A free, open-source 3D chess training game.
Kaissa is not another place to play chess online. It is a game whose purpose is to make you a stronger player. It does this by building your chess pattern recognition through adaptive, spaced practice that is embedded in gameplay rather than presented as lessons to memorize.
The name is a placeholder and may change.
Alpha. The learning core is complete and tested, and there is a playable Unity client on desktop covering training, puzzle blitz, daily puzzle, themed practice, play-vs-bot, endgames, openings, an analysis board, post-game review, stats, board-vision and coordinate drills, and calibration. The interface was rebuilt in Unity UI Toolkit with a consistent nav-rail/board/panel layout, and the board can be shown flat (2D) or in 3D. A real modelled piece set and further polish are still pending. See docs/.
Headless core (no Unity needed) - run the training loop or drive the engine from the terminal:
dotnet test Kaissa.sln # run the test suite
dotnet run --project src/Kaissa.Training.Cli -- --simulate # simulate the learning loop
The 3D client (Unity 6 LTS):
./scripts/fetch-stockfish.ps1 # get the engine (GPLv3, not committed)
./scripts/build-unity-plugins.ps1 # build the core into the client and stage the engine
Then open client/ in Unity Hub and press Play. See client/README.md for details, and docs/client.md for the client design and setup.
Getting better at chess usually means explicit study: memorizing opening lines, grinding puzzle sets, reading theory. It works, but it feels like homework and most improvers give up. Research on chess expertise (Chase & Simon and later work on chunking and template theory) shows that strong players mostly recognize good moves from a large store of learned patterns rather than calculating from scratch. That pattern store is built through exposure and practice, not memorization. Kaissa treats improvement as a training problem. You play; the system tracks which patterns you have encountered and how well you retain them, and shapes upcoming positions and opponents to practice the ones that need work, at a difficulty where learning is fastest. There are no flashcards and no "memorize this line" screens.
Most chess software is built around playing games, with learning offered separately as a library of lessons and puzzles to work through. Kaissa inverts that: it is a 3D game built around an adaptive, spaced training loop, so improvement happens through play rather than as homework on the side. It is free and open source, in the spirit of the best community-run chess projects.
- Unity (Universal Render Pipeline), targeting Windows/macOS/Linux and mobile from one codebase.
- Stockfish as the chess engine, run as a separate process over the UCI protocol.
- FSRS (Free Spaced Repetition Scheduler) for scheduling practice.
- The learning core is plain C# with no Unity dependencies, so it can be tested on its own.
Details in docs/architecture.md.
GPLv3. See LICENSE. Networked server components, if added later, will be AGPLv3. The project is free to use and always will be; see docs/vision.md for how it is meant to stay sustainable without charging for the parts that make you a better player.
See CONTRIBUTING.md.
docs/vision.md- goals, audience, scope, sustainabilitydocs/research.md- the research and prior art the design rests ondocs/architecture.md- system structure and data flowdocs/learning-engine.md- how the training loop worksdocs/client.md- the Unity client and UI design, and how to set up to build itdocs/features.md- the feature set and roadmapdocs/mobile.md- plan for the engine on Android/iOSdocs/release.md- building a desktop build and cutting a releaseCHANGELOG.md- release notes