TouchGrass: a phone referee for outdoor games, with an open-weight Game Master

Description and use

TouchGrass uses an open‑weight AI to suggest creative twists for outdoor games, while a deterministic engine and safety checks ensure only legal events are applied, freeing players from manual refereeing.

Problem

Friends, families, and youth groups often argue about rules, scoring, or timing during outdoor games, preventing them from playing freely.

Outcome

TouchGrass demonstrates that an open‑weight AI can suggest creative game twists while deterministic code ensures legality and safety, enabling seamless outdoor play without human referees.

Workflow

• 1. Players start a game session on their phones. 2. The game server creates a persistent session and notifies participants. 3. At each turn, the Game Master agent receives a trimmed game context. 4. The agent calls two read‑only tools: game_history and game_rules. 5. The agent returns a structured JSON proposal validated by Zod. 6. The engine checks the proposal for legality. 7. A deterministic safety layer validates the proposal against safety rules. 8. If all checks pass, the engine applies the event and broadcasts the update. 9. Players receive the updated game state and continue playing.

Evidence

The Game Master can propose a twist for a round: a safe zone rule, a bonus challenge, or a little more time.

Source

The model never directly changes the game state. Its proposal must pass three checks implemented in ordinary code before it can be applied.

Source

The project has a game engine, a multiplayer server, persistent game sessions, and an AI Game Master powered by an open weight model.

Source

The project is MIT licensed. The README contains the commands needed to install dependencies, run verification, start the server, and try the local demo.

Source