Engines¶

Engines are standalone game-playing programs that MOSAIC integrates as built-in opponents. Unlike Workers, engines do not train policies. They provide a fixed-strength adversary so humans or RL agents can play against a competent opponent without launching a full training run.

        graph LR
    subgraph "MOSAIC GUI"
        HC["Human Control"]
        RL["RL Agent"]
    end

    subgraph "Engines"
        SF["Stockfish<br/>Chess"]
        KG["KataGo<br/>Go"]
        GG["GNU Go<br/>Go"]
    end

    HC -- "play vs" --> SF
    HC -- "play vs" --> KG
    HC -- "play vs" --> GG
    RL -- "train vs" --> SF
    RL -- "train vs" --> KG

    style HC fill:#4a90d9,stroke:#2e5a87,color:#fff
    style RL fill:#50c878,stroke:#2e8b57,color:#fff
    style SF fill:#ff7f50,stroke:#cc5500,color:#fff
    style KG fill:#ff7f50,stroke:#cc5500,color:#fff
    style GG fill:#ff7f50,stroke:#cc5500,color:#fff

Integrated Engines¶

Engine	Game	Service Module	Notes
Stockfish	Chess	`gym_gui.services.chess_ai.stockfish_service`	UCI protocol. Requires `sudo apt install stockfish` or `brew install stockfish`. Used as opponent in PettingZoo Chess (`chess_v6`).
KataGo	Go	`gym_gui.services.go_ai.katago_service`	GTP protocol. Neural-network Go engine. Model weights stored in `var/models/katago/`.
GNU Go	Go	`gym_gui.services.go_ai.gnugo_service`	GTP protocol. Classical Go engine (no neural network). Lighter alternative to KataGo.

Engine vs Worker¶

	Engine	Worker
Purpose	Fixed-strength opponent	Trains RL policies
Learns?	No	Yes
Protocol	UCI / GTP	gRPC + JSONL (Fast Lane)
Example	Stockfish, KataGo	CleanRL, XuanCe, Tianshou