Installer pachi avec Homebrew, apt, dnf

Liens du projet

• Page d'accueil
• Dépôt
• Page du gestionnaire de paquets

Commandes et alias

• pachi

Historique du projet

Petr Baudis originated Pachi as both a modular framework for Go-playing programs and a strong engine within that framework. Jean-loup Gailly contributed extensively to distributed search, tuning, memory allocation, and time management; later maintenance continued under lemonsqueeze.

The 2010 paper by Baudis and Gailly presented Pachi as a state-of-the-art open-source Go program centered on RAVE-based Monte Carlo Tree Search, with work on time control, dynamic komi, criticality, and parallel search. Later releases added pattern and deep-learning guidance while retaining the project's GTP-oriented architecture.

Historique d'adoption

Pachi became one of the notable open-source engines from the Monte Carlo Tree Search era of computer Go. Its official history and research material describe competitive online rankings and distributed-cluster play.

The engine is packaged as a command-line GTP program rather than a self-contained desktop game. That encouraged use with independent frontends and Go servers, including GoGui, Sabaki, Lizzie, and kgsGtp.

Modes d'utilisation

Run Pachi as a GTP engine behind a compatible graphical frontend or connect it to a Go server through a GTP bridge. The defaults provide an initial configuration, while command-line parameters control engines, logging, networks, rules, and analysis behavior.

Pachi does not document a persistent application configuration or credentials file. Internet-server authentication belongs to the selected frontend or bridge rather than the engine.

Upstream warns that its GTP parser is not secure and should not be exposed directly to untrusted users.

Pourquoi les passionnés de paquets s'y intéressent

Pachi preserves an important generation of computer-Go engineering: UCT search, RAVE statistics, pattern-guided playouts, dynamic komi, and distributed search in a modular C codebase.

Its package boundary illustrates the composable Unix model particularly well: an engine speaks GTP, a separate frontend supplies the board and analysis interface, and optional networks or cluster resources improve play.

Chronologie

• 2010: Baudis and Gailly published “Pachi: State of the Art Open Source Go Program.”
• 2010s: Pachi developed distributed search, pattern guidance, and deep-learning-assisted play.
• 2026: The project released Pachi 12.90 and remained available through Homebrew.

Related projects

• GNU Go represents the earlier knowledge-driven open-source Go-engine tradition.
• KataGo represents the later neural-network and self-play generation of strong Go engines.
• Sabaki and GoGui are GTP frontends capable of presenting Pachi's play and analysis.

Sources

• http://pasky.or.cz/go/pachi-tr.pdf
• https://github.com/pasky/pachi#readme
• https://github.com/pasky/pachi/releases
• https://pachi.or.cz/