Open-source BYOK playground forces LLMs to output legal moves across five board games
wilsonye · reddit · 2026-09-08
The author built the opposite of typical "play chess against an LLM" demos: models must emit a parseable legal move applied by a rules engine, and three illegal or unparseable outputs in one match means a loss.
- Covers Chess, Go, Xiangqi, Gomoku, and Othello, with human-vs-model and model-vs-model spectate modes
- Two optional prompt knobs: tactical guidance and a legal-move list (Chess/Xiangqi/Othello only); reasoning effort Off–High, temperature left at vendor defaults since several reasoning models reject forced values
- BYOK, keys never written to disk; supports OpenAI, Anthropic, Google, xAI, DeepSeek, Kimi, Qwen, GLM, MiniMax and more; MIT-licensed and self-hostable
The author wants data on which models stay legal for a full game, whether listing legal moves changes strength or just syntax, and where Xiangqi/Go collapse compared to Chess.
More from coding & agent
- Open-source GPT-6 Astra + Seedance 2.5 pipeline swaps characters in any video — matchaman11 · 2026-09-08
- GPT-6 Astra drives Blender via MCP: load-model animation escapes uncanny valley — teortaxesTex · 2026-09-08
- When did 3D map generation become the de facto demo benchmark for coding models? — gethackteam · 2026-09-08
- danluu evaluates how well agents use test/verification techniques; Astra impresses — teortaxesTex · 2026-09-08
- Dev calls on OpenAI and Anthropic to donate compute to rewrite C libraries in Rust — cramforce · 2026-09-08
- Commit agent memory to GitHub: markdown files for cross-machine agent continuity — ariG23498 · 2026-09-08