sync progress across devices.
How do I switch the difficulty level?
The lobby has a difficulty dropdown — from a casual bot for beginners to a strong engine playing at club level. The chosen level is fixed for the current game; you can switch again before the next one. See the FAQ on the home page for details.
Can I train against the AI?
Yes. After every game the analyzer shows a review: each move is graded (best / doubtful / error / blunder), equity-loss is computed for every decision, and you can see what the AI considered the best move in each position. It is the same network that played against you — the analysis is not synthetic, it comes from the very opponent you faced.
Is there a mobile app?
Yes, for Android — download the app directly from the site or via the download button under the lobby on the home page. The iOS version is in development. The nardex.ai web version is fully adapted for mobile browsers: you can play directly from your phone without installing anything.
What kind of AI is this? Not just another script?
It is a custom neural-network engine. Not if-else logic of the «if the roll is X, move there» kind, but a full neural network trained on 20+ million self-play positions and tens of thousands of expert games. The architecture is based on TD-Gammon (Tesauro 1995) but rewritten from scratch in Rust + ONNX and updated for modern GPUs. The same engine is available via the Analysis API for developers.
Ready to play?
One click — and you are at the board against Nardex AI. No signup, free.
Play right now →